My ComfyUI workflows for using Wan 2.2
This workflows are used by me to create my art.
They are optimized for my checkpoints and created of my latest knowledge to enhance the outcome.
"If this workflow leveled up your day, I'd purr-eciate a like! 😻"
Versions & Information👇👇👇👇👇👇👇👇
👉 Please read below and the file descriptions "About this version" for more info's.
🎬🐈Click me! HowTo + Video
💡Some WAN 2.2 versions use high+low checkpoints, other like S2V use only a single checkpoint - make sure to read the descriptions and use the correct checkpoints.
⚠️ Do not use the workflows with the "Nodes 2.0 beta" from ComfyUi or it will mess up things.
👇👇👇👇👇👇👇👇
What you get from the comfy workflows:
♨️ Easy controls
✅ As less as possible dependencies
🪧 Detailed documentation
⛓️ Highly automatic logic
✨ Optimized results
🎬 Fully automated resolution logic
🔖 Bookmark-Shortcuts with number keys
Types of workflows
FastFidelity C-AiO
🖼️ I2V and FLF2V
🧩 Automatic aspect‑ratio calculation and fitting
✨ Multiple Upscalers
Torchlanc (very fast, color correct, low VRAM)
Upscale with Model (additional detail, high quality)
RTX Super Resolution (ultra fast, very accurate)
🤝 Video resolution matching - Fully automatic scaling and resolution calculations
🕠 Length automation - Fully automatic calculation of frame count
🫥 Watermark option
🧮 Color match feature
👾 MiniMeme feature - Create small gif's
🎭 NAG - Negative prompting with CFG1
🪄 Interpolation feature
🌀 Perfect loop feature
🃏 Last Frame Extraction
🔖 Bookmark-Shortcuts - with number keys
FastFidelity C-SVI
🖼️ SVI (SVI 2.0 up to 10 samplers)
🧩 Automatic aspect‑ratio calculation and fitting
✨ Ultra‑fast, color‑correct upscaling (torchlanc) OR Upscale with Model
🤝 Video resolution matching - Fully automatic scaling and resolution calculations
🕠 Length automation - Fully automatic calculation of frame count
🫥 Watermark support
🪄 Interpolation feature
🔖 Bookmark-Shortcuts - with number keys
Swarm Basic
Absolute basic SwarmUI preset
Not recommended over ComfyUI for video generation
Backend Test
If this does not create the example video your ComfyUI backend is broken.
If this works, but your other workflow not, the other workflow is broken or missing dependencies
🩻 Known issues and advice's
⚠️ Some workflows may set on webp av1 encoding (VHS node) - If your computer/setup missing drivers use any other like H265 or H264!
Install ffmpeg!
Update Comfyui and custom_nodes!
Update pytorch 2.9+cu128 or higher
Make sure to read where files/models should be placed inside the workflow
Check if the filepath for model/clip/vae match your system like Linux/Windows
The plugin ComfyUI-DD-Translation can break node connection (avoid)
All older Versions are available inside my GitHub Repo.
Spacial thanks to @Abyss_Games for a really good idea and help with the "loop"!
I got some really good ideas from @Gladas workflows!
YOU are responsible for outputs as always! If you make ToS violating content and I get aware I WILL report this.
Description
Requirements
Comfyui-WhiteRabbit
+ packaging, torchlancSage Attention (optional)
Upscaler model (optional)
Features
🗣️ S2V - Audio file, Text2Audio, Voice clone
🔊 Audio trim - Automatic audio trim to frame length and video to audio
🖼️ Anti 1st frame burn - Latent cut
🧩 Automatic aspect‑ratio calculation and fitting
✨ Ultra‑fast, color‑correct upscaling (torchlanc)
🤝 Video resolution matching - Fully automatic scaling and resolution calculations
🕠 Length automation - Fully automatic calculation of frame count
🫥 Watermark support
📢 Soundmark option
♾️ True infinite batch as long as you want
🧮 Color match feature
🎭 NAG - Negative prompting with CFG1)
🪄 Interpolation feature
🔖 Bookmark-Shortcuts - with number keys
Changes / Fixes
Added CozyVoice
FAQ
Comments (168)
Ooh, an update! I've been messing with this flow a bit and it's great. Question: What do you mean by "true infinite"? One thing I was trying to force into the flow (unsuccessfully bc I don't know how a lot of this actually works) is context windows. I really like using them to create ~10sec clips instead of 5-6 without having to to splice with last frames. Is that something possible with this checkpoint and workflow? Thank you again!
