⚡ Recommendation
You can use my custom checkpoint V2 NSFW, Lightning included:
➡️ WAN 2.2 Enhanced – Lightning Edition
for wf lightningedition long v2.1
https://civarchive.com/models/2053259?modelVersionId=2477539
Or for wf svi , no Lightning included:
➡️ SVI NSFW
https://civarchive.com/models/2053259?modelVersionId=2609141
Update: 2 or 3 KSamplers
The triple-KSampler workflow added dynamism to avoid slowdowns, but it could cause issues in certain sections, such as high brightness or other bugs. To fix this, simply enable the Double KSampler in the problematic section and change the seed.
🟣 uptade SVI LoRAs – Strengths & Weaknesses
⚡ Overview
✅ Strengths
Best solution for making long videos
Excellent transitions between video segments
Reduced degradation compared to other solutions
Strong character coherence: the model retains information from the previous video, helping maintain consistency
⚠️ Weaknesses
Weaker prompt understanding
Weaker camera understanding
Videos are less dynamic
Sometimes slow-motion effect
(can be improved with proper Lightning LoRAs and dynamic prompts)
🟣 SVI LoRAs – Download Links
⚡ Download
Note: Both LoRAs must be loaded manually in your workflow.
🟣 Suggested Lightning LoRA Combos (Optional)
⚡ Overview
You don’t have to use these Lightning LoRA combos. They are optional and allow you to fine-tune motion and degradation.
You can also use other Lightning LoRAs or assign different combos per video for more control.
🔥 Combo 1 – More Motion (Rapid Video Degradation)
High LoRA →
https://huggingface.co/Kijai/WanVideo_comfy/blob/709844db75d2e15582cf204e9a0b5e12b23a35dd/Lightx2v/lightx2v_I2V_14B_480p_cfg_step_distill_rank128_bf16.safetensors
Weight: 4
or
💜 Combo 2 – Less Image Degradation
or
or
🟢 Combo 3 – Balanced Motion / Moderate Degradation
High LoRA →
https://huggingface.co/Kijai/WanVideo_comfy/blob/709844db75d2e15582cf204e9a0b5e12b23a35dd/Lightx2v/lightx2v_I2V_14B_480p_cfg_step_distill_rank128_bf16.safetensors
Weight: 3
Low LoRA →
https://huggingface.co/Kijai/WanVideo_comfy/blob/709844db75d2e15582cf204e9a0b5e12b23a35dd/Lightx2v/lightx2v_I2V_14B_480p_cfg_step_distill_rank128_bf16.safetensors
Weight: 1.5
If you want to create several-minute-long videos while maintaining high quality, it is possible, but it will take a very long time. You just need to avoid using Lightning LoRAs and use the full model instead.
🧠 Advanced Usage Tip
You can disable global Lightning LoRAs and assign different combos per video:
Combo 1 for Video 1, Combo 3 for Video 2, Combo 2 for Video 3.
Each combo produces different motion and degradation behavior.
If you don’t use Lightning LoRAs, degradation will be even lower, but generation time will be very long.
🧠 Dynamic Prompts – Better Control & More Motion
You can use dynamic prompts to have better control and this helps to make the video more dynamic.
You just need to give this example prompt to an LLM like ChatGPT. It will be enough for it to describe the image you have and what you want as a video while keeping the same structure of the prompt in the following example.
⚠️ This will be a NSFW prompt; ChatGPT will not accept it.
You can use GROK https://grok.com, which can make NSFW prompt modifications.
For more examples of prompts with different poses, check:
Enhanced FP8 Model.
Give these prompts to GROK with the dynamic prompt structure if you want.
Example 1
(At 0 seconds: Wide shot showing a slightly overweight man casually walking down a city street, camera fixed in front, urban environment with buildings and cars.)
(At 1 second: Suddenly, a massive shark bursts from the pavement ahead, looking terrifying at first, pavement cracking, dust and debris flying, camera from side angle.)
(At 2 seconds: Medium shot from the side, the man stumbles backward in shock, while the shark dramatically slows down and strikes a comically exaggerated sexy pose, revealing large, exaggerated shark breasts, covered by a colorful bikini.)
(At 3 seconds: Close-up on the man’s face, eyes wide in disbelief, as he turns to look at the shark, small cartoon-style hearts floating above his head to emphasize his amazement, camera slightly low-angle.)
(At 4 seconds: Dynamic travelling shot showing the man frozen in the street, the shark maintaining its sexy pose, water splashes and debris still moving realistically, urban chaos around.)
(At 5 seconds: Wide cinematic shot pulling back, showing the man standing in the street, staring at the bikini-wearing shark with hearts above his head, epic perspective highlighting absurdity and humor.)
Example 2 – Anime NSFW
(At 0 seconds: The couple in a cozy bedroom, anime style, soft lighting highlighting their intimate embrace, her back arched slightly as he positions himself.)
(At 1 second: The man’s hips moving rhythmically, the head of his penis sliding effortlessly into her vagina, her body responding with a gentle, fluid motion, anime-style motion lines emphasizing the smooth penetration.)
