(10/7) NEWEST COMFYUI FRONTEND BROKE impact-pack. If you get the t.findInputSlot error, download the v2.5.1 WF update.
I just made a Discord finally! Please join to connect with other AI Filmmakers or if you have questions about my workflows/process: https://discord.gg/PGAXv4xq7
[ free patreon for workflows/tutorials/support me: patreon.com/FoxFurEssence ]
[ new patreon for NSFW content I make: patreon.com/FoxFurSpice ]
This advanced filmmaker's workflow cleanly and quickly generates 3 low resolution sample preview gens, allowing you to select your favorite to then take to upscale finalization. The workflow has every feature an AI filmmaker could need, and easily produce the highest quality and resolution Minimax is capable of. It comes equipped with all quality of life features, the top known speed ups, and many options, from seamless video continuation to custom audio, to frame interpolation to go from 24 FPS to 48-60. Everything is here!
v2.5's Dialogue / Voice Cloner tool Quick Guide:
v2.0 Usage Guide:
Shortfilm entirely made using this workflow:
Some nodes in this workflow are too new to be in the Comfy Registry to be installed automatically with Comfy-Manager! Therefore, they need to be installed manually.
.: Latent Upscaler
- Node: https://github.com/LBH-123-AI/Comfyui_Minimax_h3_latent_Upscaler
- Latent Upscale Model: https://huggingface.co/LBH-123-AI/Minimax_h3_latent_Upscaler/blob/main/minimax_h3_latent_upscaler_3d_conv_v1/minimax_h3_latent_upscaler_3d_conv_v1_fp16.safetensors
(put into /latent_upscale_models/ folder and, since 9/17, this needs to be reselected in the WF's selection for latent upscaler choice due to the author renaming the file)
.: Pixaroma Node Suite
https://github.com/pixaroma/ComfyUI-Pixaroma
.: Prompt IDE
https://github.com/ethanfel/ComfyUI-H3-Prompt-IDE
.: Resources Loader (MiniMaxH3-PromptBuilder)
https://github.com/Adudeguyman/ComfyUI-Fantastic-MiniMaxH3-PromptBuilder
.: RefMods
https://github.com/Luisacaotica/ComfyUI-MiniMaxH3Mod
.: Turbo LoRA
NEW [Kijai's DMAD 4-step - for both FL2VA/REF2VA - 8 steps 1.0 strength recommended] https://huggingface.co/Kijai/MiniMax-H3-experimental/blob/main/loras/minimax_h3_DMAD_4step_full_lora_avg_rank_39_bf16.safetensors
Older:
This particular Turbo LoRA is my favorite. I recommend 6-8 steps if at 1.0 strength, or 8-10 steps if using at 0.6-0.8 strength. If you're upscaling, 6 first pass + 3 upscale pass works well.
If you use ref2va or Singularity, this is great too:
.: Kijai's int8 Video VAE (much faster than default!)
Everything else should be easily available from the ComfyUI-manager Install Missing Nodes screen!
.: TaeH3 (for Kijai's Model Preview Override):
https://huggingface.co/Kijai/MiniMax-H3-TAE/blob/main/vae_approx/taeh3.safetensors
(note, using Kijai's Model Previewer will slow gens down by 6-16% thereabouts, I have it disabled by default now, but you can turn it on to test for yourself).
If you support what I'm doing, please consider following or subscribing to my Patreon! It's free (if you want it to be) and all my content and updates go there!
Description
FAQ
Comments (204)
For some dumb reason, CivitAI's internal AI image rating detector has marked all of my samples as X rated. Please help by voting for them to be PG or PG 13.
Awesome!
How does it handle v1.1 of the LTX 4-step lora?
You mean Lightx2v LoRA, LTX stands for Lightricks lol. I tried it but wasn't a fan. I like the 8 step LoRAs.
ERROR IN FINAL UPSCALE PASS
I had to rename the folder created in custom nodes to: MiniMaxH3LatentUpscaler3D.
Also changed device to cuda and precision to fp16.
Hope that info can help someone else.
That's... odd, those are default settings for most users.
I hit the issue "TypeError: t.findInputSlot is not a function in /extensions/ComfyUI-Impact-Pack/impact-pack.js" when opening the workflow. Anyone got idea around it?
uninstalling impact pack worked
update: scratch that, caused issues for upscale pass
I just had claude code fix it locally
"Root cause: GitHub issue #1223 — confirmed, still open upstream (there's an unmerged PR #1228). Impact Pack's ImpactSwitch/LatentSwitch/SEGSSwitch handling in impact-pack.js:639 had a leftover "slot-0 hack" calling node.connect(link_info.origin_slot, node.id, 'input1') — passing a raw numeric node ID as the connection target. ComfyUI's frontend ≥1.47 (you're on 1.49.6) expects an actual node object there, and internally calls target.findInputSlot() on it — since it got a number instead, that throws and aborts the whole workflow load.
Your version gap explains why you're hitting it now: ComfyUI core was updated 2026-08-15, but this Impact Pack checkout hasn't been git-pulled since 2026-04-20 — well before the frontend crossed 1.47.
What I changed: removed the broken hack block in impact-pack.js (only present when a switch node has a *-typed slot-0 input). It's dead code anyway — the type-propagation loop directly below it already handles origin-type propagation correctly on the current frontend, no forced reconnect needed."
Impact pack should be okay, plenty of other people have used this workflow without error. Sounds like claude made up a bunch of crap to solve something that isn't impact-pack related at all. Though I can't say for sure, outside of the fact that you're the only one who's reported this error so far.
@foxydits it solved the issue so i thank you overload claude, please dont smite foxy
@isit lol, well, claude is quite good at coding. good stuff then! no smiting required :)
Unfortunately this workflow wont work if you use the latest version of comfyui, v0.33.3 only works with v0.33.0
MiniMaxH3AddGuide seems to be removed in 33.3
also impact pack seem be breaking this workflow for some reason and keeps throwing the error
Loading aborted due to error reloading workflow data
TypeError: t.findInputSlot is not a function
This may be due to the following script:
/extensions/ComfyUI-Impact-Pack/impact-pack.js
i never had an issue with it on my other comfy portable install but as soon as i added it here, i get this issue
EDIT:
after uninstalling and reinstalling and switching comfyui versions and disabling impact pack in many various patterns......it might be working
SECOND EDIT
It only works in firefox for some reason.
EZI or chrome it breaks, this is weird, i cant use in firefox as its slow as shit lmao ohwell ill ignore this workflow
Weird, but add guide is not required for the workflow. it is just an option for first frame last frame, totally uncritical. just run with it turned off like it is by default.
@foxydits are you working on a fix?
@millerwarhammer384 I updated to the most recent comfy / updated all my nodes just to test to see if this was true. The workflow works fine. This is the user's problem, if I'm being completely honest. There is no fix besides that the reviewer's configuration differs from the 99%+ people who are using this workflow without issue.
I'm having the exact same issue and can't find a way around it. Too bad... was looking forward trying it out :/
@KirmesFabi What version of comfy are you running? Because the user reporting this claims to be running a version that isn't even publicly available based on the research I was able to ascertain (he says 33.3 but I am only able to update to 33.0, with a PR for 33.1 being barely available on github). You two are the only ones who are having trouble so far out of a bunch of early reports of the workflow being completely functional, so I am curious as to what the ACTUAL issue is.
@foxydits Hi mate, so 33.3 is the newest stable version
MiniMaxH3AddGuide missing error disappeared when I installed the https://github.com/PlagueKind/ComfyUI-PlagueKind-Nodes.git node
@Zeb101 for desktop comfyUI, yes. for portable (which is the version of comfyUI most people use and the one I always recommend) it's 33.1, which still has the node in perfect working order. I'm not sure why they decided to remove or change the name of that node, but again, the node is not required to run this workflow. it's part of an optional functioning in the workflow that only like 5% of users would even turn on.
im not so much as worried about the MiniMaxH3AddGuide node not working
but the impact pack node "any switch" is what is making it fully nonfunctional and destroying the workflow and then infects the other tabs and breaks them too, the one that you used to switch the 3 latents
i had to load the workflow up in firefox as it was the only place that let me load the workflow with out it imploding immediately, delete that node and replaced it with a rgthree anyswitch
Even with the pack uninstalled it breaks something core to the comfy ui workflow loading
it is not anything you did, but a reported issue on the impact packs github approx 2 weeks ago, they will either have to fix it or you will get this issue soon enough too
https://github.com/ltdrdata/ComfyUI-Impact-Pack/issues/1223
here are the pictures of it doing its thing
https://imgur.com/a/bLEWP0n
@rabbitything That's pretty spooky. Impact pack is a pretty popular node suite, used by thousands of workflows. I try to avoid unnecessary packs but impact has nodes that are required for several parts of my workflow's functionality. The node that parses which latent you selected from stage 1 to pass to stage 2 upscale uses it. I suppose the same thing could be done with groups though. I'll consider workarounds if there is any more trouble with impact from users.
@foxydits that might explain it... I'm using the Desktop Version v0.33.3 since my entire setup is build around it. Probably should switch to portable at some point to avoid stuff like this happening. Thanks for the explanation. Didn't meant to be ungrateful for your work though! It's highly appreciated!!
Would it be possible to integrate this into ur filmmaking workflow? Seems very helpful
this workflow far surpasses the function and flexibility of my filmmaker workflow
@foxydits That's odd, I get OOM using this one but never had a problem with v2.0 and earlier WFs. For the other WF I had everything turned on Sage, ComfyKitchen, Sol-Attn, Spectrum. 0.5 MP videos of 14s would finish in 15 minutes. The default settings on this workflow give me 13 minutes for 5s and it is using 0.3 MP by default.
you don't need ref2v model for this workflow? how would that work?
Simple, as explained in my video, the fl2va just basically IS the ref2va model but with better visuals and audio. the devs even confirmed this. Try it out and you shall see lol
@foxydits actually you are right this works even better for character replacements, I rarely see identity bleed.. this is a very good workflow btw! thanks a lot
I'm getting the error: "Final Video is missing a required input: images" when I click the Manual Random Seed button. I'm using this for basic i2v and have the Enable <Picture 1> is First Frame setting enabled and an image loaded into Load Image & Crop node for <Picture 1>.
Hi, as explained in my video, the 'red outline errors' when you run your first pass are fake and don't actually stop any progress towards your gen. The first pass will create those errors/red outlines as comfy's way of just saying "Hey, none of the images are gonna make it to finalization"... but that's normal, because this is a multi-stage workflow! Your first pass finishes, you select which of your gens you like best, you ENABLE the finalize upscale pass, then click run from there.. and tada! those nodes that gave errors before? Now they have input because you're on the finalization pass for the workflow! Please watch the video. It all makes sense, I promise.
This is working great, thanks so much. Really appreciate it. However, I'm getting garbled sound/dialogue with the default settings. Any ideas on what I might be doing wrong? I tried with 0.3 mp and 0.6mp initial gens with no change. Also tried wring in Spectrum Apply MiniMax H3 after the sampler shift and before the power lora loader—but no luck. Any ideas on what I might be doing wrong?