True infinite is the looping logic for S2V I implemented, you can add as many batches as desired. The only thing what is stopping you is your RAM/VRAM.
This enables longer videos as just setting high frame counts.
What exactly do you mean by context windows?
Thanks a lot for your flows. I'm trying to optimize the I2V to make more use of my 5070ti, with sage attention and block off loading, but it means changing the whole design with WanVideo sampler. Does it work or not with a distilled GGUF checkpoint like yours? Am I even making any sense ?
Since I do not use WanVideoSampler, I can not provide any information on that.
Block offloading will things make slower/worse, since you are offloading to CPU/RAM from VRAM. - So only use it if you have to.
I can not tell you how to change the whole design to that.
Wow, I tried the new AiO v2 that you redesigned and it's like 50% faster than the I2V or the previous AiO !!! I don't know, maybe the new sampler. Anyway, fantastic !
@Elmer588 I would not assume such a speed up just from optimising the workflow. But good if it works that good for you :)
I’m not sure what went wrong... I tried the new workflow, and since then, I haven’t been able to generate a video anymore.
The PC freezes so badly that I have to do a hard reset. I used to be able to handle a 5-second 720p clip on my 4080 Super, but now I can’t even process a 1-second 360p clip without it freezing and forcing a reset. I’ve tried multiple old workflows that worked before, as well as fresh installs and older versions of ComfyUI, but the problem remains.
I can not see how a workflow could somehow destroy your ability to generate videos. For sure is something wrong with your setup or PC.
You should try to make a fresh comfyui install, imho.
@darksidewalker Yepp, I’ve tried a fresh install and even older versions of ComfyUI. But I gave it a second chance before I gave up and tried a fresh install once again after like the 5th time. Suddenly it works!! I'm so confused...
hello Master, You finally joined GGFU for good, but in my system they are too much unstable with unofficial knots and basic patterns, I mean close compatibility with PyTorch, builds that break more often than other nodes, have you left definitively the old guard?
Can you explain what you mean? I do not understand.
in your latest C-AiO workflows, you always use GGFU nodes, but to install these nodes I worry to breaking my system, am I wrong to fear this problem?
@dreamswalker you can not break your system by installing this custom node
Hello Dasiwa, Thanks for your new workflows but they fail. With a fresh install comfyui 0.5. => AIO 1.9 is ok, S2V 1.2 is ok
but C-EX 1.2 not ok, AIO 2.0 Not OK and C-S2V 1.3 not ok. For these 3 workflows, It seems video isn't decoded. Should you have any idea please ?
I generated some videos today with these WF's, one just seconds before. On my end they are all functional.
@darksidewalker Hello Dasiwa, thanks for your responses. I have found the bug's origine : This is the Ksampler node. I had to destroy the 2 ones, insert 2 new nodes of ksampler and join this last one. Now that's ok for AIO 2.0 and C-EX 1.2. Cool !!!
For s2v 1.3 even though the ksampler seems to be good, i had to do the same and it is ok.
wow v2 is extremely clean
Will a way to do Last Frame only be added? Now that v2 is out, there's not an easy way to disable input for the first frame like there was previously.
There's plenty of times where I'd like to make a video that ends on a specific image, and don't want to specify how it starts. It works great on the workflow versions before 2.0 as I can just disable the first frame input node.
Please consider adding this functionality.
Well, I did not think to use it this way. But I'll consider it :)
I did try it with the old workflow. FLF with disabled initial frame gives error, obviously because the input is missing.
So I do not know what you did exactly. I could set a empty image as first frame if none is provided, but you would start from a black image.
No clue what you did :)
@darksidewalker all I do is disable the "Initial Image" node and then the "Batch Resize" node that errors. With those two disabled it works great and ends on the "FLF2V" image.
@darksidewalker he wanted to test the technology of humanity.