(At 2 seconds: Her back arching deeply against him to intensify the pleasure, hips swaying with each thrust, breasts bouncing subtly, small hearts floating around them to capture the erotic energy.)
(At 3 seconds: Her face, eyes closed in bliss, a soft moan escaping, hands resting behind her head, anime-style blush on her cheeks, the air filled with a seductive aura.)
(At 4 seconds: The man penetrating her deeply, her body moving in sync with his, the bed sheets slightly rumpled, the room’s warm lighting enhancing the intimate, lustful atmosphere.)
(At 5 seconds: The couple locked in a passionate embrace, the scene exuding vibrant, seductive energy, anime style with smooth lines and soft shadows.)
findperfectresolution
https://github.com/ashtar1984/comfyui-find-perfect-resolution
rgtree
https://github.com/rgthree/rgthree-comfy
kj-node
https://github.com/kijai/ComfyUI-KJNodes
iamccs
https://github.com/IAMCCS/IAMCCS-nodes
🔧 What’s New – V2.1
The workflow is now much simpler than before.
You can create videos up to 20 seconds.
New sections can be added very easily by copying and pasting an existing section.
🎨 New Node: PainterI2VAdvanced
🔗 https://github.com/princepainter/ComfyUI-PainterI2Vadvanced
Main Parameters
color_protect
Enables or disables color drift correction.
Recommended: True.
Prevents green tint, darkening, and color distortion.
correct_strength
Controls the strength of color correction.
Higher value = stronger color correction
findperfectresolution
https://github.com/ashtar1984/comfyui-find-perfect-resolution
_______________________________________________________________________________________________
New FunVace workflow: it's extremely easy to use. If you have separate videos and want to join them with a seamless transition, use this! Thanks to Bob once again, excellent work: https://civarchive.com/models/2277993?modelVersionId=2563927
Workflow Update – Lightning Edition V2
📌 New Version Available
The Lightning Edition V2 is the latest update of my workflow.
This version is cleaner, better organized, and scene transitions are improved for smoother results.
🔧 Fixes & Improvements – Lightning Edition V2
Workflow fully reorganized and clarified for better readability.
Scene transitions improved for smoother results.
Optimized nodes and settings for easier use.
Confirmed compatibility with the Lightning Edition V2 model.
🔹 V1
Version without Super Power LoRA
A version without Super Power LoRA has been added for those who are more cautious.
If you want to try the Super Power LoRA without using the folder included in the ZIP, check the GitHub page:
➡️ GitHub – rgthree workflow
Tutorial
The workflow is divided into 6 sections:
Section 1: generates a 5-second video.
Section 5: handles transitions and interpolation.
Important notes:
Running only sections 1 + 2 will not merge the video.
To create a 15-second video, run sections 1 + 2 + 5.
Interpolation can also be enabled in section 1 if you prefer.
You can use different prompts and LoRAs for each section.
Resolution is automatically adjusted to avoid cropping.
💥 Super Power LoRA – V1
Requires the custom rgthree node:
➡️ rgthree ComfyUIInstall via git clone or the Manager, then replace the folder with the one included in the ZIP.
Allows you to save and load templates, very useful when working with multiple LoRAs, especially for WAN 2.2.
A version without Super Power LoRA is also available for those who prefer.
For the version with Super Power LoRA, see: GitHub – rgthree workflow
✅ Tips – V1
Keep the default settings for fast results.
Increase the steps if you want better quality.
Description
FAQ
Comments (159)
When creating your workflow second video, do you use sample images as a reference, or do you use the final scene from the first video as a reference?
It takes the last image, but also the last 5 frames to understand the motion. You can increase this value to more than 5.
@taek75799 When a woman is giving a handjob, and especially when ejaculating, she lowers her head and closes her eyes.
Even if I add a negative prompt or use LORA, the video always contains scenes of her lowering her head and closing her eyes. How can I fix this?
@OQQO Try it in the positive prompt instead. Wan understands head movements very well, such as 'she lifts her head,' 'looks at the man,' or 'looks at the camera.' Also, try facial expressions like 'she smiles' or 'she shouts'
Great job, Workflows, thank you!
The easiest way to activate it is simply add the --use-sage-attention flag in the .bat/.sh file you use to launch comfy
Correct. Install sage and triton via pip using python - do that in venv mode. Then do as the other comment mentioned and append your .bat file - you do not need to do anything in ComfyUI. If you do not understand what i said, have GPT translate it and walk you through exact steps. When ComfyUI log starts up it will show sage attention is in use, that means its activated
@wodz30 thank you
If possible, could you create a version of the workflow with the integrated SageAttention node?
Hello, click the small icon at the top right of the subgraph node (the one with the 1st prompt). The subgraph will open, and you will be able to enable Sage Attention.
Thanks, man! I'm not yet familiar with subgraph nodes; I didn't know you could do that. The whole workflow is much clearer now, thanks!
awesome workflow, simple and provides great results...