Adding spectrum won't help with anything. It's a speedup node that helps only with high-step (16+) gens. As far as garbled dialogue, what do you mean? Babbling characters? That comes down to prompting skill. Try ditching the <d>[English]This is what I say</d> method and just do: <Subject 1> says "Fuck, this is a lot easier". After they finish talking, they do X -- giving a character action after dialogue prevents babble, and currently (until fixed soon), the tokenizer misinterprets <d>[English]</d> prompts (which is how the prompting guide says to write dialogue). So until fixed, quote are better. Hope that helps.
@foxydits Thanks for the response. No, it's not the prompts. I have an amazing prompt engine I built and have no issue with my other workflows. And it's not just dialogue, it's garbled sound in general. But I only played with it for 20-30 minutes, so I'm sure I'll figure it out. I understand about Spectrum, but for some reason it's positively effected sound quality in other workflows—and I actually initially came across it because it was mentioned in a reddit post as having improved sound quality.
@jayhartford Other thing is make sure your audio VAE is properly selected in the model loader options. Had another user who was reporting missing audio/garbled audio and it turned out their audio VAE just wasn't actually loading, needed to be reselected.
@foxydits Thanks. Gonna fire it up and I will definitely check that right now. Regardless, amazing workflow, much appreciated
It's absolutely bizarre. I've spent hours trying to diagnose the garbled sound and can't figure out the source. The workflow loads, as downloaded, no problem and works great (except for sound of course). All files are in order etc. Nothing I do seems capable of fixing the audio. Was wondering if it's some unknown issue with my Comfy build, but ran my other H3 workflow and audio is great. I've A/B tested all the nodes I could, and even tried swapping in alternate text encoders or diffusion models (hybrid), all with no luck. I found one other guy on reddit who seemed to be having a similar issue a few days ago—and he didn't seem to solve it either. If someone more capable and smarter than me reads this and has any ideas I am all ears. I'm generally able to solve issues, but this one's got me.
@jayhartford Really sorry you've had to struggle so hard with this one. I'm a bit mystified too. As far as the latent path goes for the 1st stage, it's just a normal workflow when everything's turned off like custom audio and the like.
@foxydits Totally, it's really strange. I'm guessing I'll stumble across the solve at some point. I turned off all attention nodes, no turbo, increased steps etc etc. It doesn't make sense in the way workflow problems typically do. I must be overlooking something. However, I am using the downloaded workflow with zero changes and failing where others are having success. I'll keep you posted in case its somehow helpful
Quick update. Did a fresh Comfy install on a different Runpod persistent storage and it worked first go. Wish I understood what the issue was, but glad it's working now
@jayhartford Were you on a runpod before? Maybe the config of that runpod was outdated or weird. Anyway, glad it's working as intended with the fresh go.
none of the switching node seem to work for me, each node only has one option, the right number of switches are there they are just all the same
The workflow is great; generation speed is really good -- I only have one issue in that audio seems to not be generated at all I tried reloading the workflow defaults and can't figure it out at all
What are your video settings and what options do you have turned on?
In the 2.0 Filmmaking workflow it took me 15 minutes to generate 14-15s videos on ref2v, i2v even shorter. Now I only get OOM with this new workflow.
@udizjxvo just a reminder that 2x upscale is 4x the pixels, so if you're doing 14-15s at any resolution above 0.4 MP, it takes a beefy system to render. 0.3 MP -> 2x latent upscale = 1.2 MP, 0.4 MP -> 2x = 1.6 MP, 0.5 MP -> 2x = 1080p. These are huge latents for minimax's typical 768p target res to handle. 14-15s is a big ask for that size unless you've got a 5090 + lots of system ram.
@Patikitten In my video I talk about how VHS will, by default, save two videos each output; one that has audio, and one that doesn't. you might wanna check that's not the case for you. i discuss it in the video towards the end.
@foxydits I don't have audio in any video generated with this workflow I checked all videos come with blank audio tracks
@Patikitten Is it possible you've accidentally switched on an audio setting such as the "Force Custom Audio" option? or maybe just have an audio ref enabled with no file loaded?
@foxydits The only options I have turned on are Comfy-Kitchen | Sparse (SLA) | Finalization LoRA for upscale pass only | Image_1 | Video_1
Anything custom audio wise is disabled
It's not a huge deal since I don't need audio right now for what I'm doing since it's motion graphics but it's a bit of a bummer for other things
I might reload it again and see if it still does it from fresh but I've tried that twice now -- I did look to enable the audio things but it won't work w/o anything piped into it errors out instantly so it rules it out
@Patikitten It's certainly a weird one, and so far you're the only one who's reported this. I can't imagine what the issue is since the audio is pretty straight forward.. like, it's connected to the video, which is generating and loading correctly for you. Last thing I can think of is maybe check to make sure your audio VAE is correct / isn't corrupt? minimax_h3_audio_vae_fp32.safetensors
@foxydits I ended up fixing it by replacing my VAE with a freshly downloaded one; I guess it was corrupted it's weird since I was getting audio in other workflows
@Patikitten Stranger things happen in comfy every day. I knew after you confirmed the audio options weren't incorrectly configured it had to be something with VAE settings. Who knows, though. But I'm glad it's working.
@foxydits But should this be a problem when I am only at the stage of generating previews? I watched your video and it seems like the latent upscale doesn't happen until I generate the seed for it. I have also tried lowering latent upscale to 1x and I still get OOM.
consider me in that 1% of people with problems (someone said that 98-99% of people are running this without issues, so I must be the 1-2%)
Too busy in my life to get something this with this many components working by backtracking my entire comfy setup. Looks like a good workflow. If i ever do a reinstall of my entire setup i'll try this immediately.
I wish you told me what the problem was though. A lot of people seem to have an issue because they run the newest version of Comfy Desktop, which messes with one very optional node in this workflow.
Amazing workflow, works great!
Beautiful workflow, works amazing. Thank you!
One of the best MiniMax workflow I've come across so far. T2v, I2v is fast, except the ref2v generation and upscaling
15 sec I2V and upscaling with 1.5x multiplier takes about 10 min or less.
15 sec Ref2v + 1.4x upscaling takes about 35-40 min
Thank you! The slowness when using several references is just the model though, to be fair. You have to be kind of picky about how high resolution your references are. You want them to be high enough that the upscale pass gets nice quality references, but if they're above 2MP it can really bog down VRAM for long gens like 15s. 15s is pushing the model super hard with upscaling past 768p. I don't know what your 1st pass MP is, but for example a 0.6 MP first pass upscaled 1.4x is roughly 3x bigger in terms of pixels, coming in at a 1.68 MP finalized resolution. The balance is hard to find, but it might be better to set a target megapixels setting for your upscale to control everything more finely.
If anyone has the issue of impact pack causing this error and destroying the workflow:
Loading aborted due to error reloading workflow data
TypeError: t.findInputSlot is not a function
This may be due to the following script:
/extensions/ComfyUI-Impact-Pack/impact-pack.js
i ended up fixing it by going into the impact-pack.js file and deleting lines 635-641
https://imgur.com/a/MgbNDUL
fixes everything up nice
The best workflow right now, thank you for this, and congrat to give informations on wf notes it's very helpfull.
Ill will redacted. It would appear you and Muse Collective are equally credited for Seed Hunting workflows and progression. I removed negative comment and support your reasoning on your second response. I would definitely say your first response was short fused. But I am good if you are good. I will dig in to your workflow and test it.
Uhh, well I'm definitely the one credited as the creator of the Seed Hunting concept. I'm the first to bring it to the community as far as I know, and you're certainly the first to suggest otherwise. I've never heard of Muse Collective.. if there's some workflow or announcement of his that predates LTX 2.3's release, please let me know. And thanks for deleting your previous statement. My response to it was short fused because no one appreciates falsely being accused of theft.
@foxydits I mean yeah Muse made a video before yours came out. but just a general credit googling it, credits you both. I think he does film work and you are here. Just out of curiosity can I dump Sage Attention in favor of Comfy Kitchen? More of a hardware limitation on my end. Sage attention and Triton do not play nice with my rig.
@irishpc491 The workflow already uses comfy kitchen by default (enabled).
Hey, why sparse attention, and not the sol attention?
Now that sol has implemented Sparse I'd prefer to switch back to it, but at the time of making this wf, sol had fallen out of fashion a bit and SLA was the new hotness. Now they're basically the same efficiency and I don't wanna update the workflow for just that switch.
please could you check the speeds between the 2 again? because it would be really nice to have the fastest, thanks!
@dungaipara532 Just use Sol Attention. It's like "Sparse Attention 2.0". Same speed (And almost the same thing), i think, but better quality overall.
@dungaipara532 I gave both sol and sparse a go, 10s 1 MP. Ref2V w 5 refs
Sol: 444s
Settings: Node default
Sparse: 435s
Settings: default for this WF + Sparsity ratio 0.85
Visual quality seemed the same, with 2 samples each, Sparse seemed to follow the prompt better and produced a more coherent output but it may just be luck with the seed.
Again, this is amazing. Thank you. I had some issues that were just because I didn't follow your directions. =)
Anyone else here getting OOM while trying to use this as a ref2v flow? I am using the same minimax quantized models, no latent upscaling portion yet. Same prompt, same inputs as what I used previously in the v2.0 filmmaking one, same output settings (14s output), really makes no sense why the other workflow completes in 15-20 minutes and I cannot even get this to even run.
The only time I got this workflow to run was at a target output of 5s, even then it took 25 mins.
I have Comfy-Kitchen, Sparse turned on.
Something's definitely wrong with your setup then. Nothing but the highest res gens should take 25 minutes lol. I unfortunately don't know how to trouble shoot because, you are correct, this is identical to the Filmmaker 2.0 workflow in terms of brass tacks: they each are performing the same general steps to load the models and begin sampling stage 1. But the difference between your experience with the last workflow and this one could be a million things, don't really know where to start suggesting fixes besides make sure that your models are loading correctly, and that you're not using ultra-sized refs without downscaling to <2 MP, stuff like that which would sometimes sneakily bog down performance since the h3 node is set to "max".
@foxydits Thx for your input. Appreciate your workflows :)
there is a known issue with the Fast Groups Bypasser by rgthree (options)
If you're using ComfyUI with Nodes 2.0 enabled, the Fast Groups Bypasser may not display or function correctly. Several users have reported this issue. If you're experiencing problems, try disabling Nodes 2.0 in your ComfyUI settings — this resolves the display and functionality issues.
Yeah, Nodes 2.0 is a plague upon comfyUI. It's the same with mxtools' sliders. I forgot to mention it because so many people have Nodes 2.0 turned off already that I don't even think about it sometimes.