@JakeJames use black or white image at init image :)
Hello DaSiWa, i was using your workflow but i have a bit of a problem. Videos generate fast, but.. somehow everything has a very limited range of motions. Its at best wiggle, little motions, nothing drawn out no long movements. What is the problem?
Hi! ✌️
I would guess your prompt is the issue. Too less or over extended prompting will not work good.
You sir, are a mad genius of truly epic considerations. I am running 6G Vram - Using your Q5 Version. Tip of the hat Nods works a treat, thank you. I will make sure I donate in the near future
The last image is not respected, the video generation does its own thing and never ends where the last image indicates; in fact, it's as if it wasn't there. Any tip?
Yes, enable flf2v feature 😁 I assume you forgot that.
I just kind of skimmed over notes/explanations and was about to ask same question. These notes are actually so helpful. Greatly appreciate all the work.
I was given a "Model in folder 'vae' with filename 'Wan/wan_2.1_vae.safetensors' not found." even though I have it in the VAE folder. I'm pretty new to this so I'm not sure what I'm doing wrong
nvm figured it out, I was given "Wan2_2-I2V-High-LS-DaSiWa-TastySin-fp8-v81" right after. I'm not sure which one I have to download and put in my checkpoint folder. Please help as soon as you can
You need to select VAE, clip, High-model and Low-model in the nodes.
this updated one dosent work i allready did your requirements in the about this version
For sure it does work with my workflow. Check again if you really use comfyui 0.4.0 or higher.
@darksidewalker im on comfyui 0.6.0 is that my issue? cause the output is just the distorted flickering thing it does when it dosent work
Make sure you use the correct checkpoint for high and low
i have some questions, i saw a few versions of wan 2.2 model on hugging face, and i don't know what's the difference, i think maybe u made these models?
1. tastysin
2. midnightflirt
3. lurenoir
4. radiantcrush
some of 'em have different version, like Q5_K_M, Q5_K_S, Q6_K, which should i choose? what's different? and what does the letter mean?
Q (Quantization) higher number means better quality but more VRAM usage. download what fits in your VRAM (I have 16gb, so I use Q8 model)
there is no way to give more references other than 1st and last frame? I'm trying to snitch several videos to make scenes, but I'm logically losing memory of the stuff that isn't in the 1st frame.
There is no way.
Sorry to bother you. I'm using the C-AIO 2.0 workflow and it failed to load with the error "TypeError: Cannot set properties of undefined (setting 'mode')". My ComfyUI version is 0.6.0, last updated on December 24. I rolled back the version to 0.5.1, the version updated on December 18, and it's still the same error. Did I do something wrong?
others is done, only aio 2.0 = =!
I do not know that error. But last comfy/custom_node updates messed something up. I uploaded a new version, maybe it is fixed now.
@darksidewalker Thank you so much—it worked! and I really like it
i'm using your C-CA v1.5 workflow and trying to figure out if theres a way to add a particular voice style to the mmaudio, i notice in comfy there is a mma vocoder loader too. i want a specific characters voice to be used in the audio generated by mma, is there some way to do that?
Not that I'm aware off.
@darksidewalker i see, thank you for your response
Thanks for your work!
I have a problem with the newest version workflow, Backend and Features, Node #945 is always unhappy, unless I activate FLF2V and Exact Loop.. Any Ideas? Thanks!
console:
Failed to validate prompt for output 644:
* easy ifElse 894:945:
- Required input is missing: on_false
Output will be ignored
Failed to validate prompt for output 63:
Output will be ignored
Did you install all dependencies?
Thanks for your work!.
After comfyui update, and all dependencies installed, with FLF2V and Exact Loop i get: "TypeError: WanModel.forward() got an unexpected keyword argument 'control'. Did you mean 'controlnet'?". With all feature disables, i get the "easy ifElse" error.
@Cuajoneta I do not know this error, I can only assume that a dependency is missing, everything works on my end.
Double check if everything is installed.
@darksidewalker Testing with the GUFF models, on a clean installation of ComfyUI 0.4.0 and 0.6.0, with all dependencies installed: the error is that the Ksampler is entering a parameter with the value 'control', which the Ksampler doesn't recognize. However, if I use your .Safetensor v7 models, the error disappears. It was working with the GUFF models until ComfyUI and the node library were updated.