Top workflow, but i recommend using euler beta57 with shift 5.7, 6 steps (2 hi and 4 low steps). Why is the motion value set to 1.3 when you recommend 1.15 (and why do you recommend this value), and the colour value set to 0.05 instead of 0? Small typo: 2nd Lora low has a wrong name (3LORA ...).
Hello, thanks for your feedback! These are the default values from Painter's repo, but of course, they can be changed as desired. This value adds more dynamism. The number of steps and other settings are a major debate, lol! Let's just say everyone adjusts them in their own way. And thanks for catching that little mistake, haha!
Yeah, the default amount of steps is not enough to get a decent video, I get artifacts. I just tried 6 steps (3 and 3) and it seems like that might be good enough. I use the Q4 version of the model though, so maybe that affects it? Or maybe it's the resolution?
The motion seems to be a little too fast with the default value. I feel like it's faster than using 1.15 with the old painter node.
@jszgwogylxjpqoahgx346 Hi! Thanks for your feedback. I think the base image plays a huge role, and of course, the video resolution as well. All my previews are done in 4 steps, but I get better results with 6 steps. They were made just to show that you could get something decent with only 4 steps. If you can try the Q6 and Q8 versions, you will get even better results.
Sorry for the beginner question.
The workflow didn't run without re-setting the "Unet Loader (GGUF) HIGH" and "Unet Loader (GGUF) LOW" nodes, so I downloaded files with matching names myself and configured them. When I tried generating, the video came out blurry with unnatural movement. Switching the "Unet Loader (GGUF) HIGH" node to the "NSFW FAST MOVE V2 Q8 H checkpoint" improves it significantly, but is this the correct/intended way to use the workflow?
Thanks in advance!
Interesting. I was just going to post a Thanks saying it worked perfectly for me? I just downloaded the WF and had to add the missing Painterly thing nodes and voilà, it worked. I didn't get any of the issues you are having. Is your Comfy up to date?
@yajukun
Thanks for the reply!
Yes, my ComfyUI is fully up to date.
I’m running it through StabilityMatrix.exe (the Stability Matrix launcher).
So far everything works perfectly for me when I replace the original
wan2.2_i2v_high_noise_14B_Q8_0.gguf and wan2.2_i2v_low_noise_14B_Q8_0.gguf
in the Unet Loader with
wan22EnhancedNSFWCameraPrompt_nsfwFASTMOVEV2Q8H.gguf and
wan22EnhancedNSFWCameraPrompt_nsfwFASTMOVEV2Q8L.gguf.
I just wanted to double-check whether this substitution is the intended/correct way to use these enhanced models.
@Jalkina @Jalkina Ahh, ok, I see what you're saying. I also downloaded and used his special models for HIGH and LOW with this workflow. My previous workflow also just uses the basic Q6 or Q8 models. This WF runs faster than my normal WF (which has 3 stages, 1 HIGH w/o lightening, 1 HIGH w/light and 1 LOW w/light). So I normally get very good results with my 3stages but it takes forever to run. Esp now that I gave up my desktop 4090 RTX for a 5090m laptop.
@izaya1 All the Enhanced NSFW | camera prompt adherence files are here, just find the correct version https://civitai.com/models/2053259
Отвал головы от такого рабочего процесса! огромное спасибо! работает из коробки))
fantastic,. thx very much!!!!
New FunVace workflow, it's extremely easy to use. If you have separate videos and want to join them with a seamless transition, use this! Thanks to Bob once again, excellent work: https://civitai.com/models/2277993?modelVersionId=2563927
Cool workflow! I'm still a beginner to this but I could get it running, works great and is really well organized.
One thing confuses me a bit; the instructions for longer videos say that if you turn on a second video and leave the seed for the first video unchanged the first video will not be re-generated again, only the second video. But following those instructions seem to still re-generate the first video as well, since the seed is unchanged it produces the same result but it's still time that is "wasted", no?
Have I just misunderstood the instructions and how it works or is something not working correctly? Thanks for the hard work!
Hi! On the subgraph node, click the small icon in the top right corner to open it. Set all noise seeds to 'fixed' on all KSamplers. Do the same for the other subgraphs. Normally, with the Seed RGTree node, this should have no impact since it is already set to 'fixed,' but I believe it depends on the ComfyUI version, and you likely have a version where this issue occurs
Thanks for the quick answer. Turns out I'm an idiot and had added a RAM cleanup node at the end of the workflow. This workflow seems to handle RAM much better than what I've been using before so I added it without thinking, but seems to work well without it here, oops lol.
Suggestion: Make the middle parts loopable/repeatable. So that the final clip could consist of parts 1, 2, 2, 2, 3, 3, 3, 3 and 4. So that stages 2 and 3 could repeat (some chosen number) times with a different seed to make longer clips without making new nodes. Great workflow, BTW. I'm having lots of fun with it now that I learned how to use it somewhat properly.
noticed: when something goes invisible or out of frame in between videos ( e.g. nipples get covered up ) and gets revealed again, it changes it´s likeness completely because of missing reference I guess. ) any idea for a workaround?