How much vram is recommended to run this? I have 16gb vram and 64gb of ram, the finale upscale pass has been running for 2 hours and seems to be hung up. video length was set to 10 sec at .3mp
Definitely shouldn't hang up like that for 10 sec at 0.3. Possibly your VRAM was being held by something else? If you have all 16 GB of VRAM to be utilized you'll get roughly 50-100second steps on the upscale from 0.3 MP -> 1.2 MP (2x latent upscale)
Upscale fails for me too (can't even do 1 pass). Its faster to generate in the res i plan on upscaling to, so i assume something is not quite working
@Zombovich Oh it works just fine. It's more a question of latent size. What you two aren't understanding is that the sampler's no different than any other workflow's sampler. The latent upscale has already occurred by the time it gets to the sampler that you're getting stuck on. At that juncture, there's literally 0 difference between it and another workflow's sampler. Your systems are simply bogged down, likely by the size of the latent you're trying to gen. It's easy to forget that 0.5 MP upscaled 2x is actually 2.0 MP, it's 4x the pixels. So if you're going for something that high res you have to make sure your system has enough resources to handle it without bottlenecking.
@foxydits I wasn't clear. I was meaning its more an issue of comfy, node, or something specific with the hardware/setup.
@Zombovich I see. I guess I'm just responding generically 'cause a lot of people think 0.5 MP -> 2x upscale = 1 MP, when it's really 2 MP, and so there are a lot of "wow I can't even upscale, something's wrong with the workflow!" comments circulating.
Try lowering the upscale factor, starting with 1.5x and raising it gradually each new generation, until it freezes.
I could get my 16GB VRAM up to 0.5MP with x1.7 upscaling without freezing for hours.
The same thing happens to me, but not all of the time. Sometimes the upscaler pass takes 200 seconds. Sometimes it takes a almost an hour if I don't stop it and run it again. I also have a 16GB GPU and 64 GB system RAM.
@potatometer350 It all just depends on how much system resources you have at the moment you start the upscale. When you push against the edge of what your system can handle, it either holds on to the cliff's edge, or falls off. The model wasn't designed to run 10+ second videos at 1.5+ MP. It's designed around 3-15s at 1.0 MP for most consumer cards. So when we're upscaling to 1080p it can be quite a shock to your system when the sampler gets that fat juicy latent.
@foxydits I was clearing the RAM and VRAM before starting the upscale step. Most of the time it still doesn't work. Just need more firepower, I guess.
i got this error : Node threw an error during execution.
# ComfyUI Error Report
## Error Details
- Node ID: 133:129
- Node Type: SamplerCustomAdvanced
- Exception Type: AttributeError
- Exception Message: AttributeError: 'NoneType' object has no attribute 'show_progress_bar'
## Stack Trace
```
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\execution.py", line 545, in execute
output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\execution.py", line 344, in get_output_data
return_values = await asyncmap_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-Lora-Manager\py\metadata_collector\metadata_hook.py", line 177, in async_map_node_over_list_with_metadata
results = await original_map_node_over_list(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\execution.py", line 318, in asyncmap_node_over_list
await process_inputs(input_dict, i)
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\execution.py", line 306, in process_inputs
result = f(**inputs)
^^^^^^^^^^^
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\comfy_api\internal\__init__.py", line 149, in wrapped_func
return method(locked_class, **inputs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\comfy_api\latest\_io.py", line 1990, in EXECUTE_NORMALIZED
to_return = cls.execute(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\comfy_extras\nodes_custom_sampler.py", line 1052, in execute
callback = latent_preview.prepare_callback(guider.model_patcher, sigmas.shape[-1] - 1, x0_output)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\latent_preview.py", line 117, in prepare_callback
previewer = get_previewer(model.load_device, model.model.latent_format)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-videohelpersuite\videohelpersuite\latent_preview.py", line 100, in get_latent_video_previewer
previewer = get_latent_video_previewer.__wrapped__(device, latent_format, args, *kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\latent_preview.py", line 99, in get_previewer
taesd.first_stage_model.show_progress_bar = False
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
```
## System Information
- ComfyUI Version: 0.33.0
- Arguments: ComfyUI\main.py --windows-standalone-build
- OS: win32
- Python Version: 3.12.10 (tags/v3.12.10:0cc8128, Apr 8 2025, 12:21:36) [MSC v.1943 64 bit (AMD64)]
- Embedded Python: true
- PyTorch Version: 2.8.0+cu128
## Devices
- Name: cuda:0 NVIDIA GeForce RTX 3090 Ti : cudaMallocAsync
- Type: cuda
- VRAM Total: 25756696576
- VRAM Free: 24405475328
- Torch VRAM Total: 33554432
- Torch VRAM Free: 25034752
## Logs
```
2026-08-24T04:25:16.986092 - Premultiply image with mask2026-08-24T04:25:16.986092 -
2026-08-24T04:25:16.986092 - Image Resize Factor (mtb)2026-08-24T04:25:16.986092 - : 2026-08-24T04:25:16.987093 - Extracted mostly from WAS Node Suite, with a few edits (most notably multiple image support) and less features.2026-08-24T04:25:16.987093 -
2026-08-24T04:25:16.987093 - Save Image Grid (mtb)2026-08-24T04:25:16.987093 - : 2026-08-24T04:25:16.987093 - Save all the images in the input batch as a grid of images.2026-08-24T04:25:16.987093 -
2026-08-24T04:25:16.988093 - Load Image From Url (mtb)2026-08-24T04:25:16.988093 - : 2026-08-24T04:25:16.988093 - Load an image from the given URL2026-08-24T04:25:16.988093 -
2026-08-24T04:25:16.988093 - Save Gif (mtb)2026-08-24T04:25:16.989093 - : 2026-08-24T04:25:16.989093 - Save the images from the batch as a GIF2026-08-24T04:25:16.989093 -
2026-08-24T04:25:16.990093 - Export With Ffmpeg (mtb)2026-08-24T04:25:16.990093 - : 2026-08-24T04:25:16.990093 - Export with FFmpeg (Experimental)2026-08-24T04:25:16.990093 -
2026-08-24T04:25:16.990093 - Latent Lerp (mtb)2026-08-24T04:25:16.991094 - : 2026-08-24T04:25:16.991094 - Linear interpolation (blend) between two latent vectors2026-08-24T04:25:16.991094 -
2026-08-24T04:25:16.991094 - Image Remove Background Rembg (mtb)2026-08-24T04:25:16.992093 - : 2026-08-24T04:25:16.992093 - Removes the background from the input using Rembg.2026-08-24T04:25:16.992093 -
2026-08-24T04:25:16.993093 - Float To Number (mtb)2026-08-24T04:25:16.994093 - : 2026-08-24T04:25:16.994093 - Node addon for the WAS Suite. Converts a "comfy" FLOAT to a NUMBER.2026-08-24T04:25:16.994093 -
2026-08-24T04:25:16.994093 - Int To Bool (mtb)2026-08-24T04:25:16.994093 - : 2026-08-24T04:25:16.995092 - Basic int to bool conversion2026-08-24T04:25:16.995092 -
2026-08-24T04:25:16.995092 - Int To Number (mtb)2026-08-24T04:25:16.995092 - : 2026-08-24T04:25:16.995092 - Node addon for the WAS Suite. Converts a "comfy" INT to a NUMBER.2026-08-24T04:25:16.996137 -
2026-08-24T04:25:16.996137 - Transform Image (mtb)2026-08-24T04:25:16.996137 - : 2026-08-24T04:25:16.996137 - Save torch tensors (image, mask or latent) to disk, useful to debug things outside comfy2026-08-24T04:25:16.996137 -
2026-08-24T04:25:16.997138 - Load Image Sequence (mtb)2026-08-24T04:25:16.997138 - : 2026-08-24T04:25:16.997138 - Load an image sequence from a folder. The current frame is used to determine which image to load.2026-08-24T04:25:16.997138 -
2026-08-24T04:25:16.997138 - Save Image Sequence (mtb)2026-08-24T04:25:16.998137 - : 2026-08-24T04:25:16.998137 - Save an image sequence to a folder. The current frame is used to determine which image to save.2026-08-24T04:25:16.998137 -
2026-08-24T04:25:17.000137 - [1m[33m[WARNING][0m Traceback (most recent call last):
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\nodes.py", line 2263, in load_custom_node
module_spec.loader.exec_module(module)
File "<frozen importlib._bootstrap_external>", line 999, in exec_module
File "<frozen importlib._bootstrap>", line 488, in callwith_frames_removed
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfy_mtb\__init__.py", line 183, in <module>
from .endpoint import endlog
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfy_mtb\endpoint.py", line 9, in <module>
import requirements
ModuleNotFoundError: No module named 'requirements'
2026-08-24T04:25:17.000137 - [1m[33m[WARNING][0m Cannot import C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfy_mtb module for custom nodes: No module named 'requirements'
2026-08-24T04:25:17.002137 - [1m[33m[WARNING][0m Traceback (most recent call last):
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\nodes.py", line 2263, in load_custom_node
module_spec.loader.exec_module(module)
File "<frozen importlib._bootstrap_external>", line 999, in exec_module
File "<frozen importlib._bootstrap>", line 488, in callwith_frames_removed
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfy_vid2vid\__init__.py", line 8, in <module>
from .sd import load_checkpoint_guess_config, load_lora_for_models
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfy_vid2vid\sd.py", line 3, in <module>
from comfy.sd import load_model_weights, ModelPatcher, VAE, CLIP, model_lora_keys_unet, model_lora_keys_clip
ImportError: cannot import name 'load_model_weights' from 'comfy.sd' (C:\comfyui\ComfyUI-Easy-Install\ComfyUI\comfy\sd.py)
2026-08-24T04:25:17.003137 - [1m[33m[WARNING][0m Cannot import C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfy_vid2vid module for custom nodes: cannot import name 'load_model_weights' from 'comfy.sd' (C:\comfyui\ComfyUI-Easy-Install\ComfyUI\comfy\sd.py)
2026-08-24T04:25:17.003137 -
2026-08-24T04:25:17.003137 - Initializing ControlAltAI Nodes2026-08-24T04:25:17.003137 -
2026-08-24T04:25:17.041263 -
2026-08-24T04:25:17.041263 - 🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱2026-08-24T04:25:17.041263 -
2026-08-24T04:25:17.041263 - 🌱 Eden ComfyUI Pack maintained by 2026-08-24T04:25:17.041263 - https://eden.art/ 🌱2026-08-24T04:25:17.041263 -
2026-08-24T04:25:17.041263 - 🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱🌱2026-08-24T04:25:17.041263 -
2026-08-24T04:25:17.041263 -
2026-08-24T04:25:17.224908 - [32m[INFO][0m Using pytorch attention
2026-08-24T04:25:17.241423 - (RES4LYF) Init
2026-08-24T04:25:17.242423 - (RES4LYF) Importing beta samplers.
2026-08-24T04:25:17.263435 - (RES4LYF) Importing legacy samplers.
