Post "WAN ANIMATE 2.2 V2 IMAGE TO VIDEO TO VIDEO KING!" by "Aitrepreneur" from patreon

Stuart Galloway

the auto install bat wont run on windows 11 as it is flagging as a unsupported 16 bit application

Wallosh Wolfmann

If you use Wan2.2 you need both High and low noise lora. If you use Wan2.1 you only need the low noise one.

aradfuton

I just became a member to support your content and I must say it is frustrating to hunt down the workflow jsons. Please update your posts to include their respective workflow jsons.

Timeless1175

Hi Aitrepreneur, I am late to the party but I want to know if i add a WAN lora, how to do it? the loras i downloaded are to be used with High and Low noise in pair. so how do i do it in your workfow? thanks!

Avishek

where do the vitpose models go into? which folder?

SebasMad Gau

Hey, with the problem PoseAndFaceDetection you mention in the post. I cant find the folder comfyui windows portable with Runpod , pls help

caca drk

what settings can i tweak in this workflow to get the highest possible quality output? the faces are coming out a bit bland/soft... not showing enough emotion like the original videos im using. not enough eyebrow lifts, smiles, winks etc... how can i tweak this to make sure its copying the original video properly and not giving me a soft emotionless face on the output?

jack2pot

Just downloaded it today, same issue torch.AcceleratorError: CUDA error: out of memory

Ez

This no longer works for me. Originally it was great, churned out 1280x720 vids on my 4090 no issue. I can still churn them out on the image/text wan2.2 workflow (as well as having no issues on any of my own). Now though, It doesn't matter what I update, clean install, etc, or how tiny the vid and image are, it will always run out of memory.

Louse

Hello, I have installed the workflow I am very happy with it and I am glad I have the opportunity to use it, I just have a problem regarding the physics of the upper body of a person such as jiggling and realistic movements of the upper body and chest, how can I fix that? Thank you!

M B

Hi there, is there a possibility to change the workflow to GGUF Model of Wan2.2 etc. so it would work with my RTX3060 12GB VRAM. I installed everything like in the instructions and the workflow starts doing its thing, until i get a reconnecting message from comfy ui. In the past it meant for me that i ran out of space. Can someone help me please. Thanks a lot!!

troy X

Nevermind, it was a Chrome browser issue. Switch to Edge and the manager came back.

troy X

On Runpod anyone know how to get the ComfyUI manager back after the update all and restart. It's gone and no longer on the top right corner.

a pretty parrot

i made a loader script that runs everything after pod deploys. the one thing i always have to do separately is install librosa. you can do that using web terminal, this command should do it: source /workspace/ComfyUI/venv/bin/activate && pip install librosa EDIT: this is on runpod

Ryan B

Wondering if anyone else is struggling with the runpod install for the latest Wan Animate 2.2 workflow? Ive been struggling aall day, restarted like 4 times now, following instructions to a T and I cannot get it working. When following instructions exactly, I end up with an updated ComfyUI interface, but cannot access the manager (ui button never appears, multiple reboots dont help. In one of my attempts I was able to get the manager to work after manually installing it (have not successfully recreated this since) however even with that working, I could not get the custom nodes from the workflow to install. In particular, WAN Vide Wrapper refuses to import, no matter what i do. The only other real valuable info as to whats happening that I have come across so far is that a whole lot of the following warning for a number of nodes: "[DEPRECATION WARNING] Detected import of deprecated legacy API: /scripts/ui.js. This is likely caused by a custom node extension using outdated APIs. Please update your extensions or contact the extension author for an updated version." Any help would be appreciated, though my guess is that something got changed/updated and the install script needs updating, but IDK. Thanks in advance for any help. I love the channel and wish Aitrereneur all the best!

James IV

I can never get any of these to work. I guess I need to learn actual AI coding because none of these are ever just 1 click install and load the workflow and go. Its just error after error and hour after hour of trying to fix them for nothing.

Taiga

close the running comfyui and go to ComfyUI_windows_portable/updata and run the update_comfyUI.bat file. if it's still a problem, try running the models nodes bat file.

Fizzy

Hey I tried creating a video and for some reason I am running into the following issue: WanVideoTextEncodeCached Error while deserializing header: incomplete metadata, file not fully covered I have installed everything but the runpod as I'm not using that. I'm not quite sure how to fix this issue. I get to around 30% and have that issue. If anyone knows how to fix this please let me know

Louse

Hey would you be able to please help me set it up as well I am having some problems with it and I do not know how to fix them.