Hi! Yes, that's normal. The Wan long-video workflows are made in 5-second segments. The second video simply retrieves the last frame of the previous one. You just need to ensure that at the end of each video, you keep the elements you want for the next part.
Use SVI
Hi there, I'm currently using your Wan 2.2 Lightning Edition workflow and I'm having some trouble.
I cannot find where to adjust the Sampling Steps and CFG values. Are they locked within a specific subgraph or controlled by a specific node (like ModelSamplingDiscrete)? I'm running an RTX 3090, so I'd like to experiment with different settings to see if I can improve the detail.
Could you please guide me on where these parameters are located? Thank you!
Edit the subgraph then
I was trying to use the upscale feature but I am unable to get it working. Any ideas? Thanks again.
UpscaleModelLoader Weights only load failed. In PyTorch 2.6, we changed the default value of the `weights_only` argument in `torch.load` from `False` to `True`. Re-running `torch.load` with `weights_only` set to `False` will likely succeed, but it can result in arbitrary code execution. Do it only if you got the file from a trusted source. Please file an issue with the following so that we can make `weights_only=True` compatible with your use case: WeightsUnpickler error: Unsupported operand 60 Check the documentation of torch.load to learn more about types accepted by default with weights_only https://pytorch.org/docs/stable/generated/torch.load.html.Thanks, this is great. But breast bounce is a little over the top (even when walking slowly its bouncing a lot) and immersion breaking, anyone found a way to reduce it? I like the bouncy breasts when it makes sense, but its too much.
I also think it would be awesome if a workflow like this could be adjusted to use an input video, and then comfy generates a continuation of that video. Just like now the videos are being stitched together by taking the last frame of a generated video.
Hi! The normal version has less of that breast bounce issue, lol.
if i want to get into wan, with a 4060, how long would it take to generate a 5 sec vid?
Depends on resolution and steps (Quality) -> 10 to 30 minutes.
What @vicaut said is true - really depends on res / steps / etc. I am currently on a 4080 (16VRAM) with small portrait resolution of 512x768. Having Enable VIDEO2 10 SEC - my queue is showing about 100 second runtime with an average of 14.5 VRAM needed. Running res 512x512 still has it at 14.5vram. I am not super savvy when it comes to ComfyUI / Wan gguf BUUUUT I wonder if this workflow would benefit from having WanVideo Block Swap to help offload VRAM requirements (and use CPU - causing extra time per iteration).
I'm using Q4 gguf on my RTX 4090 16gb ram laptop and it takes roughly 145 seconds for 540x960 resolution and looks great.
@robomaraiart 4090 is 16gb on laptop? I thought those suckers were 24gb vram. Either way, yeah that's around how long it takes for me too. I'm starting to notice that video's 3 & 4 tend to hallucinate and not do what my prompt is asking. (video 1, 2 being a man waving, sitting) and video 3 askign the man to stand up again - I end up getting some weird handstand / zoomed in on face with a hand coming out of their face. HAHAHA. I guess I should be more detailed or something.
whatever i do i just cant get the 2nd video to run.
i flip the switch and the workflow just ignore it and doesn't run the 2nd video generation. (just finish the job in 2 seconds...).
Were you able to get it to work?
@rawrasaurussss134 no, i just switched to "Wan2.2 - continuous generation" which just works OOTB.
@nowsafe https://www.youtube.com/watch?v=uOZ0wLq1XoI&lc=UgwvKoecymzxPQyCoXR4AaABAg.ASP5pXyuvEeASQp4ook-sp
[If you don't want to click link, you can search "WAN 2.2 SVI 2 Pro — Long Continuous AI Video Generation (Beginner to Advanced) by Atelier Darren
Try this workflow with SVI PRO. It honestly is a gamechanger. I've made some edits to do a lora on a PER VID segment and added RIFE 47 for 60FPS.
Hi, thank you for the workflow. I have one question, is it possible to also have a first/end frame system ? Because I want my first 5s video to loop. Would it be possible to add it to this workflow? Is it easy to add?
Hi, it's not possible, but you can go to Painter's repo. Those are the nodes I use. He made a First/Last Frame workflow if you're interested. https://github.com/princepainter/Comfyui-PainterFLF2V
@taek75799 Thank you
@taek75799 "it's not possible"
Not working at all.
Works perfectly, you are doing something wrong.
@vicaut He is not working
What exactly isn't working?
Hello, the first 5-second video works, but not the other segments?
@taek75799 when you have the terminal open, do you have any errors the populate?
@rawrasaurussss134 None on my end, do you have any?
@taek75799 honestly - it's working just fine for me. Besides the obvious (did you enable it) and made sure its not bypassed - are you also checking to see if the "seed" is -1 so that you are 100% sure that its trying to generate a new pass. A good alternative is to take the workflow json that is "failing" and upload that to ChatGPT along with any terminal info/notes and see what it tells you. Sometimes a node may have disconnected
Your workflow has no SVI pro in it, so the transitions from one video to the next is not seamless, so if you add this node to video 2,3 and 4 the workflow would even be better.