2026-08-24T04:25:17.282146 -
2026-08-24T04:25:17.282146 - [rgthree-comfy] Loaded 48 epic nodes. 🎉2026-08-24T04:25:17.282146 -
2026-08-24T04:25:17.282146 -
2026-08-24T04:25:17.282146 - [rgthree-comfy] ComfyUI's new Node 2.0 rendering may be incompatible with some rgthree-comfy nodes and features, breaking some rendering as well as losing the ability to access a node's properties (a vital part of many nodes). It also appears to run MUCH more slowly spiking CPU usage and causing jankiness and unresponsiveness, especially with large workflows. Personally I am not planning to use the new Nodes 2.0 and, unfortunately, am not able to invest the time to investigate and overhaul rgthree-comfy where needed. If you have issues when Nodes 2.0 is enabled, I'd urge you to switch it off as well and join me in hoping ComfyUI is not planning to deprecate the existing, stable canvas rendering all together.
2026-08-24T04:25:17.282146 -
2026-08-24T04:25:17.285143 - [💾 save_image_extended]2026-08-24T04:25:17.285143 - AVIF is not supported. To add it: pip install pillow pillow-avif-plugin2026-08-24T04:25:17.285143 -
2026-08-24T04:25:17.285645 - [💾 save_image_extended]2026-08-24T04:25:17.285645 - JXL is not supported. To add it: pip install jxlpy2026-08-24T04:25:17.285645 -
2026-08-24T04:25:17.285645 - [💾 save_image_extended]2026-08-24T04:25:17.285645 - You will need a valid MSVC env to build the wheel2026-08-24T04:25:17.285645 -
2026-08-24T04:25:17.286649 - [💾 save_image_extended]2026-08-24T04:25:17.286649 - version: 2.882026-08-24T04:25:17.286649 -
2026-08-24T04:25:17.293645 - [1m[33m[WARNING][0m Traceback (most recent call last):
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\nodes.py", line 2263, in load_custom_node
module_spec.loader.exec_module(module)
File "<frozen importlib._bootstrap_external>", line 999, in exec_module
File "<frozen importlib._bootstrap>", line 488, in callwith_frames_removed
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\teacache\__init__.py", line 1, in <module>
from .nodes import NODE_CLASS_MAPPINGS as NODES_CLASS, NODE_DISPLAY_NAME_MAPPINGS as NODES_DISPLAY
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\teacache\nodes.py", line 12, in <module>
from comfy.ldm.lightricks.model import precompute_freqs_cis
ImportError: cannot import name 'precompute_freqs_cis' from 'comfy.ldm.lightricks.model' (C:\comfyui\ComfyUI-Easy-Install\ComfyUI\comfy\ldm\lightricks\model.py)
2026-08-24T04:25:17.293645 - [1m[33m[WARNING][0m Cannot import C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\teacache module for custom nodes: cannot import name 'precompute_freqs_cis' from 'comfy.ldm.lightricks.model' (C:\comfyui\ComfyUI-Easy-Install\ComfyUI\comfy\ldm\lightricks\model.py)
2026-08-24T04:25:17.297156 - [VNCCS] Automatic legacy migration is disabled. Use the VNCCS Migration Assistent node to migrate legacy sheets.2026-08-24T04:25:17.297156 -
2026-08-24T04:25:17.834906 - WAS Node Suite: 2026-08-24T04:25:17.834906 - OpenCV Python FFMPEG support is enabled2026-08-24T04:25:17.834906 -
2026-08-24T04:25:17.834906 - WAS Node Suite 2026-08-24T04:25:17.834906 - Warning: 2026-08-24T04:25:17.834906 - ffmpeg_bin_path is not set in C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\was-node-suite-comfyui\was_suite_config.json config file. Will attempt to use system ffmpeg binaries if available.2026-08-24T04:25:17.834906 -
2026-08-24T04:25:18.250876 - WAS Node Suite: 2026-08-24T04:25:18.250876 - Finished.2026-08-24T04:25:18.250876 - 2026-08-24T04:25:18.251874 - Loaded2026-08-24T04:25:18.251874 - 2026-08-24T04:25:18.251874 - 2202026-08-24T04:25:18.251874 - 2026-08-24T04:25:18.251874 - nodes successfully.2026-08-24T04:25:18.251874 -
2026-08-24T04:25:18.251874 -
2026-08-24T04:25:18.251874 - "Success is not about being better than someone else. It's about being better than you used to be."2026-08-24T04:25:18.251874 - - Unknown2026-08-24T04:25:18.251874 -
2026-08-24T04:25:18.251874 -
2026-08-24T04:25:18.266213 - [32m[INFO][0m
Import times for custom nodes:
2026-08-24T04:25:18.266213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\Image-Captioning-in-ComfyUI
2026-08-24T04:25:18.266213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\websocket_image_save.py
2026-08-24T04:25:18.267213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI_bnb_nf4_fp4_Loaders
2026-08-24T04:25:18.267213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\Comfyui-Memory_Cleanup
2026-08-24T04:25:18.267213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-mxtoolkit
2026-08-24T04:25:18.267213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\SeedVarianceEnhancer-main
2026-08-24T04:25:18.267213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-ZImage-LoRA-Merger
2026-08-24T04:25:18.267213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-image-selector
2026-08-24T04:25:18.267213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\wlsh_nodes
2026-08-24T04:25:18.267213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyLiterals
2026-08-24T04:25:18.267213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\janus-pro
2026-08-24T04:25:18.267213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\loadlorawithtags
2026-08-24T04:25:18.267213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI_JPS-Nodes
2026-08-24T04:25:18.267213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\cg-use-everywhere
2026-08-24T04:25:18.267213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-inpaint-cropandstitch
2026-08-24T04:25:18.267213 - [32m[INFO][0m 0.0 seconds (IMPORT FAILED): C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui_nvidia_rtx_nodes
2026-08-24T04:25:18.267213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-obvpm
2026-08-24T04:25:18.267213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\topazvideoai
2026-08-24T04:25:18.267213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI_Steudio
2026-08-24T04:25:18.267213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-SCAIL-Pose
2026-08-24T04:25:18.268213 - [32m[INFO][0m 0.0 seconds (IMPORT FAILED): C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfy_vid2vid
2026-08-24T04:25:18.268213 - [32m[INFO][0m 0.0 seconds (IMPORT FAILED): C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\teacache
2026-08-24T04:25:18.268213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI_Ib_CustomNodes
2026-08-24T04:25:18.268213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-QwenImageLayeredToLength
2026-08-24T04:25:18.268213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-ReservedVRAM
2026-08-24T04:25:18.268213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\Comfyui-QwenEditUtils
2026-08-24T04:25:18.268213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-WD14-Tagger
2026-08-24T04:25:18.268213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-Florence2
2026-08-24T04:25:18.268213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-TiledDiffusion
2026-08-24T04:25:18.268213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-VFI
2026-08-24T04:25:18.268213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-MelBandRoFormer
2026-08-24T04:25:18.268213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-AutoCropFaces
2026-08-24T04:25:18.268213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\WWAA-CustomNodes
2026-08-24T04:25:18.268213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-various
2026-08-24T04:25:18.268213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\SmartSaveImage
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfy-image-saver
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\save-image-extended-comfyui
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-WanAnimatePreprocess
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyMath
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI_essentials
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-Custom-Scripts
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-Image-Saver
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-GGUF
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\WhatDreamsCost-ComfyUI
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds (IMPORT FAILED): C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI_Searge_LLM
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui_image_metadata_extension
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\gguf
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI_UltimateSDUpscale
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-VoxCPMTTS
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\controlaltai-nodes
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-sol-attn
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI_SaveImageWithMetaDataUniversal
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\rgthree-comfy
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI_Qwen3-VL-Instruct
2026-08-24T04:25:18.269213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI_Comfyroll_CustomNodes
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-QwenVL
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui_controlnet_aux
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-Koolook
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\vnccs
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-PlagueKind-Nodes
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\eden_comfy_pipelines
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-kjnodes
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-videohelpersuite
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\vnccs-utils
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI_LayerStyle_Advance
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-LTXVideo
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-IG-Nodes
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.0 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI_tinyterraNodes
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.1 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI_RyanOnTheInside
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.1 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-Impact-Subpack
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.1 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI_LayerStyle
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.1 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-Impact-Pack
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.1 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-MediaMixer
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.1 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-vrgamedevgirl
2026-08-24T04:25:18.270213 - [32m[INFO][0m 0.1 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-GIMM-VFI
2026-08-24T04:25:18.271213 - [32m[INFO][0m 0.2 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-Lora-Manager
2026-08-24T04:25:18.271213 - [32m[INFO][0m 0.2 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\RES4LYF
2026-08-24T04:25:18.271213 - [32m[INFO][0m 0.3 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-SeedVR2_VideoUpscaler
2026-08-24T04:25:18.271213 - [32m[INFO][0m 0.4 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-itools
2026-08-24T04:25:18.271213 - [32m[INFO][0m 0.5 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-WanVideoWrapper
2026-08-24T04:25:18.271213 - [32m[INFO][0m 0.7 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-Crystools
2026-08-24T04:25:18.271213 - [32m[INFO][0m 0.9 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\was-node-suite-comfyui
2026-08-24T04:25:18.271213 - [32m[INFO][0m 1.1 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-EasyColorCorrector
2026-08-24T04:25:18.271213 - [32m[INFO][0m 1.2 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-manager
2026-08-24T04:25:18.271213 - [32m[INFO][0m 1.2 seconds (IMPORT FAILED): C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfy_mtb
2026-08-24T04:25:18.271213 - [32m[INFO][0m 1.3 seconds (IMPORT FAILED): C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-nunchaku
2026-08-24T04:25:18.271213 - [32m[INFO][0m 1.4 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-HackAfterDark-Nodes
2026-08-24T04:25:18.271213 - [32m[INFO][0m 1.6 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-rmbg
2026-08-24T04:25:18.271213 - [32m[INFO][0m 3.2 seconds: C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-Easy-Use
2026-08-24T04:25:18.271213 - [32m[INFO][0m
2026-08-24T04:25:18.273213 - [32m[INFO][0m Context impl SQLiteImpl.
2026-08-24T04:25:18.273213 - [32m[INFO][0m Will assume non-transactional DDL.
2026-08-24T04:25:18.353599 - [32m[INFO][0m Using RAM pressure cache.
2026-08-24T04:25:18.371610 - [32m[INFO][0m [LoRA-Manager] Deduplicate: removed 0 history rows, 0 queue rows, 0 orphaned queue rows
2026-08-24T04:25:18.373607 - [32m[INFO][0m Starting server
2026-08-24T04:25:18.418124 - [32m[INFO][0m [LoRA-Manager] Recipe cache miss: performing full directory scan
2026-08-24T04:25:18.447644 - [1m[33m[WARNING][0m [LoRA-Manager] Duplicate filename conflict detected: 10 lora filename(s) are shared by 20 files total, causing ambiguity in Lora resolution. Open the Doctor panel to resolve one-click.