@darksidewalker Did a fresh install, pulled all missing nodes with manager, same issue. in some cases, when the easy ifelse node gets the on_false input from a bypassed node, this error occurs.
The comment from ErufuDaisuki seems to mention the same.
When i reconnect to bypass the ifelse node, everything works fine.
I would assume that your install is somehow corrupted. I double checked everything and if all dependencies are installed it works without issues.
@darksidewalker allright, I'll figure it out eventually. Thanks anyway!
I changed the switch logic with C-AiO 2.2, if this is not working with the standard comfy node, your install is broken.
@darksidewalker ty for answer!! For some reason, ComfyUI Manager detects "ComfyUI-GGUF-FantasyTalking" as a missing node when I load the workflow for the first time; after deleting it, your gguf models started working again. It's strange that the Manager asks to install it on the first workflow load when it's not required. After deleting it, it no longer asks for it.
@Cuajoneta well that's strange, there is no node from fantasy talking inside and I do not have this installed ether.
On a freshly installed ComfyUI 0.6, your C-AiO v2.1 doesn't seem to work due to the following error:
-> Backend and Features is highlighted red
-> The If else node before F4# Interpolation is highlighted red, the error message is Required input is missing: on_false (there's a connection leading to the connector from F3)
This error happens no matter which options are enabled.
Did you install all dependencies?
@darksidewalker I have used the "Install missing nodes" button of Comfy Manager:
First batch (since there were no conflicts listed)
- rgthree-comfy
- ComfyUI-Easy-Use
- ComfyUI-VideoHelperSuite
- ComfyUI-Frame-Interpolation
- WhiteRabbit
Afterwards I installed ComfyUI-GGUF and then ComfyUI-KJNodes.
The manager had listed a third package ComfyUI-GGUF-FantasyTalking which I didn't install since all missing nodes errors went away after installing the other two packages.
Have the same issue. It works while FLF2V is enabled.
The boolean node (True/False) it's not working with all these ComfyUI and nodes updates. I tested it on ComfyUI 0.4.0 and 0.6.0, fresh installation with all dependencies installed.
@Cuajoneta It is working, even the if-else nodes, if you install all custom nodes correctly. I mentioned on a other post, that I downloaded the workflow now 3 times and tested every setting on its own, all is working.
My comfyui setup:
version 0.6.0, python 3.13.8, torch 2.9.1+cu128, [email protected]
Dependencies:
* https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite
* https://github.com/rgthree/rgthree-comfy
* https://github.com/Artificial-Sweetener/comfyui-WhiteRabbit
packaging, torchlanc
* https://github.com/kijai/ComfyUI-KJNodes
@darksidewalker Ah, I'm using python 3.10 and I think my torch version is different as well. When I find the time later or tomorrow I'll try reinstalling with your dependency list and let you know if that worked.
I changed the switch logic with C-AiO 2.2, if this is not working with the standard comfy node, your install is broken.
Uploaded C-AiO v2.5 - This should fix everything.
hey, I have all plugins installed and once trying to run it, i'm getting a following error:
Prompt outputs failed validation: easy ifElse: - Required input is missing: on_fale
You need to set always F2# and F3# inside "Backend and Features" subgraph, at least that's how I tried to make it run
I really do not know what you are doing, but you obviously did not install everything correctly. I downloaded the workfow 3 (!) times now and tried every setting. All is working.
My comfyui setup:
version 0.6.0, python 3.13.8, torch 2.9.1+cu128, [email protected]
@SKlunge that is not needed and would not even function for i2v, it is only for flf22 with loop.
Here are all needed dependencies for AiO v2.1:
* https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite
* https://github.com/rgthree/rgthree-comfy
* https://github.com/Artificial-Sweetener/comfyui-WhiteRabbit
packaging, torchlanc
* https://github.com/kijai/ComfyUI-KJNodes
@darksidewalker
I posted wrong. I had the same error as @veladorn802 . Frame nodes F2-FL2V and F3 were disabled in backend and features. Watermark option - very useful !!. F1NAG is for Nodes 2.0 beta? should it be on or off?
@Kwiwi Nodes 2.0 beta will kill the workflow any most of custom nodes available, just do not use it.
