This is a very easy and quick workflow to use. It's I2V and gives good results.
Description
FAQ
Comments (19)
In all workflows with wan 2.2 (including this one), I get the same error "
KSamplerAdvanced
Given groups=1, weight of size [5120, 36, 1, 2, 2], expected input[1, 32, 21, 104, 58] to have 36 channels, but got 32 channels instead
" I downloaded the correct models, but I still get this error. I updated comfyui to the latest version and still haven't solved the problem. I have all the models and loras. Can anyone help me with a solution?
Solving these types of multi workflow problems can be difficult. Take the full error you quoted there and search it in Google, you will find a number of links and posts about possible solutions, etc. Good luck.
Hmm maybe try downloading a different clip model. umt5 xxl has different versions like a scaled version etc
What vae Clip, Unet, you using? sometimes you have to reselect the model in the node because it takes the first one you have in the models folder. That error is usually because one of those things isn't compatible with the others. Copy the full error to ChatGPT or Grok, and they'll tell you how to fix it.
@saehara151 I downloaded exactly the recommended models from the workflow. We cannot rely on AI models such as ChatGPT or Claude to resolve these situations. I tried. I even uploaded the workflow to ChatGPT, and it was a significant waste of time. I hope that the next WAN model will be ok. I am not the only one experiencing these issues. That's it, time to move on.
@drak0n Did you try downloading a different clip like I mentioned? Takes a minute, will possibly solve everything. Also yeah like Saehara said make sure you physically select each model in the workflow, the ones shown when you open a workflow might be in a different location to your computer, so it doesn't actually load until you select them
@Light2020462 Thank you for your reply. Yes, I checked each model. Normally, if they have the same name and are in the correct location, they are selected correctly in the workflow from the start. For "clip," I selected "umt5_xxl_fp8_e4m3fn_scaled.safetensors."
@drak0n Ah damn, well if you ever figure it out let me know out of curiosity.
Yeah like you said I guess you can wait until the next big video model in the future, there's always gonna be new things.
@drak0n This isn't a workflow problem, it is a problem with something on your setup. Like you said originally "In all workflows with wan 2.2 (including this one), I get the same error" so the common denominator is your setup. Something is wrong, missing, invalid, corrupt, it is a mystery that only you can solve. We have no idea what you have done, other than report you have done everything right, but obviously something is wrong. If it were simply a wf problem then there would be more people complaining. Now, despite me saying all of that, I've not used the wf and it very well could be a wf problem by the author. But this is the nature of the beast, it is a million things interfacing together to form ai video titties, not just this one wf doing all the work.
So, here is what I would say, run the official wf first. use GGUF if you can't run it with full models. If you can not get official workflow to run, then begin dissecting your setup, the versions of things you have installed, making updated, looking at the server logs, error logs, searching Google and AI for answers. Remember, this is a million things coming together, so AI in this case is your best friend, you need to piece together years of work done by thousands of people across millions of lines of code, there is no person who can whip up the answer. AI and search and trial and error are your tools. Its your only tools.
@ToxicBot I think you're confused. I didn't say that these workflows are to blame and that I'm perfect. I just asked this question, hoping someone could help me with an answer. Anyway, I'll stick with WAN 2.1 until something new comes along. I have nowhere else to look. Don't worry, I searched online, I asked ChatGPT and Claude for solutions. Nothing worked. It's not a tragedy. I'll wait. By the way, I'm not the only one with this problem. Also, at the risk of repeating myself, it's clear that the workflows are fine.
@drak0n copy me your complete error and I'll ask grok, I always find solution with grok.
@saehara151 Thank you for your reply. This is the error: "KSamplerAdvanced
Given groups=1, weight of size [5120, 36, 1, 2, 2], expected input[1, 32, 11, 120, 68] to have 36 channels, but got 32 channels instead". There are also these details: "!!! Exception during processing !!! Given groups=1, weight of size [5120, 36, 1, 2, 2], expected input[1, 32, 11, 120, 68] to have 36 channels, but got 32 channels instead
Traceback (most recent call last):
File "F:\ComfyUI\ComfyUI_windows_portable\ComfyUI\execution.py", line 496, 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, hidden_inputs=hidden_inputs)".