2026-08-24T04:25:18.459142 - [32m[INFO][0m [LoRA-Manager] Lora cache hydrated from persisted snapshot with 908 models
2026-08-24T04:25:18.463142 - [32m[INFO][0m [LoRA-Manager] Checkpoint cache hydrated from persisted snapshot with 71 models
2026-08-24T04:25:18.467646 - [32m[INFO][0m [LoRA-Manager] Embedding cache hydrated from persisted snapshot with 79 models
2026-08-24T04:25:18.467646 - [32m[INFO][0m To see the GUI go to: http://127.0.0.1:8188
2026-08-24T04:25:18.507183 - [32m[INFO][0m [LoRA-Manager] Recipe cache initialized in 0.13 seconds. Found 0 recipes
2026-08-24T04:25:18.512184 - [32m[INFO][0m [LoRA-Manager] AutoV3 backfill: updated 0/2 models for checkpoint
2026-08-24T04:25:18.527931 - [32m[INFO][0m [LoRA-Manager] FTS index validated, reusing existing index with 0 recipes
2026-08-24T04:25:19.642879 - [1m[33m[WARNING][0m [DEPRECATION WARNING] Detected import of deprecated legacy API: /scripts/ui/components/buttonGroup.js. This is likely caused by a custom node extension using outdated APIs. Please update your extensions or contact the extension author for an updated version.
2026-08-24T04:25:19.645996 - [1m[33m[WARNING][0m [DEPRECATION WARNING] Detected import of deprecated legacy API: /scripts/ui.js. This is likely caused by a custom node extension using outdated APIs. Please update your extensions or contact the extension author for an updated version.
2026-08-24T04:25:19.646996 - [1m[33m[WARNING][0m [DEPRECATION WARNING] Detected import of deprecated legacy API: /extensions/core/clipspace.js. This is likely caused by a custom node extension using outdated APIs. Please update your extensions or contact the extension author for an updated version.
2026-08-24T04:25:19.647793 - [1m[33m[WARNING][0m [DEPRECATION WARNING] Detected import of deprecated legacy API: /extensions/core/widgetInputs.js. This is likely caused by a custom node extension using outdated APIs. Please update your extensions or contact the extension author for an updated version.
2026-08-24T04:25:19.648794 - [1m[33m[WARNING][0m [DEPRECATION WARNING] Detected import of deprecated legacy API: /extensions/core/groupNode.js. This is likely caused by a custom node extension using outdated APIs. Please update your extensions or contact the extension author for an updated version.
2026-08-24T04:25:19.956409 - HackAfterDark: Loaded 1 presets from C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-HackAfterDark-Nodes\presets.2026-08-24T04:25:19.956409 -
2026-08-24T04:25:21.640467 - FETCH ComfyRegistry Data: 5/1772026-08-24T04:25:21.640467 -
2026-08-24T04:25:21.680998 - [VNCCS] list_characters: failed to list new character path 'C:\comfyui\ComfyUI-Easy-Install\ComfyUI\output\VNCCS\Characters': [WinError 3] The system cannot find the path specified: 'C:\\comfyui\\ComfyUI-Easy-Install\\ComfyUI\\output\\VNCCS\\Characters'2026-08-24T04:25:21.680998 -
2026-08-24T04:25:21.680998 - [VNCCS] list_characters: failed to list new character path 'C:\comfyui\ComfyUI-Easy-Install\ComfyUI\output\VNCCS\Characters': [WinError 3] The system cannot find the path specified: 'C:\\comfyui\\ComfyUI-Easy-Install\\ComfyUI\\output\\VNCCS\\Characters'2026-08-24T04:25:21.680998 -
2026-08-24T04:25:21.680998 - [VNCCS] list_characters: failed to list new character path 'C:\comfyui\ComfyUI-Easy-Install\ComfyUI\output\VNCCS\Characters': [WinError 3] The system cannot find the path specified: 'C:\\comfyui\\ComfyUI-Easy-Install\\ComfyUI\\output\\VNCCS\\Characters'2026-08-24T04:25:21.680998 -
2026-08-24T04:25:21.680998 - [VNCCS] list_characters: failed to list new character path 'C:\comfyui\ComfyUI-Easy-Install\ComfyUI\output\VNCCS\Characters': [WinError 3] The system cannot find the path specified: 'C:\\comfyui\\ComfyUI-Easy-Install\\ComfyUI\\output\\VNCCS\\Characters'2026-08-24T04:25:21.680998 -
2026-08-24T04:25:21.681995 - [VNCCS] list_characters: failed to list new character path 'C:\comfyui\ComfyUI-Easy-Install\ComfyUI\output\VNCCS\Characters': [WinError 3] The system cannot find the path specified: 'C:\\comfyui\\ComfyUI-Easy-Install\\ComfyUI\\output\\VNCCS\\Characters'2026-08-24T04:25:21.681995 -
2026-08-24T04:25:21.681995 - [VNCCS] list_characters: failed to list new character path 'C:\comfyui\ComfyUI-Easy-Install\ComfyUI\output\VNCCS\Characters': [WinError 3] The system cannot find the path specified: 'C:\\comfyui\\ComfyUI-Easy-Install\\ComfyUI\\output\\VNCCS\\Characters'2026-08-24T04:25:21.681995 -
2026-08-24T04:25:21.681995 - [VNCCS] list_characters: failed to list new character path 'C:\comfyui\ComfyUI-Easy-Install\ComfyUI\output\VNCCS\Characters': [WinError 3] The system cannot find the path specified: 'C:\\comfyui\\ComfyUI-Easy-Install\\ComfyUI\\output\\VNCCS\\Characters'2026-08-24T04:25:21.681995 -
2026-08-24T04:25:21.681995 - [VNCCS] list_characters: failed to list new character path 'C:\comfyui\ComfyUI-Easy-Install\ComfyUI\output\VNCCS\Characters': [WinError 3] The system cannot find the path specified: 'C:\\comfyui\\ComfyUI-Easy-Install\\ComfyUI\\output\\VNCCS\\Characters'2026-08-24T04:25:21.681995 -
2026-08-24T04:25:21.681995 - [VNCCS] list_characters: failed to list new character path 'C:\comfyui\ComfyUI-Easy-Install\ComfyUI\output\VNCCS\Characters': [WinError 3] The system cannot find the path specified: 'C:\\comfyui\\ComfyUI-Easy-Install\\ComfyUI\\output\\VNCCS\\Characters'2026-08-24T04:25:21.681995 -
2026-08-24T04:25:21.681995 - [VNCCS] list_characters: failed to list new character path 'C:\comfyui\ComfyUI-Easy-Install\ComfyUI\output\VNCCS\Characters': [WinError 3] The system cannot find the path specified: 'C:\\comfyui\\ComfyUI-Easy-Install\\ComfyUI\\output\\VNCCS\\Characters'2026-08-24T04:25:21.681995 -
2026-08-24T04:25:21.681995 - [VNCCS] list_characters: failed to list new character path 'C:\comfyui\ComfyUI-Easy-Install\ComfyUI\output\VNCCS\Characters': [WinError 3] The system cannot find the path specified: 'C:\\comfyui\\ComfyUI-Easy-Install\\ComfyUI\\output\\VNCCS\\Characters'2026-08-24T04:25:21.681995 -
2026-08-24T04:25:21.681995 - [VNCCS] list_characters: failed to list new character path 'C:\comfyui\ComfyUI-Easy-Install\ComfyUI\output\VNCCS\Characters': [WinError 3] The system cannot find the path specified: 'C:\\comfyui\\ComfyUI-Easy-Install\\ComfyUI\\output\\VNCCS\\Characters'2026-08-24T04:25:21.681995 -
2026-08-24T04:25:21.681995 - [VNCCS] list_characters: failed to list new character path 'C:\comfyui\ComfyUI-Easy-Install\ComfyUI\output\VNCCS\Characters': [WinError 3] The system cannot find the path specified: 'C:\\comfyui\\ComfyUI-Easy-Install\\ComfyUI\\output\\VNCCS\\Characters'2026-08-24T04:25:21.681995 -
2026-08-24T04:25:21.682994 - [VNCCS] list_characters: failed to list new character path 'C:\comfyui\ComfyUI-Easy-Install\ComfyUI\output\VNCCS\Characters': [WinError 3] The system cannot find the path specified: 'C:\\comfyui\\ComfyUI-Easy-Install\\ComfyUI\\output\\VNCCS\\Characters'2026-08-24T04:25:21.682994 -
2026-08-24T04:25:21.684994 - [VNCCS] list_characters: failed to list new character path 'C:\comfyui\ComfyUI-Easy-Install\ComfyUI\output\VNCCS\Characters': [WinError 3] The system cannot find the path specified: 'C:\\comfyui\\ComfyUI-Easy-Install\\ComfyUI\\output\\VNCCS\\Characters'2026-08-24T04:25:21.686490 -
2026-08-24T04:25:21.686490 - [VNCCS] list_characters: failed to list new character path 'C:\comfyui\ComfyUI-Easy-Install\ComfyUI\output\VNCCS\Characters': [WinError 3] The system cannot find the path specified: 'C:\\comfyui\\ComfyUI-Easy-Install\\ComfyUI\\output\\VNCCS\\Characters'2026-08-24T04:25:21.686490 -
2026-08-24T04:25:26.587585 - [1m[33m[WARNING][0m [DEPRECATION WARNING] Detected import of deprecated legacy API: /scripts/ui/components/button.js. This is likely caused by a custom node extension using outdated APIs. Please update your extensions or contact the extension author for an updated version.
2026-08-24T04:25:28.424297 - FETCH ComfyRegistry Data: 10/1772026-08-24T04:25:28.424297 -
2026-08-24T04:25:28.711965 - [32m[INFO][0m got prompt
2026-08-24T04:25:28.733506 - [1m[31m[ERROR][0m Failed to validate prompt for output 34:
2026-08-24T04:25:28.733506 - [1m[31m[ERROR][0m * (prompt):
2026-08-24T04:25:28.733506 - [1m[31m[ERROR][0m - Required input is missing: images
2026-08-24T04:25:28.735007 - [1m[31m[ERROR][0m * VHS_VideoCombine 34:
2026-08-24T04:25:28.735007 - [1m[31m[ERROR][0m - Required input is missing: images
2026-08-24T04:25:28.735007 - [1m[31m[ERROR][0m Output will be ignored
2026-08-24T04:25:28.813085 - [32m[INFO][0m VAE load device: cuda:0, offload device: cpu, dtype: torch.float32
2026-08-24T04:25:29.021123 - [32m[INFO][0m VAE load device: cuda:0, offload device: cpu, dtype: torch.float16
2026-08-24T04:25:29.094829 - [32m[INFO][0m Found quantization metadata version 1
2026-08-24T04:25:29.096340 - [32m[INFO][0m Using MixedPrecisionOps for text encoder
2026-08-24T04:25:30.688039 - [32m[INFO][0m CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
2026-08-24T04:25:30.718589 - [32m[INFO][0m Requested to load MiniMaxH3VideoVAE
2026-08-24T04:25:30.809292 - [32m[INFO][0m Model MiniMaxH3VideoVAE prepared for dynamic VRAM loading. 4965MB Staged. 0 patches attached. Force pre-loaded 128 weights: 348 KB.