Arkadiy Grigoryants

Thank you so much for your amazing work! This is honestly the best thing I’ve seen in the world of AI. I’ve been working in computer graphics for over 20 years - 3D, compositing, live-action integration - and what I was able to create using your workflow absolutely blew my mind 😍 It’s incredibly powerful and inspiring. Thank you for your effort and for everything you’re doing for the community 🫂

James Inage

Used the triton/sage attention installer bat and it seemed to have installed properly or it said it did but every time I try to use sageattention in one of your workflows i get the error PY_SSize_T_Clean macro must be defined for "#" format Anyone know the issue or possible fix?

Einziger

It did work for me with a 5090. First time I have nothing to tweak or modify.

JOHN

is there someone like this guy that makes 1 clicks that work?

Chris Rose

Do you know what would be great for this? If you could create a version that is like your wan extended video workflow. If you could upload a video regardless of how long it is, and have it broken down into sections, vram offloads and then goes on to the next section.

Emiliano

Ditto AI video editor (built on Wan, by the Wan team) dropped, hope to see a video on it soon. Our AI overlord always has insanely good workflows.

Mr Morphin Time

I was able to fix it. I didn't know how to restart comfyui in runpod without the manager button. But port 8000 has a start and stop button so I restarted comfyui from there and then manager showed up.

Nicholas Byford

Great work. Anybody know how to keep the black blocks from showing up in the final video when using the SEC masking?

Extreme Mojo

Hi and Thanks for all the work you do. I'm trying to use this workflow on runpod. I do the entire instilation as instructed, however it is not loading the following nodes: OnnxDetectionModelLoader, PoseAndFaceDetection, and DrawVitPose. The manager is not finding these three nodes when I try install missing nodes. Is there something I'm missing?

chibialien

He mentions it in the video. If you use the WanVideo Block Swap node, that is a part of the workflow, you should be able to offload from VRAM to system RAM

newtoai

Is there a place is the workflow to change the text Encoder used?

Frank B

Welp, in this workflow, Video seems to be a no go on a 3070 (8GB) I just get 'out of VRam' half way though generation, using the Q4KS gufs, I know its possible, Im sure there is a comfy node off off loading to System RAM, anyone?

whitebread

Hi, do i still need to put both low and high lora// or just only one of them

TheRangor

I'm having issues with the SEC masking too. The auto masking seems to work more consistently. With the SEC mask, more often than not, it will fail to completely mask the character and their face. Perhaps I'm just doing it wrong, I don't know.

James Inage

Used the triton/sage attention installer bat and it seemed to have installed properly or it said it did but every time I try to use sageattention in one of your workflows i get the error PY_SSize_T_Clean macro must be defined for "#" format Anyone know the issue or possible fix?

Tinkering

To some of the people having issues. I have found a couple things. First, the installs are not always perfect. Run issues or missing nodes have been cleaned up by just a fresh install. I also had to check the models individually. Re-downloaded one that got corrupted. I did have Mask issues several times. First you need make sure you are following the steps exactly. I had that user error a couple times. However, even after correcting my errors I had masking mess ups. I have found that restarting the program and the web browser fixed it - most of the time. However, when that did not work, reload a clean copy of the workflow. Just had an issue where I could not get the video to mask properly. Background would appear and disappear. reboot did nothing. New video did nothing. I redownloaded the workflow. All was back to normal. Just my two cents.

Pizza

Having problems with this. It worked at first, but now all of my videos that I load are skewed in the masking area down below. They are zoomed in a lot and I can't mask since it's so skewed. Do I need to match the h x w exactly with the video or picture?

Vaivij

Also, in the loras, do we put high noise wan2.2 loras or lownoise ones?

Vaivij

Is there a way to change the mask color? If I am replacing anything black, a portion of it comes back as the mask 'blocks' that don't go away, especially at the edges.

nurtu

Im not getting the best likeness to my ref, what do you recommend besides training a lora? Up the steps?

newtoai

Hi Guys. Is there an active discord channel where we can all chat and help each other with throubkeshooting , as supportes for this channel?

Mr Morphin Time

When using rtx 5090 or rtx 6000 ada on runpod, the manager button no longer shows up and I can't install custom nodes. Anyone know a fix for this?

Tinkering

fixed. did a file check of the three installs on my machine and the Wan2_2Animate-14B-fp8_scaled_e4m3fn_KJ_v2.safetensors had three different sizes. Ya that would be an issue. Downloaded another from Huggingface and now it workss!!!