@drak0n 1º: Grok said me this: 1. Update ComfyUI and its dependencies
This error is often resolved by updating ComfyUI to the latest version, as Wan 2.2 requires support for specific channels in the model (36 instead of the standard 32 for SD). The basic update script is sometimes not enough; you need the one that includes Python dependencies.
If you're using the portable version of ComfyUI (such as ComfyUI_windows_portable):
Go to the update folder within your installation. ComfyUI.
Run the update_comfyui_and_python_dependencies.bat file (right-click and "Run as administrator" if necessary).
Restart ComfyUI after the process completes.
If you're using ComfyUI via Git:
Open a terminal in the ComfyUI folder and run:
textgit pull
.\update_comfyui_and_python_dependencies.bat
If you don't have the .bat file, download it or use the ComfyUI Manager to update.
Why this helps: Recent updates to ComfyUI (around July/August 2025) fix compatibility issues with Wan 2.2, including channel handling in the sampler. Users on Reddit report that this resolves the issue immediately.
If you're using Stability Matrix or another launcher, update the ComfyUI package from there and restart.
2. Verify and use the correct VAE for Wan 2.2
The error can occur if the VAE (Variational Autoencoder) does not match the model. Wan 2.2 has its own optimized VAE, and using one from Wan 2.1 or the wrong fp32 causes channel mismatches.
Download the Wan 2.2 VAE from Hugging Face (search for "Wan_2.2_ComfyUI_Repackaged" or the official Comfy-Org repo).
Alternative test: If the workflow is for WAN 2.1, switch to its VAE (253 MB or bf16), but for 2.2, stick with the 2.2 one. Some users report that trying different VAEs (2.1 vs. 2.2) resolves mismatches.
3. Disable or remove conflicting custom nodes
Many WAN workflows use custom nodes such as "ComfyUI-WanVideoWrapper" or "WanImageToVideo (Flow2)", which can interfere if they are outdated or incompatible.
Temporarily disable all custom nodes:
Add the --disable-all-custom-nodes argument to the ComfyUI startup .bat file (edit it with Notepad).
Restart and test the workflow. If it works, the problem is a custom node.
Specific nodes to check/remove:
Find and remove the ComfyUI-WanVideoWrapper or Wan2Flow folder in ComfyUI/custom_nodes/ (or any variant like "flow2-wan").
If you use ComfyUI Manager, update or uninstall Wan-related nodes (e.g., WanVideoSampler).
Restart ComfyUI after removing.
Why this helps: In GitHub and Reddit discussions, removing "WanImageToVideo (Flow2)" or similar wrappers resolves the error because these nodes alter the channel flow incorrectly in older versions.
4. Other quick checks
Correct model: Make sure you're using Wan 2.2 files (e.g., wan_2.2_14B.safetensors or version fp16/fp8). Place them in ComfyUI/models/unet/ or the folder specified in the workflow. Avoid mixing with Wan 2.1 models.
@drak0n 2º :Hardware and Memory: With [1, 32, 11, 120, 68], it seems like a small batch, but if you have low VRAM (less than 12 GB), reduce the video size or use quantized (GGUF) versions of Wan 2.2.
Check that the input image has the correct dimensions (multiples of 8, ideally 512x512 or 768x768 for Wan 2.2).
VRAM and Optimization: If you have a GPU with little VRAM (less than 12 GB), use the fp8 version of the model and reduce the number of frames generated (try 16-24 frames).
@saehara151 Thank you for your efforts. I updated ComfyUI and its dependencies and am still getting the same error. I don't think I have a hardware problem, especially with an RTX 4090 graphics card. So, I'll wait until the next WAN update. Thank you again for your efforts.
@drak0n activate your confyui virtual environment and use: pip check
does it tell you that there is a conflict?
@saehara151 It gives me the message "No broken requirements found." Anyway, it seems very strange to me. Wan 2.1 works, but I have problems with wan 2.2 on any workflow. It's clear that I have an incompatibility somewhere, or who knows what the reason might be.
@saehara151 I finally managed to get WAN 2.2 working. I removed some nodes, made some updates, and succeeded. Thank you for your support. I have one request: starting halfway through the video, it begins to lose color and experience color changes. Can you advise me on how to resolve this issue?
@drak0n I'm glad you were able to make videos with the workflow! What CFG do you have set? If it's too high, it can do that. Sometimes it's due to the length of the video.