@darksidewalker
I had the same issue. I noticed that in version 2.0, the node was "Any Switch (rgthree)". I switched back to it, and now it works without errors.
I changed the switch logic with C-AiO 2.2, if this is not working with the standard comfy node, your install is broken.
@darksidewalker
I downloaded the latest version you posted. I reinstalled comfyUI fresh install separately because somehow I managed to break something by installing extensions, it wouldn't start anymore.
After a fresh install i have this this:
Installed frontend version 1.34.9 is lower than the recommended version 1.35.9.
Version Compatibility Warning: Frontend version 1.34.9 is outdated. Backend requires 1.35.9 or higher. Visit https://docs.comfy.org/installation/update_comfyui#common-update-issues for update instructions.
update_comfyui >>>>Successfully installed comfyui-frontend-package-1.35.9
update_comfyui_stable >>>Successfully installed comfyui-frontend-package-1.34.9
Using 1.34.9:
I can make it work but it's a bit strange.. I have F1 F4 F5 F8 enabled but I have to disable it with CTRL+B in the front end FLF2V(Last-Frame) and in the back end the batch images node(switch logic).
It's probably from the white rabbit node (2025-11-17 latest update)@Kwiwi
If the switch in the front gui panel does not work https://github.com/rgthree/rgthree-comfy is not correctly installed.
The frontend version of comfy itself does not matter that much, it is just the gui version, the internals should function non the less. Also this has nothing to do with the workkflow, it is the comfy install itself.
The "white rabbit node" needs 2 dependencies (torchlanc and packaging), maybe you missed that?
Uploaded C-AiO v2.5 - This should fix everything.
@veladorn802 Try this https://imgur.com/a/Cwsy66i
Facing the same issue as others. I have installed all the required dependencies again. But it still shows the on_false error.
I have disabled nodes 2.0 as well.
It only seems to work when FLF2V is enabled but the output is bad.
I changed the switch logic with C-AiO 2.2, if this is not working with the standard comfy node, your install is broken.
Uploaded C-AiO v2.5 - This should fix everything.
@abggsop586 Try this https://imgur.com/a/Cwsy66i
very good workflow and wan2.2 models. The explanations of use are very good and useful. It took me a while to realize that there are on/off buttons there o_o
Really fast and high-quality video generation, just what I need for my RTX 4060 Ti. But there is one problem...The animation speed is very fast, how can I make it slower? I am using FastFidelity C-AiO 2.2
Can you describe what animation speed you mean?
@darksidewalker Well, as if the animation of breathing, chest, hip movements, in short, any animation of movements, in the video output, as if accelerated 1.5 - 2 times
@_Jarvis_ That sounds off to me, maybe you set 24fps on a short frame time?
Or maybe you used extra "speed-up" loras?
@darksidewalker No, I didn't use speed-up loras. The fps is set to 16. Maybe it's because of the interpolation? I set the fps to 8, enabled interpolation, and it seems to be working better.
@_Jarvis_ I cannot imagine what your problem is, 8 fps is really low and interpolation will set it to 16 again, so that is almost the same as 16 native. I do not understand what that should bring on the table.
If you want slow motion you should prompt for it, I think this is just a prompting issue.
First of all, thank you for your hard work. I made changes to your version 2.3, as some nodes in the ComfyUI-Easy-Use package aren't working correctly, even on clean installations of ComfyUI portable and updated with testing workflow. https://imgur.com/a/mjidlZ9
Hi, thanks!
1# This are correct, I replaced them, too
2# and 3# rgthree-switches there do not work as expected for me they will trigger a error or the color-match can not prevented or will be made every time.
The boolean triggers for me just fine.
Uploaded C-AiO v2.5 - This should fix everything.
@darksidewalker fun or not but still not working for me. I removed everything related to if else and boolean. I tried something similar as @Cuajoneta .I added an F9 that is activated or deactivated together with F8
I think it's a problem with comfy manager. I tried to install SD prompt reader a few times but it didn't work. I have to delete the folder from comfiUi portable and go to the link:https://github.com/receyuki/comfyui-prompt-reader-node to install it manually.I think it also affects the installation of other nodes.
@Kwiwi Yes, it's happened to me before that sometimes when installing nodes from the manager, they mess up your Torch version. I'll have to start doing everything manually again, what a pain.