2026-08-24T04:25:31.697030 - [32m[INFO][0m Model MiniMaxH3VideoVAE prepared for dynamic VRAM loading. 4965MB Staged. 0 patches attached. Force pre-loaded 128 weights: 348 KB.
2026-08-24T04:25:31.772161 - [32m[INFO][0m Requested to load MiniMaxH3TEModel_
2026-08-24T04:25:31.925482 - [32m[INFO][0m Model MiniMaxH3TEModel_ prepared for dynamic VRAM loading. 14956MB Staged. 0 patches attached. Force pre-loaded 410 weights: 4572 KB.
2026-08-24T04:25:36.003162 - FETCH ComfyRegistry Data: 15/1772026-08-24T04:25:36.005197 -
2026-08-24T04:25:37.964007 - [32m[INFO][0m Found quantization metadata version 1
2026-08-24T04:25:37.964007 - [32m[INFO][0m Detected mixed precision quantization
2026-08-24T04:25:37.965007 - [32m[INFO][0m Using mixed precision operations
2026-08-24T04:25:37.965007 - [32m[INFO][0m Native ops: int8_tensorwise, convrot_w4a4, asym_w4a8_int8 , emulated ops: float8_e4m3fn, nvfp4, float8_e5m2, mxfp8
2026-08-24T04:25:37.986278 - [32m[INFO][0m model weight dtype torch.bfloat16, manual cast: torch.bfloat16
2026-08-24T04:25:37.988278 - [32m[INFO][0m model_type FLOW_AV
2026-08-24T04:25:38.222224 - [32m[INFO][0m [H3Utils] SLA installed | sparsity=0.90 | BLK=64x64 | min_seq_len=8192 | dense_last_steps=0 | protect_audio=True
2026-08-24T04:25:38.331680 - [32m[INFO][0m Model MiniMaxH3VideoVAE prepared for dynamic VRAM loading. 4965MB Staged. 0 patches attached. Force pre-loaded 128 weights: 348 KB.
2026-08-24T04:25:38.495397 - [32m[INFO][0m Model MiniMaxH3VideoVAE prepared for dynamic VRAM loading. 4965MB Staged. 0 patches attached. Force pre-loaded 128 weights: 348 KB.
2026-08-24T04:25:38.601348 - [1m[33m[WARNING][0m WARNING: No VAE weights detected, VAE not initalized.
2026-08-24T04:25:38.613907 - [1m[31m[ERROR][0m !!! Exception during processing !!! 'NoneType' object has no attribute 'show_progress_bar'
2026-08-24T04:25:38.619410 - [1m[31m[ERROR][0m Traceback (most recent call last):
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\execution.py", line 545, in execute
output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\execution.py", line 344, in get_output_data
return_values = await asyncmap_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-Lora-Manager\py\metadata_collector\metadata_hook.py", line 177, in async_map_node_over_list_with_metadata
results = await original_map_node_over_list(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\execution.py", line 318, in asyncmap_node_over_list
await process_inputs(input_dict, i)
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\execution.py", line 306, in process_inputs
result = f(**inputs)
^^^^^^^^^^^
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\comfy_api\internal\__init__.py", line 149, in wrapped_func
return method(locked_class, **inputs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\comfy_api\latest\_io.py", line 1990, in EXECUTE_NORMALIZED
to_return = cls.execute(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\comfy_extras\nodes_custom_sampler.py", line 1052, in execute
callback = latent_preview.prepare_callback(guider.model_patcher, sigmas.shape[-1] - 1, x0_output)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\latent_preview.py", line 117, in prepare_callback
previewer = get_previewer(model.load_device, model.model.latent_format)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-videohelpersuite\videohelpersuite\latent_preview.py", line 100, in get_latent_video_previewer
previewer = get_latent_video_previewer.__wrapped__(device, latent_format, args, *kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfyui\ComfyUI-Easy-Install\ComfyUI\latent_preview.py", line 99, in get_previewer
taesd.first_stage_model.show_progress_bar = False
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'NoneType' object has no attribute 'show_progress_bar'
2026-08-24T04:25:38.621410 - [32m[INFO][0m [32mPrompt executed in 9.88 seconds[0m
2026-08-24T04:25:43.835796 - FETCH ComfyRegistry Data: 20/1772026-08-24T04:25:43.835796 -
```
## Attached Workflow
Please make sure that workflow does not contain any sensitive information such as API keys or passwords.
```
Workflow too large. Please manually upload the workflow from local file system.
```
## Additional Context
(Please add any additional context or steps to reproduce the error here)
Which sampler threw that error? 1st pass right when you started your first gen? If somehow you've turned off all the samplers in 1st pass maybe that could happen. I don't know though, you're the first to report this specific one. So it gives this error if you load up a fresh version of the workflow and hit run?
Sampler #2 at the subgraph
@foxydits you put 3 VAE model link at the left side but workflow model loader has 2 VAE loader
@ehsanbbaxe the int8 video vae is optional choice to use over the regular official video vae.
Your sampler crashes because the tiny "preview" model that ComfyUI uses to show you live previews during sampling is missing. The code blindly tries to use it without checking it exists. The fix for this is one file: grab "taeh3.safetensors" (~9.8 MB) from huggingface.co/Kijai/MiniMax-H3-TAE (vae_approx subfolder) into models/vae_approx/, restart ComfyUI, rerun.
@miaoying The thing you told me to do, I had already done it before. Now that you asked me to double-check it again, I checked it, but I am still getting the same error.
@ehsanbbaxe Thanks for checking, that tells us the file's presence isn't the fix. Your log shows why: "WARNING: No VAE weights detected, VAE not initalized."
My gues is, it looks like ComfyUI found the file but couldn't get usable weights from it (or looked in the wrong folder), so it treated it as "no preview model" and crashed. This error is only in the preview path, the sampling itself is fine.
Quickest fix: Settings (gear icon, bottom-left) → Preview Method → None → rerun. The render completes normally; you just lose the preview strip in the browser.
One more thing: your log shows you're loading a different H3 repack than the standard Comfy-Org one, which can also mean the model never references the preview file, in that case, previews off is the answer.
@miaoying thanks so much bro fixed! musch love
*** UPDATE EDIT ***
Correction to my earlier warning — the author's right, and his numbers were the missing piece.
The real budget isn't megapixels alone — it's resolution × clip length (how much total video the upscale pass must hold alongside the ~20 GB model). On a 24 GB card, per his own tested numbers:
0.5mp: fine up to ~8–9 s
0.4mp: fine up to ~15 s
My crash was a 10 s clip at 0.5mp — past the 0.5mp ceiling, stalling at the upscale pass's model reload (GPU busy-waiting at ~24 GB, frozen on "Model Initializing").
And fair on setup: I run --reserve-vram 1 --disable-async-offload, which tightens headroom vs a stock launch — startup flags matter. If anyone stalls: shorten the clip → drop to 0.4mp → relaunch without those flags, in that order.
***********
Heads-up for 24 GB cards — the 0.5 to 2.0MP settings need ~32 GB
Great workflow, but I hit a hard stall running the settings quoted in the video attached (0.5mp / 8 steps / 4-step upscale) on an RTX 4090 (24 GB), and wanted to warn others before they burn an hour on it.
Symptom: first pass renders fine, upscaler completes, then the upscale pass freezes at "Model Initializing" — GPU at 100% but only ~125 W (busy-wait, not computing), VRAM pinned at ~23.7 GB, zero progress for 20+ minutes. The interrupt button didn't even work; I had to kill the whole ComfyUI server.
Cause (as far as I can tell): the H3 model itself needs ~20 GB of the 24 GB card, leaving only ~4 GB for the upscaled video. At 0.5mp the 2× upscaler produces a ~2.0mp upscale pass, which doesn't fit in that headroom, so the GPU spins forever instead of failing cleanly.
What works on 24 GB: 0.4mp base / 6 steps / 3-step upscale — same workflow, no stall, and honestly hard to tell the difference in output. The step counts don't seem to be the issue — it's specifically the 0.5mp that tips it over the VRAM wall.
Would be great if you could note the GPU/VRAM you tested with in the description — I suspect 0.5mp is fine on a 5090 or larger card, just not 24 GB.
I have a 3090 (24 GB as well) and have no problem 2x upscaling 0.5 MP up to around 8-9 seconds. 0.4 MP can go up to 15 seconds easily. While I can appreciate your personal analysis, you're leading others who have 24GB by saying they'll get stuck. I don't know your setup but the reality is that not everyone will run into the same issues as you. Plenty of people with 16 GB cards not getting stuck either. Sometimes it comes down to just not multitasking and using your computer for other things while genning, sometimes it's a startup flag that gives VRAM more headroom, causing system RAM to take over, sometimes it's having the VRAM chunking option turned on at a high setting.
Fair, I appreciate the correction. Looking at it again, the piece I under-weighted is clip length: my workflow is 10 s at 0.5mp, which is past your own 8–9 s ceiling for 0.5mp on a 24 GB card, so that's the actual boundary, not resolution alone. Also fair on startup flags, I run --reserve-vram 1 --disable-async-offload, which tightens headroom vs a stock launch. So maybe the warning should read: 0.5mp ≈ 8–9 s max on 24 GB; beyond that, drop to 0.4mp or shorten the clip. That fits your numbers. Want me to update my comment to say it that way?
@miaoying You updating is appreciated. I'm pretty sensitive to reviews that aim to scare people off with headlines. What I wish I explained better in my video or made more clear up front is that by default, a 0.5 MP latent upscale with this model leads to a 2 MP output, which is waaaay larger than what people normally run if they're just doing a native gen.
Why is the background in my video always distorted?
Just the background? I'm not sure. Are you using references or just T2V? Is it blurry in the 1st stage as well as the final result?
@foxydits im having a similar issue, the output is worse than the sample! if i use the h3 upscaler on its own then the output is fine
@pup_alph What do you mean on its own? How are you running it on its own? Also what upscale model do you have selected in the "Minimax H3 Latent Upscaler (3D)" node? the _bf16.safetensors version?