This is an awesome workflow! thanks a lot!
I noticed 2 issues: the motion frame is always used as 1 because you have in the subgraph a "Select Images" set to -1 (it should be set to "-1 * motion frame") and following the same "select images", the "Scale Image" is not very useful because the image will be scaled down using bilinear https://github.com/princepainter/ComfyUI-PainterLongVideo/blob/main/nodes.py#L120
Dunno why it is so difficult to get anything to work in comfy... so i wanted to try 10 seconds video, i followed the indication to generate one first and activate the 10 seconds option...whenever i do this something shows up
Prompt execution failed
Prompt outputs failed validation: CLIPTextEncode: - Required input is missing: clip CLIPTextEncode: - Required input is missing: clip PainterLongVideo: - Required input is missing: vae VAEDecode: - Required input is missing: vae
I do not know what to do here, according to the workflow i shouldn't have to do anything other than activate the 10 seconds node.
Gotta AI, search for answer...that's how I learn and a lot screenshot for better guidance of what you want to do. Advise that you have a smart AI tool, that can guide you through it. Not something anyone can teach but Ai.
Hello, when you activate the 2nd subgraph (10sec), it becomes active. Check below to see if all the 'gets' are also activated.
@taek75799 best advice I use Grok Ai to help me organize my workflow ;) it works for me
I saw the clip joiner workflow which you recommended and found it interesting. But I am not sure how you are really using it for Iv2. In the first video itself, by the 3/4th 5s segment the face will drift a lot. In that case, how are you creating the second video (scene) from the original image?
@taek75799 Any thoughts?
Hello, this is a workflow solely for joining two videos together. It is only used to create perfect transitions without freezes or other undesirable speed changes.
@taek75799 Got it. My question was in general about drift as well. How do you solve it? I am trying out multiple start images for each scene with QWEN I2I
@rahulr1199 It's difficult to make a full tutorial here as there are several ways to go about it. Let's say the simplest method for perfect consistency without quality loss is, for example, using a woman in a doggystyle position. From there, you just need to change the head position with qwen 2511 or flux2 (i prefer qwen), clothing, or other elements. Generate all your images, then all your videos, and finally use the 'Clip Joiner' WF. You could also use the basic 'First/Last Frame' WF. In my opinion, these are still the methods that yield the best results.
Do you have workflow for T2V using your models?
I2V is the best model i used and the quickest.
Hello and thank you for your feedback. I don't do T2V, and when I do, I use the official ComfyUI one.
was there an update regarding sageattention yesterday or something? didn´t change my system in any way but getting "pathchsageattentionkj dll load failed while importing _fused..." workflow worked fine and fast yesterday.
I'm getting the same crash since the most recent update.
good news, updating sage attention to match the ?new pytorch version 2.9.1+cu130 fixed the problem. bad news I also picked up some weird issues with custom nodes with the most recent comfyui update and the RIFE at the end won't play nicely.
@yamavishnu892 Hello, there are other custom nodes if you want. Maybe that will work; however, they will be slower: https://github.com/Fannovel16/ComfyUI-Frame-Interpolation"
https://github.com/DazzleML/comfyui-triton-and-sageattention-installer wish I´d known this sooner
@taek75799 I don't think it's that node in particular that's the issue, I think comfyui is just doing something stupid with custom node management, as it circularly claims they aren't installed, installs them, says they are installed, then says they aren't installed.
How can I add prompt into the workflow? I saw the text encoder but it goes directly into the set clip node. I don't know where to add my prompts.
FindPerfectResolution seems like a great node. it would be more great if there was a link to download it.
https://github.com/ashtar1984/comfyui-find-perfect-resolution Hello, just type 'findperfectresolution comfyui' on Google; since the 6th of this month, it no longer appears on the manager
@taek75799 thanks bro. you're the best
So cool but...strange problem... It works great for video1 and video 2. Then in video 3 (10 seconds in) it goes wildly off-script. Then it obeys me again and goes back to my script for video 4.
Edit: Prompt fields are inverted for video3 node.
If I bypass video 3 it works fine for a 15second film. It's as if the video 3 prompt is not being processed.
Thanks for the workflow, I'm going to have so much fun with this!
Redownloaded the file and loaded it again. It confirms that the video 3 workflow is broken. But it works great otherwise. Thanks!
I know what went wrong, it's the prompt problem, in video3 subgraph u can see the postive prompt and negative prompt are locate in the opposite place, so u should put the postive prompt in the "negative prompt"
@liudaisong1995986 You're right, the prompt fields are inverted for video 3! Works great now. :)
Thanks for a great workflow!
I'm getting significant quality drift in the second video, becomes blurrier and darker. The third video already becomes really low quality and I find myself stopping consistently after the 2nd video which is still acceptable. Any tips on how to get it to work better?
Hello, what resolution are you using?
@taek75799 First image is 1024x1536, and I use the workflow as-is with find perfect res
@Derpxten Hello, you are using a high resolution, so that is not the problem. Try downloading one of the two preview videos; they are 10 seconds long. You might see the parameters used, which could help.