I'm really fighting windmills here, every day comfy breaks other custom nodes....
I haven't updated your workflows in a while. I was on 1.9, now I've switched to 2.5 and I'm getting this error:
got prompt
Failed to validate prompt for output 644:
* easy ifElse 911:1003:
Required input is missing: on_false
Output will be ignored
Failed to validate prompt for output 907:903:
Output will be ignored
Failed to validate prompt for output 907:906:
Output will be ignored
Failed to validate prompt for output 958:
Output will be ignored
Try this https://imgur.com/a/Cwsy66i
@Cuajoneta Thanks, I'll give it a try later, If the author doesn't come up with a fix.
The error does not quite help me. It would mean you did not set anything. Can you look inside the checkpoint subgraph if there are disabled switches? Once comfyui did disable them out of the blue for me.
Please try C-AiO 2.6 and tell me if this works for you
@darksidewalker 2.6 works as it should. Thank you.
Good afternoon, I haven't tried your version 2.5 yet, I was playing around with v2.4 XD
I think I found the problem with the logical nodes; it's not just an issue with the Easy-Use If-Else node, it's a general ComfyUI problem or a new feature. Some signal must be sent to the True/False node, even if it's a null signal. I'll solve it with this little trick https://imgur.com/a/Cwsy66i . I have to leave the Any Switches nodes enabled, or the signal is lost, just like you did in v2.5 in the checkpoint back-end. The If-Else node in Color Correction was working fine, because it receives signals through both entrances, so I left it as it was in v2.4. In conclusion, with v2.4 with that fix i posted, and Any Switch enabled, everything seems to be working without errors now for me. TYVM4 all
I also updated everything with update_comfyui_and_python_dependencies.bat and saw that some dependencies were updated; be aware that sometimes it can break more than it fixes.
https://imgur.com/2pB0dBN @Cuajoneta @darksidewalker maybe like this for V2.5 ?The only thing is that F3 also activates F2 but F2 does not activate F3
@Kwiwi F3 is optional, that's why F2 doesn't force F3 to be activated. That's fine in my opinion.
Right flf2v do not have to be a loop at all times
Please try C-AiO 2.6 and tell me if this works for you
@darksidewalker TYVM! Everything works fine for me.
My setup, in case it helps someone:
ComfyUI 0.6.0 portable nvidia
ComfyUI frontend version: 1.35.9
Python version: 3.13.9 with libs+Include
pytorch version: 2.9.1+cu130
sageattention-2.2.0+cu130torch2.9.0 (work on checkpoint/model)
xformers version: 0.0.33.post2 (work on VAE)
NVIDIA Driver: 591.59
Cuda 13.0
@Cuajoneta Try the NVIDIA Studio Driver 591.44 .
@Kwiwi Hello, why? Is there something wrong/good with 591.59/591.44?
The most well made workflow i've seen in comfyUI until now
Hi, Tried to create smth with your workflow; I read replies and didn't find my problem: ModelSamplingSD3
'NoneType' object has no attribute 'clone'.
It's poping out on WanFirstLastFrameToVideo phase. I dont know how to fix it. Maybe smbd had same issue?
Please look inside the checkpoint subgraph if there are the switch nodes disabled, comfyui seem to disable them out of the blue. In 2.5 this should be fixed
Thanks a lot! However I have the following error for the 2.5 workflow when not selecting FLF2V option: "Failed to validate prompt for output 644:
* easy ifElse 911:1003:
- Required input is missing: on_false
Output will be ignored
Failed to validate prompt for output 958:
Output will be ignored" So basically the workflow only works if you also provide a last frame you want. But in pure I2V it crashes.
I checked another comment and there is a "Any Switch" node missing before the "true/false" node
Well, technically it is not missing... But it seems some comfyui update messed up the function of the custom nodes...
Please try C-AiO 2.6 and tell me if this works for you
@darksidewalker Yes it works OK thank you
Thanks for much for this, it is incredible! I just have one question on the workflow functionality, and I'm probably being stupid... On the "DaSiWa WAN 2.2 i2v FastFidelity C-AiO-25" Workflow, how do I do regular I2V?