@foxydits i mean running Minimax H3 Latent Upscaler (3D)" node, _bf16.safetensors version in a standalone workflow, not in your seed hunter workflow,, its amazing work, and i most likely need to play with setting more, its not quick to test using my 3080ti lol (the 64gb of ram i luckily bought before the ram prices went through the roof helps though)
Im not sure if this is a bug or if I'm just doign somethign wrong. I would run the seed hunter and find on i liked, then i selected it from the top 3 option selector, then the button to the right in red. And it renders all 3 videos, or sometimes not the one i selected
Make sure Nodes 2.0 is turned off in your comfy settings. The WF will never regenerate stage 1 sample gens unless you've changed something that would cause comfyUI to think they need to be regenned. How comfyUI works is, if the produced latent is the exact same, clicking run again simply does nothing since it detects that nothing has changed and it would be a waste of time. If you alter a prompt, change a setting (vid length, step count, the strength of a LoRA, ANYTHING pertaining to how stage 1 generated), it will regen because that alters their outcomes. So just like in my tutorial video, just do a T2V super fast low MP, low duration test: make sure the "Final Upscale Pass" toggle is off, then run the model, let all 3 samples in stage 1 finish until the WF says "Idle" and all 3 samples are showing. Then select sample #1 or #2 or #3 from the toggle up to, turn on "Final Upscale Pass", and then click run again. Don't even use the Manual Random Seed buttons, I don't want you accidentally clicking the green one on the left which forces Stage 1 to regenerate.
@foxydits yep it was definitely off and I didn't change anything. When I do the manual roll on the second column to the right i see the green bar process the other two; is that normal?
@The_Last_Goblin_King Yeah, comfyUI just reloads the VHS previews on all 3. It's quick, it's not regenning 1st stage. Afterwards it takes your selected latent to finalization.
I don't know if this can be fixed or not, but apparently even if you just open one of the previews to get a better look, it will start generating everything again when you do the upscale phase.
That is not true. ComfyUI only regens first stage if something materially changes in terms of values or settings that would affect stage 1. Comfy is designed not to 'redo' work it's already done -- if you finish genning something and click 'run' again, it will instantaneously complete again if everything would result in the same outcome. The reality is, you have accidentally changed something material to stage 1's output, or something in your comfy's setup is wrong. But right clicking -> Open Preview in tab is not a material change, and therefore cannot/will not ever be the reason stage 1 regens when moving to Final Upscale pass. Keep in mind, previous samples will quickly decode again from stage 1, causing the green loading bar to appear very quickly as the VHS node loads the images again. Perhaps you're seeing that and thinking they're regenning without realizing it's all normal.
@foxydits Nah it was doing a whole 10 step regen. I must have changed something without realizing it. Thanks for the reply!
I’ve been a fan since the LTX days.
This Minimax version is truly amazing!
I was deeply impressed.
There’s no need to pay any attention to low-quality comments.
I’ll continue to support you.
Oh it's fine, there are some legitimate problems with different peoples' setups, but mostly it's just people who aren't as experienced with heavy workflows like this. Anyway I really appreciate your support!
Absolutely the best workflow I've tested so far for minimax. Overall, I don't see the point of turbo lora; the speedup is about 20-30%, the sound is terrible, and motion is much worse, as is prompt understanding. Which, on a 6th generation, can result in 5 bad and 1 good. It's better to spend those "few" minutes and usually have 2/3 good generations.
Apparently, there are already methods for improving sound, such as ComfyUI-H3-AudioRefine, but I haven't tested it yet. If it works, you can add it to the workflow someday.
If you want to make money and improve our lives at the same time, I recommend creating your own template on Runpod, which will automatically download all the models and nodes and install things like Sage Attention – I promise I'll be one of the first users.
Try 12 steps at 0.5 strength for the turbo lora, it'll improve sounds and motion, possibly a good middle ground
Appreciate the review. Currently the only way I 'try to make money' is just by offering my free patreon to people, and some of them choose to support at a higher tier. As for adding stuff, it's been tough figuring out what needs to be added or changed in the workflow. I'm gonna give it a couple days to settle before putting new stuff in that makes people go "argh why do I have to download more nodes?!"
Anyone know why my "options" node and the "select latent to proceed... " are just empty boxes?
Yes, you have Nodes 2.0 turned on. Turn them off in your comfy settings. It's a horrible option that comfyUI is pushing, even though 95% of people turn it off at some point.
@foxydits No, it's not that, as I already have that turned off.. Unless i'm missing something els?
@HaakonMild Oh, that's usually the issue. The toggles are from the rgthree node pack. I guess just make sure you have that properly installed, and that your comfyUI is up to date. I don't know what else to say.. rgthree's a really popular node pack and it should work.
My system is RTX 5080 with 64GB RAM. Idle VRAM before comfyui starts processing is around 2.3-2.5GB VRAM. I don't do anything else when comfyui is running.
I'm doing ref2va 16:9 at 15s. I have 3 refs (520 x 1412, 966 x 543, 1143 x 642). I can do 0.8MP with 6 steps, no upscale. But I can't even do 0.3MP with 6 steps + upscale 1.5/2.0 with 3 steps. It just OOM. Is this normal?
I am also getting OOM error on upscale phase meanwhile I can even do 0.7mp 10-20 steps without a problem.
0.3 MP -> 2x upscale is 1.2 MP though, which is quite large for this model at 15 seconds. Gotta keep in mind the sampler's just doing its best to handle the large latent with your available VRAM and memory. 16 GB VRAM is a bit of a minimum for an upscale workflow. Just keep in mind that 2x LATENT upscale is 4x the MP.
@foxydits How do you calculate the 1.2 MP? And what settings do you recommend for upscale with 16GB VRAM?
@teiji25762 2x latent upscale = 4x the megapixels. so 0.3 MP x 4 = 1.2 MP
@foxydits Thanks. I went down to 0.3 10s and can do upscale 1.9. I guess this is my limit without sacrificing too much quality.
@teiji25762 Yeah, I feel you. It's easy to look at tiny numbers like 0.3 MP and forget they balloon with upscale so quickly to much bigger resolutions than the model can handle on normal consumer hardware. It feels like little baby gens turn into giants so quickly with latent upscaling in this case.
Good workflow, only issue is that when Its finished generating the first passes and i select one i like, i press the manual random seed on the second pass node and it regenerates the first 3 passes and doesnt keep the ones from the previous run. I havent changed anything other than enabling the 2nd pass and selecting the one I like the most. any way I fix this?
The only way comfyUI will regen stage 1 is if it detects something prior to it has changed. It's programmed that way. So according to comfy, you DID change something that affects stage 1. Now, when you do it correctly, you will see each stage in 1 light up as it reloads the VHS previews -- but that's quick. However some people think that's regenning and it's a false alarm. Curious what version of comfy you use, desktop or portable? I've never had a problem with this so without prior experience I can't say where things are going wrong.
@foxydits if i had to guess, its maybe because i had the seed on the first pass set to random instead of fixed? just a wild guess.
@LewdSpirit well.. yeah.. a random seed is probably the biggest change that could possibly exist for causing stage 1 to regen every time, lol. There's a reason I have it set to fixed. You click "Manual Random Seed" to get new stage 1's.
@foxydits Yeah that was my fault then lol. thank you for helping me out. the workflow is great otherwise!
This is a great workflow, but is there a way to get the PNG file to save the metadata (prompt, seed, etc)?
Just enable the "save metadata" toggle on the "Final Video" node on the right side of the video. It'll save the .mp4 output with the metadata.
My database won't parse the metadata from mp4 files. I'll see if I can update it.
is it possible to completely skip audio generation?
Uh, well, not really. It is a very small part of the latent processing so it wouldn't speed up your genning by very much.
This is amazing!
Is it possible that you can implement this: clicking the "Manual Random Seed" (green) will automatically turn off the "Enable Final Upscale Pass" before it runs? I keep on forgetting to turn upscale off. lol
LMAO brother it's like looking into a mirror. I understand your pain. And no, I would have if I could. Comfy is not very programmable unless you want to make your own custom nodes, which is not my jam.
What I do is sit there and watch it after starting the upscale, and once it gets past the latent upscaler node, I go ahead and turn it off so it's ready to go for the next run. You could probably turn it off right after starting, but I wait anyway.
@potatometer350 So few people have ever built 2-run workflows in comfyUI that these little hacks are kind of golden nuggets. Like for instance, if you're genning the full three stage 1's but the first one that pops out is an absolute banger and you don't want to wait for the other two to finish... you can just cancel the run, toggle off the other two stage 1 previews that never finished, and then toggle final upscale on with the selected latent and proceed. When you know the rules, it becomes really flexible and efficient.
Missing connection
Required input slots have no connection feeding them.
RIFE Frame Interpolation - images
If you watch the video I mention this. It's a false error that doesn't prevent the workflow from working whatsoever. Basically a nothing burger, a side-effect of a multi-run workflow like mine (where you run the workflow twice before you get a finalized output).
Works like a charm.. however noticed when i try to do 15 second upscale on 2nd pass it stays stuck at 0/3.. my first pass has no issue generating 15 second at 0.5mp, in my second pass Im trying to do 1.0 megapixel upscale, any idea why it would stay stalled on the upscale?
When you say 1.0 megapixel upscale, you mean you switched the upscale mode to 'megapixels' and chose 1.0 as the value? If so, and you have a beefy system, it should complete if your system resources aren't bogged down (15s at 1.0 MP is still a hefty resource requirement, but certainly doable). But if your 1st pass is 0.5 MP and you left your mode on 'scale by multiplier' at 2.0x (like the workflow's default is), that's not 1.0MP.... 0.5MP -> 2x latent upscale = 4.0 MP. Just to be clear.
@foxydits Yes that's correct, selected the megapixels upscale mode. My system is a ASUS 5070TI 16GBVRAM / Intel(R) Core(TM) i9-14900K (3.20 GHz) / 64 GB RAM
I'd consider this beefy, but maybe not as beefy as I imagine lol
@foxydits Might be worth mentioning I'm using H3 Eros Max INT8 checkpoint instead of the fl2va, maybe that's the factor tbh, even though first render runs perfectly fine.
@CypriotWolf I don't think that's the major factor. Your system should be able to run a 15 sec 1.0 MP upscale. I don't know if you have a bunch of other programs running when you gen, such as Steam or a video editor or photoshop, stuff that might lock down your available RAM or VRAM. It's so hard to troubleshoot without knowing all the variables. I'd suggest trying to get one complete video, 5 seconds at your target MP. Then try 10 seconds. See where the failpoint happens when your system is fresh and no resources are being directed anywhere but comfy.
Clean install of Comfy-UI 0.33.1, but I'm missing some nodes not mentioned above.
Here is the error I'm getting:
This workflow was created with a newer version of ComfyUI (0.33.1). Some nodes may not work correctly.
Core nodes from version 0.33.0: MiniMaxH3AddGuide
--
Additionally, all the nodes in the pictures 1-6 do not load, with LoadImageCrop being from an unknown pack.
Does anyone have any ideas how to resolve this?
Thanks.
I think you need to set it to nightly. (comfyui that is).
Also if you like me a sampler problem; move "ComfyUI-MiniMaxH3-Cache" to place outside of custom_nodes.
It's a bit messy, but you need to manually install all the required nodes basically. Almost all of them can be downloaded with "git clone" commands into the custom node directory (that should solve the image crop thing), then u have to restart comfy ui for the nodes to appear. I had the same error for MiniMaxH3AddGuide, but that's a node that already should come with comfy ui, I don't remember exactly how I fixed it, but I think it prompted me something to attempt to fix it somehow.
