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:
https://github.com/Artificial-Sweetener/comfyui-WhiteRabbit
packaging, torchlanc
Changes:
🌀 Added refined "Exact Loop" feature for seamless loops
now seamless
no missing frames
🧩 Automatic aspect and ratio fitting
🖼️ monitor resolution selective upscaling
✨ ultra fast, color correct upscaling (torchlanc)
🪡 Added "Combine Video" feature (2 videos)
Added more descriptions
Streamlined nodes, compacted design
Minor adjustments and spelling
Fix:
Aspect morphing if FLF2V was used
🤌As easy as before.
FAQ
Comments (34)
Fantastic work! Amazing results with Lurenoir! Although I have an issue on 1.4, updating from 1.1. I used to be able to upscale up to 4x and get very high resolution videos. Now I have to be very modest with the upscaling because I get an OOM. Any idea what might cause that with the upscaler?
I did not try that. The upscaler is new, much more accurate and efficient. Also the method is another. How high did you set the desired resolution?
@darksidewalker I tried 3200x3200 other variant in 9:16 and it failed. 3072x3072 did work. But I could easily pull 3600x4356 on 1.1 and other upscalers. I have an RTX3090 24GB of VRAM.
@Cosmicv You tried to pull more as 4K^^ huge. I'll check and report back.
@Cosmicv I tried video 528x768 (2:3) with desired Monitor resolution 3840x2160 (4K) what resembles to 1472x2144 resulting video because of aspect and 32 divisible. Stunning clear and no OOM. That's also almost 8 times the initial resolution.
On 3200 I also git OOM. But the results with the new upscale are superior to just the old, even with lower resolutions.
You can try to reduce the precision on the upscaler from fp32 to fp16 and go higher, but will lose details.
@darksidewalker Alright thanks. I will try to do more tests with more reasonable resolutions. I guess a starting resolution from 1024x 1024 (square image) to 3200x3200 with fp32 might be overkill lol.
@darksidewalker Alright so I did a lot of testing and I guess I'm just having a hard time understanding the limits of the new upscaler. I was able to recreate your setup with the same resolutions for 5s and I had no issues. 10s gave me an OOM tho. 1024x1024 upscaled to 2048 to 2048 for 5s also gave me an OO even with fp16, like most of square ratio videos for some reason. Then I tried 1152x 832 upscaled to 3112x2440 for 5s, 7s, no issues. 10s, no go. I tried another image with the same resolutions at 7s, OOM. It behaves randomly on my side. I did then test the 3112x2440 for 10s on the 1.1 workflow and I had zero issues. This is not me complaining at all, the work is fantastic and this workflow is simply amazing! I am just reporting what I have been testing. Thank you so much for this great workflow! 😀
@Cosmicv Thank you for that test.
I can say I did implement the new upscaler, because the results are much better. The old was just lanc, cheap and fast, but no color correction and AR. The new one is faster and uses the GPU over torch, maybe that's more demanding for super high resolutions.
I'm a bit busy atm, but I'll look into it and have some tests later. Thank you for your elaboration!
Did you try with WF v1.5 or 1.4?
@darksidewalker I haven't tested the 1.5 yet but I will later. All my tests were with 1.4 and 1.1 for comparison. The upscaler does look a bit better and is so much faster indeed!
@darksidewalker I played with 1.5 a lot yesterday and it works great! No more OOM with my res output for 7s. I only had a few CPU error related saying it tried to transfer too much bytes during the Frame Interpolation after a few generations but other than that, top notch! Thank you so much :D
@Cosmicv Did some testing too, reducing that "max_batch_size" inside the upscaler also helps for higher upscales if needed. I'll maybe set it to 3 or 6 on the next release. Need some more testing.
Thank you so much for sharing this workflow—it’s incredibly useful!
However, I’ve run into an issue: when I use the last frame of the generated video to start the next segment, the two clips never match up seamlessly. The next segment always appears slightly zoomed-in compared to the previous one, even though I haven’t changed any settings, including the resolution.
I'll look into that :) maybe I got a wire wrong with all the changes -.-
Should be fixed in 1.5
@darksidewalker Thanks so much!
This workflow is awesome, thank you so much! First time using WAN and having a good time.