@taek75799 hey, yea I'm using the same parameters as you do, I just notice that the second video lowers the quality and becomes darker, the third even more, and the fourth kind of becomes really weird and smudgy. To be honest I think it also happens in your videos, but it seems that you usually stop after the 2nd video.
@Derpxten The loss of quality is completely normal. Try different combinations of Lightning LoRAs, there are plenty and I haven't tried them all. The Lightning ones contribute greatly to the quality loss. These two videos are 14 seconds long: https://civitai.com/images/117911712 https://civitai.com/images/117911828. In my experience, it also depends on the starting image.
All my videos are in 1024x1024, then FindPerfectResolution corrects them to the right aspect ratio. However, my images are never actually 1024x1024; they are much higher in resolution. Even if they are high quality to begin with, I always use an upscale: either an i2i upscale with Flux, or Topaz Photo (which is better than Topaz Gigapixel for upscaling). But very often I use SeeDvr2, which is also very important. Sometimes we have high-resolution images, but you only need to zoom in to see that they are actually blurry.
@taek75799 oh, I'm actually using your checkpoint with lightning lora merged. You suggest trying another model with lightning loras and trying to solve the quality loss?
@Derpxten Yes, I created SVI models specifically to address this issue.
Read the description, it might help
Constant error on the FindPerfectResolution node, any known fix?
I had to install some nodes manually. Didn't work from the comfyui manager for some reason.
This worked for me on windows:
cd "C:\Users\myname\Documents\ConfyUI files\custom_nodes"
git clone https://github.com/ashtar1984/comfyui-find-perfect-resolution.git
git clone https://github.com/princepainter/ComfyUI-PainterI2Vadvanced.git
git clone https://github.com/princepainter/ComfyUI-PainterLongVideo.git
@PandaRave you are a lifesaver my friend. thank you
Amazing workflow and model. Thanks a million for them.
I have a request/suggestion for one feature:
Would it be possible to have an optional node to upload a video and use last frame as the init image for generation (if video uploaded then the first subgraph merges the generated video to the end of input video and continues with 10s,15s,20s nodes if enabled)
I kinda did it already and made it to work but its my first week using ComfyUI so its really ugly and disabling the Video Upload and related nodes that are prepended to your workflow is meh (maybe there is a way to group them and bypass with one click instead of bypassing a bunch of nodes to prevent errors).
Maybe it would be possible to add this idea to it?
Hello, thanks for your idea, I will think about it for the next update.
In your opinion which has better consistency and maintains high quality out of this wf and the svi 2.0 pro continous wf?
I used painter i2v long wr for a while, but now I feel like the svi 2.0 is even better. Curious to know other people's thoughts?p.
Hello, it depends on what you want to do. I would say that SVI will be better most of the time.
Hey I'm SUPER keen to give this workflow a run, but I've hit a roadblock:
"loadWorkflowWarning.outdatedVersion
loadWorkflowWarning.coreNodesFromVersion
2643c0a2-4f8c-4b66-9bba-7305a874c2db, 322b39f0-a204-4d81-8926-eeb0309f1efc, 61498a76-d16a-4a4a-8935-0e4dfb830eab, e8912965-6b60-4198-9a54-f4816aac8f07"
I have updated comfyui through the comfy manager to nightly, stable, etc. Updated pip packages.. And no luck. What should I do? Any Ideas? There are some custom node conflicts which may be an issue, but I have no idea how to resolve those without doing a completely fresh-install.
Could be the same problem I and several other people had. I posted this on another question and many people found it helpful:
I had to install some nodes manually. Didn't work from the comfyui manager for some reason.
This worked for me on windows:
cd "C:\Users\myname\Documents\ConfyUI files\custom_nodes"
git clone https://github.com/ashtar1984/comfyui-find-perfect-resolution.git
git clone https://github.com/princepainter/ComfyUI-PainterI2Vadvanced.git
git clone https://github.com/princepainter/ComfyUI-PainterLongVideo.git
@PandaRave I have the same problem with this guy. I downloaded all your recommendations but still it doesn't work. My version is 0.9.1. Is my problem about Comfyui version?
Hi, you should try disabling each node one by one to see which one is causing the issue. Once identified, you can easily replace it with another one.
@taek75799 Sorry but I don't get it. The 4 nodes don't show their names. Because the nodes names are shown as "2643c0a2-4f8c-4b66-9bba-7305a874c2db, 322b39f0-a204-4d81-8926-eeb0309f1efc, 61498a76-d16a-4a4a-8935-0e4dfb830eab, e8912965-6b60-4198-9a54-f4816aac8f07" I can't tell the nodes one by one. The recommended node links are three, so the number doesn't match, too.
@Hdrijjgtij072373 Try to send me a screenshot so I can see what the problem is.
Works great, very easy to run. Only issue is getting the characters face to stay consistent. Any suggestions?