If I toggle FLF2V off, I get an "IfElse" node error (red box around it) in the backend and it just won't run. I have an image in my start and end frame boxes from an earlier FLF2V test, but still nothing. (Also tried bypassing the end frame with FLF2V toggled on and a diff red box highlights instead in the backend).
Also, just one final question... Would you recommend the CLIP that is in the workflow by default over the "nsfw_wan_utm5-xxl_fp8_scaled" CLIP? Or does this not matter? (No idea what the difference is, I'm new to all this).
Thanks!
EDIT: Just seen someone reporting something similar (I think), how do I fix this please to get it working again if ComfyUI has broken a node? Thank you
I did try a lot the last days, but for some nothing works. I have no idea. I can not even reproduce the error.
I did try the nsfw clip and found no use of it. You can test it, set the same seed and settings and just use any clip. If one outcome is better for you use this, I could not produce other results with this clip.
Please try C-AiO 2.6 and tell me if this works for you
@darksidewalker Thanks very much for the fast replies, appreciate it! I've tested the 2.6 version and I am now able to run normal i2v as well as flf2v :)
Thanks again, and best wishes for the new year!
Hi Dasiwa.
I'd like to ask if you have plan to replace Text Encode node and Sampler into "Wanvideo text encode + Wanvideo sampler" which have t5 input, to improve the ai's prompt understanding?
This question is a bit unclear to me. All WAN 2.2 text encoders are t5 encoders, WAN 2.2 does not understand other text encoders.
So there is no benefit from using another loader.
If you mean you would like to use other clip files, you are free to use bf16, nsfw, whatever you like.
In my experience there is barely any difference between them anyway.
If you ask me to change my workflow to the "kijai WanVideoWrapper" nodes - No, not planned, you could do that yourself if you want to bother with that extra nodes. ✌️
@darksidewalker Thank you for quick response. Kijai one is what I asked for, as the prompt seemed to be reflected better (maybe just a feeling)
Hi Dasiwa, Version 2.6 is perfect ! thanks
V2.6 works perfectly!! that error didn't appear anymore. Question about Lora for I2V in your workflow. I saw that it has Lora High and Lora Low, the same lora can be used for high and low? does it make sense to use it like that?
That depends on the lora and how the creator designed the lora. You will have to look that up per lora.
Hello,
Now I want to try WAN 2.2 and am wondering whether I should upgrade to 5070 12G or 5060 TI 16G?
5070 TI is out of my budget
Just wanted to say thank you for the Workflow, learning the ropes and was struggling to find one that didn't output an absolute mess, yours works beautifully and once I dropped the resolution for testing purposes the renders are lightning fast.
It really helps with understanding the limits & constraints when it's so easy to iterate quickly.
Hey ! Just wanted to say thank you for all the wonderful work you share ! Just testing the 2.6 workflow and it's really nice.
Hi Dasiwa and everyone using this workflow,
I think it’s a powerful workflow—I love it!
However, even when I put the image in last-frame-image, the output is completely different. What’s happening? Did I set something wrong
Hi and thx!
Maybe you forgot to activate the FLF2V feature?
@darksidewalker Oh no, you are right. Thanks
For S2V, even if I change the settings in the settings section, it only does 480x480. It doesn't apply the settings I changed. Why?
I do not know why, if I set a resolution it will use that resolution.
What is the possible cause of run times increasing after the first generation? I just finished setting this up and my first run only took about 6 minutes, but every one afterwards took at minimum 15-20 and I don't know why.
Your setup of comfyui seems to have a memory leak. Try updating everything, make sure not to use torch 2.8 and maybe lower your sample resolution.
I see you recommend h.265, did you try AV1?
I would recommend webm av1 CRF 30, atm.
H265 over H264
Hi Dasiwa how can i in FastFidelity C-AiO
change the cropping of the image which i put as a starting image? I noticed that the image in the generated videos are being cropped.
There is no logic for cropping implemented. So whatever you are facing there is not from the workflow.
Maybe you use a resolution not fitting your initial image?
@darksidewalker i was talking about this node 🐇 Batch Resize w/ Lanczos
@darksidewalker You mean that the resolution I put has to be compatible, and if I put 720x960, it's not compatible?