My question would be how would I change the scheduler? I'm going to assume it's doing Euler but I don't see any place to tell. I think there would be a K sampler node somewhere? Maybe I'm blind.
Also, how would I get my output video to 1920 x 1080? I have a 3090ti, hopefully that is enough.
The "WanMoeKSampler" nodes, there are the sampler and scheduler. You set your desired resolution to that, but you will have to make the correct aspect as input that fits 16:9
@darksidewalker Thank you very much, I figured it out! Now to figure out why the last picture keeps ending up zoomed in sometimes. I'm using 1.5 if that matters, but I assume it's me, because it doesn't always happen. I use a 400 x 400 picture and it zooms in. I use a 1920 x 1080 or a 3840 x 2160 it doesn't happen.
@Venkas 400x400 is 1:1 and nor divisible by 32, like wanted for WAN. You should use inputs and outputs that match their aspect at least remotely.
Also this zoomed in is from FLF? Normal i2v does not do this as far as I'm aware.
@darksidewalker Ah ok, thank you I will make sure to use proper ratios. The zoomed in is happening with Normal I2V, I'm going to try FLF and see.
EDIT: Yea I'm pretty sure it's the ratio I'm using, FLF doesn't do it with my 1920 x 1080 images. Good thing to know, but thank you for your assistance! Learning new stuff is awesome.
dumb question, how to I load the GGUF models with the load checkpoint you put in? Only ones that show are my safetensors checkpoints
It can not load ggufs, you would have to change the Loader node.
BatchResizeWithLanczos
The following operation failed in the TorchScript interpreter. Traceback of TorchScript (most recent call last): File "C:\StabilityMatrix\Packages\ComfyUI\venv\Lib\site-packages\torchlanc\torchlanc.py", line 356, in resample1d_jit resampled_flat = (gathered_pixels * weights.unsqueeze(0)).sum(dim=-1) else: resampled_flat = torch.empty( ~~~~~~~~~~~ <--- HERE (num_rows, out_size), device=tensor_flat.device, dtype=tensor_flat.dtype ) RuntimeError: Allocation on device
every time same error
You did not install torchlanc with pip, its in the dependencies.
OR
You run out of memory and OOM, your target resolution is too high.
@darksidewalker I'm the person who wrote TorchLanc and WhiteRabbit!
It's not necessarily about the target resolution. They could also lower the batch size. TorchLanc attempts to adjust batch size dynamically but it only works to a point. The Resize to Target node in WhiteRabbit has a batch size parameter that maps 1:1 with TorchLanc batch size and lower batch size = less VRAM per batch at the cost of speed.
@ArtificialSweetener_ Thanks for the insights!
Help, I can't figure it out. How and where do I adjust the number of steps on the samplers?
When I set it to 4, it takes 1 on high and 3 on low. If I set it to 6, it takes 2 on high and 4 on low.
How can I set it so it takes the same number of steps?
That's correct. With moe-ksampler is always uses the correct noise-to-latent ratio to swap steps. You would have to use standard k sample to change to custom step swapping.
Thank you always for sharing such valuable resources.
I have one more question. I noticed that opinions are quite divided regarding the use of SageAttention.
I’m currently using an RTX 5090, and I was wondering if I may politely ask for your thoughts on using SageAttention?
Sure! I gathered my thoughts on my Guide here: https://civitai.com/articles/20293/darksidewalkers-wan-22-14b-i2v-usage-guide-definitive-edition
TL/DR: I wouldn’t recommend.
It seems like in ComfyUI 0.3.67 - the Steps/Frames/Width/Height nodes no longer allow you to set a value, anyone know a fix? They are base PrimitiveNodes for me with no options to change them
I'm using 0.3.67 and 0.3.68 and all works, but the last days comfy seems to mess with updates and maybe destroyed python pip dependencies. I needed to recreate the venv and reinstall all dependencies.
@darksidewalker Hmm, I'm not sure how to do this in the Desktop version, I tried python.exe -m pip install --upgrade pip which did upgrade it - but still running into the same issue
I also just fully uninstalled all the custom nodes and reinstalled them
Ok, I fixed it, I needed to:
1. Fully uninstall all my custom nodes
2. Delete the custom node files in the folder
3. Trigger a Reinstall through the desktop app
4. Reinstall custom nodes
Step 3 fixed Frames, Width, Height - and Step 4 fixed Steps