Hi, and thanks for your feedback. This can be quite difficult: each section degrades the video further. The face needs to remain clearly visible at the end of each section. I'll try to release an SVI version this Saturday to improve that.
@taek75799 Awesome! Thanks man! I’ve been messing around with adding IPadapters and it hasn’t really made a difference for me unless I’m screwing something up which is possible. Overall this with the vae clip merge WF is absolutely gold.
I have had OK results applying IPAdapter-WAN to the High and Low models (you can vary the weight) then using ReActor after the video is produced.
@kmbac997 I’ve tried a few different ways without luck but I’ll give your way a shot. I did just train a ZIT character Lora so I’m going to try with that also
@taek75799 What model are you using for the CF option? The GGUF or diffusion? I read you have 2, FM and CF, I got the SVI fm fp8 H/L but wasnt sure if the NSFW V2 Q8 L/H were it or not for the CF. Also, thanks for creating these!
Thx for ur work! in V2.1 the extended parts are wired incorrectly wich results in instant oom because of 10batch generation. check ur painter nodes and the motion frames connection. after hours of confusion, it works like a charm.
Hi, and thanks for your feedback. You're not the only one having issues, even though it works perfectly for others. I think the problem might be coming from the subgraphs."
When I select fixed seed for vid 1, vid 2/3/4 stay with seed 0, am I supposed to manually copy and paste the seed to vid 2/3/4, or leave as seed 0?
Each section is a different video, so each section will have its own seed independent of the others. Therefore, the seed for one section will not affect the other sections in any way
Has anyone noticed that the positive and negative prompts for video3 might be swapped or reversed?
Another question is: why does generating video3 take longer than generating video1 and video2? I followed Usage Tips to use the same seed and generate it segment by segment. Video1 and video2 usually only take 3 minutes, but video3 takes 20 minutes.
btw I use lightingeditionlong v2.1
Hi, thanks! I just changed the negative and positive [prompts]. However, I didn't notice any anomalies with video 3 regarding the generation time. Can you access the subgraph to check if the number of steps is set to 4 on both KSamplers? Also, please verify that the resolution is properly connected.
hello I opened the workflow you mentioned and there is not any load image to do img2vid. is there another workflow you use for img2vid for nsfw??
Still new to ComfyUI - the manager says the workflow contains nodes that are in conflict with other installed nodes. Do I ignore the conflict?
Hi yes
Hi please if someone can help id appreciate it , i downloaded comfyui like a couple days ago so im pretty much braindead , i downloaded the model from this user and the workflow and everything needed
But non of the nodes are connected, even after trying to connect ,nothing works , i keep getting that vhs isnt taking image input even though i uploaded and i even tried asking chatgpt and whatnot, not a single thing works
Is the workflow bugged ? Is there something wrong on my end ? The whole thing is sections with 0 connected nodes fresh out of install, its like throwing someone right into the fire
Hi, most of the nodes are connected via 'Set' and 'Get' nodes. Look behind the main nodes; you'll see them there and understand how it works.
@taek75799 ok ill look into it, but still doesnt explain the error, its not taking images for some reason even if i do have one selected
required input is missing , images vhx_videocombinr : required input is missing images
Man I'm having the same issue. ComfyUI throws an error at the 16fps node (located between the 2 upscale nodes and the 60fps node). Even though I bypassed both the upscale and 60fps nodes, the 16fps node remains active and causes a validation error (Required input missing). What's wrong with it and how to fix?
@Kiempc Hi, I assume you have VideoHelperSuite properly installed. If so, maybe try deleting the node and adding it back again.
Maybe you should also check if it's properly connected. Go to the subgraph, select 'Unpack subgraph', and connect it directly from there.
I had the exact same issue and found the real cause. In this workflow, the image output is placed inside an upscale relay / toggle group (rgthree relay). When upscale is turned OFF, that relay disables the entire group — including the image output.
@serpensartgor751 That's my assesment too. Although i have no idea how to fix it, where should what be linked.
@jazzyreynard285 Hi, sorry about that. Look behind the nodes: they are connected to 'Set' and 'Get' nodes. Move them outside of the groups so they remain always active. Just make sure they aren't muted; they need to be active. To do this, left-click on them and press Ctrl+B. They should not be grayed out.
@serpensartgor751 So what's the best way to fix it man? Should I just pull the image wire out of that group and plug it straight into VideoCombine, or should I open the group and bypass each node manually?
@Kiempc In principle, if there is a bug preventing connections from working because the 'Set' and 'Get' nodes are disabled, you just need to reactivate them using Ctrl+B.
eagerly awaiting SVI version , he has done some decent job just need a way to remove loras since your checkpoint has https://civitai.com/models/1866565/wan22-continuous-generation-svi2-pro-or-gguf-or-32-phase-or-upscaleinterpolate-w-subgraphs-and-bus very clean setup , just need some expert optimisation
HI, I'll be releasing two versions this Saturday, and without the Lightning LoRAs; I'll just be suggesting two Lightning combos that work well.
what am i doing wrong?
i generated the first video, loved it, so i enabled the second part (the 10 second one) and started the generation again, but instead of skipping the first video, it started running on the first video again? i thought it was supposed to start the second video, continuing from where the first was made? or do i need to manually disable the first video before i run it again? so confused.