@john455454674 This node just scales the images to optimal WAN 2.2 input resolution with div16. No cropping involved.
If you use a 480x480 image and want a video with like 1280x720 WAN itself will crop it, because the aspect does not fit.
@darksidewalker ok thanks i will find a solution. The picture and the resolution has never been for me from wan2.2 cropped, even when it's small. I also used your last workflow, and it worked fine. The problem is this workflow, but thanks.
@john455454674 The logic did not change since like v1.5, so I don't think it could be the workflow.
do you plan on adding that new SVI Pro with wan 22 its sopposed to extend videos and not look bad?
Maybe if they enter a stable state. But for now they have problems with 20+sec and distillation. There is a discussion about that on GitHub. Also it will not work good with fp8, what makes it kind of pointless.
无缝循环会被错，请问为什么
'str' object has no attribute 'shape'
Update comfy and make sure all dependencies are installed correctly.
@darksidewalker 谢谢大佬回复，用的是rh在线平台，所以现在问题无解了~
I'm starting to leave the image generation realm to move into video gen.
Tried pure WAN, Smooth, and some other variants.... Sir (or lady, not sure) I have to say, I'm trully impressed by your WAN model. It produces very clean results, and particularly, I used your I2V workflow... I learned a lot with it!!! Thank you for your efforts.
I have a question though. I modified the I2V myself to make it receive a video, take the last frame, use it as start for the next and so on.... And Obviously I'm hitting the color wash/quality degradation issue after 4th ot 5th consecutive one. I read an upscale helps, but is not a definitive fix.
Did you come up with something for that?
Thanks again an congrats on this beauty!
There is no method to prevent this degration and a upscale will alter the starting frame very noticeable.
You could try SVI 2, but it may not be stable atm.
@darksidewalker Thanks for the advice. Will take a look. I know there is a lora for SVI. Is there anything else that need to change? Do you know any good sources for investigation? If not, that is ok. I'll go and search myself. Just a lazy dude tryinbg to cut corners! :) Thanks again!
@maseielite423 You will need a workflow made for SVI chaining.
thats a great WAN model and a good workflow, especially the ability to control the quality/speed ratio, allowing for quick tests without having to wait for 10 minutes
Great model and workflow. I have one issue tho, I'm trying to do img2vid, and I only have one image, but the workflow needs one starting frame and ending frame. What do I do about this? Also I tried giving the same pic in both. While I do get results, it seems like the character in the picture doesn't move much, it does but not as much as I want it to, like doesnt change positions or anything. How do I make the character move more?
Just "I2V" do not need a start AND end frame.
"FLF2V" need both.
If you set the same image to start and end it will have less motion, because it will try to maintain the same thing. This is only good for loops/exact loop.
Read the description inside the workflow and check what features you enabled.
Thank you sincerely for sharing this. It is an excellent workflow, and it actually runs faster than the official one. However, I still have the following two questions:
Q1: The first and last frames are displayed correctly, but the last frame is shown for an extremely short duration. Is this due to an issue with the CLIP stage, or is there some option/feature that I have not enabled?
Q2: This workflow currently supports only a start frame and an end frame. In the future, could it be extended to support multiple keyframes (e.g., img1 → img2 → img1, or img1 → img2 → img3 → ...)?
Hey and thank you!
Let me explain...
A1: The end-frame is 1 frame for WAN 2.2 so it will try to just end with this frame, nothing more.
A2: WAN 2.2 (i2V) does not support multiple reference images, so this is not possible.
@darksidewalker Thank you for your reply. Yesterday, after I tried using an LLM to rework the clip’s prompt/editing, I found that the ending frame could be presented more naturally and less abruptly. It seems my writing skills aren’t very good.
Thank you for this. Very detailed explanation. Quick question about enabling FLF2V. Is there a way to loosen the strictness of last frame? I essentially want the character pose to be strict but the background not to be so strict for some transitioning effects.
There is no simple way to do that.
Better edit the last frame the way you want it to be before inserting.
im having a problem the checkpoint switch , features switch and resolution have red boxes around them and aren't connected to anything and i cant load loras also node 105 is having problems
Make sure to install the dependencies and custom nodes. They are all mentioned in "About this version" and inside the workflow.