The image loader node is really quite simple: both my video, workflow, and civit page explain that you have to manually install some of these nodes.
The Add Guides node is not required for the functionality of the workflow, so even if you never solved that, you can leave those options disabled and it will never be a problem. For some reason comfyUI removed them for like one update on the Desktop version, but it seems to be fine now. Blame them.
@randyrocketcorp497 Thanks! The main branch of Comfy just added it back in (0.34.0)
@foxydits Thanks! I messed up the LoadImageCrop installation, but the video made it clear! Thanks for the help.
This workflow is magic. I previously used your other workflow and for 15 sec video at 1mp it took me 24 mins. With this workflow it is 12mins 0.5mp with 1.5x upscale. PC specs: 5950x 3090 and 64gb ram
Is this better than comfy default Ref2v workflow? I have an RTX 3090 and 32GB ram. The gen struggles if video is involved. So this will improve speed, right?
The fl2va model plugged into the ref2va node is indeed better than the ref2va model used normally, in my opinion. All the same functionality but with the visual/audio of fl2va (which is better and more compatible with turbo loras).
If your goal is to generate 1.0 MP+ videos, then yes, this helps with speed over genning at that resolution from scratch. 6-12 steps at 0.3 MP is very quick, and that will upscale to 1.2 MP at a 2x latent upscale setting, of which the model only needs about 3 steps to clarify.
RIFE Frame Interpolation is missing a required input: images
If you watched even 30 seconds of my video you'd learn that it's a fake error that doesn't matter even in the slightest. Some nodes aren't designed for workflows like mine that run in two separate segments; the 1st pass, and then the 2nd pass in different runs. The 'red outline' and error notification doesn't prevent the 1st pass or workflow from functioning whatsoever.
Great Workflow, I love it !
But is it specialized for 24Gb VRAM ? I have a 16 Gb VRAM and it takes around 5 min for 8 sec video but you show in your preview that it takes less than a minute so idk what's the problem
I have RTX 5080. For 6 steps, 0.3, 16:9, 10s, 3 ref, it takes ~5min for all 3 previews to finish. I have comfy kitchen + Sparse SLA + turbo enabled.
It depends on prompt complexity; reference images - reference videos I have a 3090 and my vram usage sits around average 22gb~
Without any references and a simple prompt I can generate all 3 previews in 120s~ at 5s / IT
If you use a high resolution video as your reference video the IT / s will take a huge hit
@teiji25762 so in this case kitchen>sage? 5090 btw
@evantopsmith They both are about the same speed with almost no quality loss. During a few tests I did, kitchen was faster for me (only a tiny bit though) and since it's official comfyui implementation, I pick it.
You should do your own tests.
@Patikitten Hello, could you please help me. I have a 4070 with 12gb of vram but always get an error that I ran out of gpu memory. I am using first to last frame to generate a 4 sec video but cant even generate one preview. Help would be much appreciated!
@Patikitten May I ask what other settings are you using? I'm using Comfy Kitchen, Sol Attn, low ram setting, 0.5 megapixel, 16:9, 3 ref images, 1 preview only and it I'm getting 38s/IT. I already disabled pinned memory which help quite a bit.
I highly recommend this workflow. if you are beginner, take the time to read all, don't be afraid to breakdown the workflow in clear visible blocks to understand. be patient, it will be worth. i'm posting a video later :)
Fantastic workflow, compared it to one I had put a week into developing and just deleted it. Cohesive outputs at .8mp and look great upscaled.
The 3 previews before committing are a gamechanger vs gambling on bulk outputs
This is a huge time saver for my RTX 3080TI 12GB. The 3 previews at 0.3 take around 3 minutes, then 1-2 minutes for the upscale. Been using this workflow for a few days already. Thanks! Looking forward to what 1.1 will bring.
Hello, could you please help me. I have a 4070 with 12gb of vram but always get an error that I ran out of gpu memory. I am using first to last frame to generate a 4 sec video but cant even generate one preview. I am assuming hen you say 0.3 you mean megapixels. I am using the same setting. Help would be much appreciated!
@uusevit It can be a lot of things. I am firstly on Arch Linux (btw) with 32GB or RAM. If I feed it videos I also get OOM. I use only image references. I have only on Comfy Kitchen and Sparse.
I just had an idea that would make this workflow the absolute king. You could have a feedback set of nodes which would let you continue a new generation, starting with the frame that the previous generation ended at (without having to manually take a screenshot and manually do everything). Not sure how hard or feasible this idea is, but it would be crazy good if possible.
That workflow exists, it's called Motion Context. It's okay. I'm a filmmaker, so 15 second shots are about all I need. Most cinematic shots never extend past 15 seconds, cuts happen on average every 3-8 seconds. I use previous shots' references in future gens to make sure everything stays consistent. Also, the next update of this workflow has a video extension feature. You load a video, it takes 22 frames from the end and seamlessly continues it in a new gen.
@foxydits oh, ok, so the next version of the workflow has this feature or something very similar?
@rombo Yep, coming a little later today, just gotta finalize it.
Amazing workflow. Thanks! and Is there any guide how to use shift?
Thank you! You'd simply follow the LoRA instructions that typically are why you would change it. Some LoRAs specify the values, such as 6.0/3.0, or 12.0/6.0, etc.
Can I use the Ref2VA model in this workflow? If so, how?
You can just load it in the top left where you select which model loads. fl2va and ref2va function identically in this workflow, except fl2va has better lora support and better visuals/audio. This workflow defaults to fl2va because it can do everything the ref2va model can, but better.
your previous workflow had "deroping", right? but thats not in this workflow anymore? I guess because its very slow?
Thanks for this workflow. I've been using it for a few days now and it's putting out amazing quality at 0.5MP with the latent upscale. I'll be posting a few videos later.
I noticed a slight change in sound and animation with the final video after upscaling. The first pass often looks and sounds a bit different than the final. Often not a big deal but sometimes it deletes a nuance I'd have loved to keep. How can I make sure the upscaled video is exactly the same as the video from the first pass?
Also, you've set the MP for the first image to 0.91 and the MP for the following images much lower. I tried setting the second image to 0.91 as well and ran a 15 seconds first pass at 0.5. Second pass gave me an OOM on a 5090. In your experience how much impact does the MP of images 2-x have on the output?
The 0.91 is just a mistake. I actually meant to release the workflow with it set to 0, which is unlimited size. I don't like limiting people, the workflow is all about flexibility and choice. That said, the references aren't likely to cause OOM's. It's almost ways the latent size. 0.5 MP -> 2x latent size -> balloons up to 2.0 MP (4x the pixels). That's 2x the pixels the model is trained for.
And yeah, technically the video is just being re-sampled during upscale -- there is new noise and it's being denoised in 3, 4 or 5 steps. It has access to the same references, but with a 2x latent size they're not applying as fully as you'd expect. I had to work around this with my new feature coming in today's update, video extension. That one works like a charm, no change from 1st pass to upscale. But that's unwieldy to achieve with normal gen, I'm afraid.
@foxydits Thank you. Looking forward to the updated workflow and the extension feature.
Amazing work, as always! You've quickly become my favorite workflow creator. Just one issue with this one that I've been having: The audio from the upscale pass tends to deviate a lot from the audio in the low-res sample. It still coheres with the video, but it can be a bit frustrating when the low-res sample nails the audio, and you just want a higher-quality version of that in the upscaled video. Is there a way to mitigate this?
Hmm, I haven't really noticed degradation this way. Perhaps it has to do with LoRAs to some degree. I've stopped using Turbo LoRAs, running 12 steps 1st pass, 4 steps final pass, with great results. If you wanted to take the audio from your 1st pass and keep it clean for your 2nd pass, you could always audio decode the split audio from the upscale pass (it's the AV split node is hidden underneath the Upscale Latent node near the top of the Upscale pass). Then just take that decoded audio and shove it into the Final Output node's audio slot.
@foxydits Yeah, I might have to try to the turboless approach as well. Oddly, when I try to use 4 steps instead of 3 on the 2nd pass on longer videos (0.5MP 1st pass * 12s+), I get OOMs even on my 5090. I didn't think the number of steps would affect VRAM usage.
@Hookflash yeah, 12s at 2.0 MP (1080p) is a lot for any system. But the OOM gets caused by an initial spike usually, so just make sure none of your system is being used for anything else if you're bumping up against the barrier like that.
Can this workflow be used for T2V ? I turned off all the image options, but using it as T2V didn't produce the results I expected
Yes, it works very well for T2V. As long as all of the <Picture> references are off. I'm curious what you mean by the results weren't what you expected?
T2V works just fine. Did you leave the steps set to 6? If so depending on what you are making you might have to go higher like 12 steps with 0.5 turbo or 25 steps no turbo
@bhopping Personally, I don't use the turbo LoRAs anymore. I do 12 steps 1st pass, 4 steps final pass. That's how I made my most recent short film and it came out looking great.
@foxydits I checked the issue, and it turns out that one of the LoRAs was causing it. After disabling the LoRA, the image no longer has that noisy-looking appearance. However, it might just be psychological, because I still feel like something looks a little strange.
@zens Yeah, I mean it all depends on your variables -- what MP you're genning at, how many steps, what sampler/scheduler, what version of the model you're using, etc.
@foxydits I'm currently using minimax_h3_fl2va_pruned_int8_convrot.safetensors.
My current settings are:
Sampler: RES
Scheduler: Simple
Steps: 14 + 3
Video length: 12 seconds
Resolution: 0.5 → 1 MP
I was wondering if the current workflow's second-pass upscaling could potentially be upgraded to a tiled/chunked upscaling method instead.
With my RTX 4080 and 128 GB of system RAM, doing the second pass at 1.2× takes roughly 1,000–2,000 seconds, which seems extremely slow.
Is there something wrong with my settings or workflow, or is this expected behavior with H3? Would it be possible to replace the current second-pass upscaling with a tiled/chunked approach to significantly reduce the processing time and VRAM usage?
@zens You are correct in that something's wrong if your upscale is taking that long. For 12 second videos, 0.5 MP -> 1.0 MP takes about 3-5 minutes on my 3090 with 3-4 steps. It sounds like something else is using your system resources. I only get speeds like that if I have other VRAM hogs open like a video editor, and so the model has no room to utilize VRAM. There's a great node/plugin called Crystools-Monitor. You should install to see within comfy what your system resources are being utilized for. There's nothing extraneous in my workflow that would slow down the upscale, it's literally just your system + the model trying to sample the newly sized latent.
The other suggestion is to make sure your nodes are updated, especially the Upscale Node. There's an option 'force unload' and 'enable temporal chunking' that should help prepare your system for the sudden large latent upscale pass.
Best!
What's this error I get every time? Invalid input
The value for Model Preview Override's tiny_vae is not available.
In the bright red note on the left side of the workflow titled "CUSTOM NODE LINKS" -- you should see a link to download taeh3.safetensors. that goes in the model preview 'tiny vae' selection.
@foxydits TX
Legit one of the best workflows I've ever used. Fast!