Hi, if your seed for the first video is the same, ComfyUI won't re-generate your video. Try this: don't enable the other sections, generate your video with a fixed seed, and run it again.
@taek75799 ahhh ok ok, thanks! ill make sure to check that. also, for the second generation (the 10 second subgraph) do i need to prompt the same as the first video or do i prompt like im continuing from the first.
like for example: if im making a video of someone jumping and then turning around and walking away, i would prompt "jumping" in the first video, and then only prompt "turning around and walking away" in the second part? or do prompts neeed to be the same for each?
thanks for the repsonce
In fact, it's quite simple: the second section just takes the last frame from the previous video. It's just a trick, so it doesn't really know what happened before. There's just a node that picks up the last few frames to understand the motion and ensure it looks natural.
So, for each new section, assume that only the last frame is taken into account. It depends: if you ask in the first prompt for a sitting woman to stand up, you can't use the same prompt again because she will already be standing. You can also change the LoRA in each section.
@taek75799 ahh ok thanks so much, i figured it was something along those lines. much appreciated. workflow has been working very well. fast and smooth gens. much love ❤️
@RPxGoon Don't hesitate to get the SVI version tomorrow; it will be much better for making long videos.
@taek75799 alright cool, ill be lookin out for it and will give it a try
on the description you have this:
"Steps: 2+2
(Jellai recommends 2+3 for even better results, and I agree)
Sampler: Euler simple
CFG: 1
....."
but using the associated workflow, i dont see where i can adjust this unless im being mega super blind? Which node does it belong to?
Im hoping this will resolve some consistency i have with eyes slightly morphing a bit and taking away facial recognition through the 5 second video
ah i found this, i'll leave comment up but add:
Right click on the node, edit subgraph widgets -> then the little eye to unhide the relevant things and adjust accordingly.
I didnt see Euler simple in the list, but is that different from Euler ?
For steps - i guess as per suggestion is that 2 + 3 as in 2 on high, 3 or low or the other way around? or can you literally type "2+3" in high and low?
@Biscuit10 Hi, you can also set it to 6 total steps. To do this, go into each subgraph and find the first KSampler (High): set Steps to 6, Start at step to 0, and End at step to 3. On the KSampler (Low), set Steps to 6, Start at step to 3, and End at step to 6
Hi, when loading the workflow into comfyui I get this message:
Some Nodes Are Missing
When loading the graph, the following node types were not found
SetNode
GetNode
FloatConstant
VHS_SelectImages
FL_RIFE
INTConstant
Power Lora Loader (rgthree)
VHS_VideoCombine
FindPerfectResolution
Fast Groups Bypasser (rgthree)
Seed (rgthree)in subgraph 'New Subgraph'
PathchSageAttentionKJin subgraph 'New Subgraph'
ModelPatchTorchSettingsin subgraph 'New Subgraph'
ImageBatchMultiin subgraph 'New Subgraph'
Do you have a fix?
Hi, you need to install the missing nodes. Have you installed ComfyUI Manager? It will make things much easier: https://github.com/Comfy-Org/ComfyUI-Manager
@taek75799 Thank you so much taek75799 it worked, I downloaded all the missing nodes through the ComfyUI manger <3
Hello! For some reason my browser is saying it could be a virus duo to file format and I can't download it. Can you please re-upload it in a normal .json file?
Hi, I can't upload JSON files directly to Civitai, so here is another link. I also got a message from Windows saying that this file is not commonly downloaded; that must be the reason. It's strange, it's the first time I've seen this on Civitai: https://www.swisstransfer.com/d/4ea3c69f-27bf-4184-a31e-f185db37e046
@taek75799 Yeah same here. Thanks again for your contribution and quick response.
Its very painfull to debug a workflow that you see for the first time :D and its not working out of the box. You need to enable fps16 from disabled upscale group to make 5 seconds. IF i turn on 10seconds thant there is another eror about missing image models that i have no clue where they come from :D heh.. you should update this...
I don't know why it's not working; maybe it's because there are subgraphs? I'm not sure. Try to enable all the nodes (meaning, un-bypass them) and see if it runs with a resolution of 300 so that it's fast.
@taek75799 yeah i found the stuff to enable but its just a bit messy :D im only 1 month in to comfyui but posible to understand. also this long video making makes no sense for me for example, because every extra 5 step kinda makes quality wors, some weird contrast or brightness gets aplied every 5s... But we learn every day something new here :)
@poisas69220 I completely understand your point of view. I limit myself to 15 seconds with this method. Other people asked me for 20 seconds, so I added that option. I don't make the workflows for myself; I'm just trying to make sure as many people as possible are happy. Try the SVI version, the degradation is much less pronounced: I estimate it at 30 seconds or even more, depending on the resolution.
Hello, whenever I use these models, it always generates me same video regardless the seed. What I am doing wrong? Thank you for any advice