Tinkering

First OVI was great. Lots of fun. Question. I have installed this as an addon to my existing Comfyui and as a new install. I have used videos created with OVI and local images and web based images. I am doing the exact same setup as your first demo section of the video (though I enter the video size 992/512 to match the input video. Note: I also updated comfyui from .bat file. Regardless of what I do I get. WanVideoModelLoader Error while deserializing header: incomplete metadata, file not fully covered.????

Lance Fanger

Will the node installer and Triton/Sage attention installers work for the base version of ComfyUI or does it have to be the portable version?

Jean Luc Go 💨

Everything in the workflow works great until I tried using the SeC masking. I followed all instructions in the video, but I get this error: SeCModelLoader Failed to load SeC model: No module named 'omegaconf' I tried downloading a fresh copy of the SeC model from the following link, but that didn't help: https://huggingface.co/Aitrepreneur/FLX/blob/main/SeC-4B-fp16.safetensors I tried (re)installing omegaconf with the following command, to no avail: pip install omegaconf This is the output: Traceback (most recent call last): File "C:\AI\ComfyUI_windows_portable\ComfyUI\execution.py", line 498, 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) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\AI\ComfyUI_windows_portable\ComfyUI\execution.py", line 316, in get_output_data return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, hidden_inputs=hidden_inputs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\AI\ComfyUI_windows_portable\ComfyUI\execution.py", line 290, in _async_map_node_over_list await process_inputs(input_dict, i) File "C:\AI\ComfyUI_windows_portable\ComfyUI\execution.py", line 278, in process_inputs result = f(**inputs) File "C:\AI\ComfyUI_windows_portable\ComfyUI\custom_nodes\Comfyui-SecNodes\nodes.py", line 435, in load_model raise RuntimeError(f"Failed to load SeC model: {str(e)}") RuntimeError: Failed to load SeC model: No module named 'omegaconf'

dan

every time i try to run the WF the resize image node for the reference image halts the process and i get no generation. any ideas how to fix?

Cristo

Would be interessting if we could upload more than one image. Let's say a person, front and back. Because with only one image the AI won't be able to guess what the image reference looks like from behind

H-BOX

Hi! I somewhat made it work on vastai, reinstalling manually the missing nodes about ComfyUI-WanAnimatePreprocess, thanks for your work ! One more question : How can I change the fps ? I do not want to use the original one, because that is too much. Wan goes well starting from 16 so I changed the force fps to 16 but I accelerates the movements in final results. Any easy tweak to change that ? Also, just wanted to say it's incredible because I was using a wananimate WF I found randomly and it sucked compared to this one, didn't even take the face etc ! Thanks !!

Einziger

This is awsome! It would be nice to have a installer with everything (Wan/Qwen/Flux...) on a regular basis. Can you do that please?

Chris Russell

Local and I will try that! And just a question outside this workflow do you have multiple versions of comfy. I have like 4 already haha

Aitrepreneur

local or runpod? if it's local just use the WAN-ANIMATE-MODELS-NODES_INSTALL.bat file inside the comfyui folder or just install the ComfyUI-WanAnimatePreprocess custom node For runpod, try again, it might have been a temporary runpod issue because everything works now with the installer

Aitrepreneur

I don't understand your question

Khajiirah

Sorry i did missunderstand, so i deleted my comment. I wrote something like "try to replace the buggy node wich gives you the error, not copy it!" But your Problem is different, i will look into that. It could mean, that inside your video, the characters head turns away and so it can not recognize the face correctly.

Khajiirah

If you cant find VitPose Custom Node, its called "ComfyUI-WanAnimatePreprocess" from Kijay, in ComfyUI Manager to download and install. After installing this you get the three Nodes. After i downloaded the "ComfyUI-WanAnimatePreprocess" with the Manager and restarted ComfyUI, it worked and all three missing Nodes where available and nomore Red. =^..^= Have Fun! If it won't work afterwards, (does nothing when clicking Start), make sure to enable the Mask Group (down below), as inside the Masking Group is the Export File inside (combine and save video), without it, it does nothing as it does not know where to export and save the file.

Tom Baron

PoseAndFaceDetection cannot convert float NaN to integer. Anyone else getting this error? UPDATE GPT told me it was my video. "on frame 296 of your input video, no face was detected. When it tries to compute a bounding box from missing face data, some math divides by zero, which creates NaN values—and that's what crashes the whole workflow."

H-BOX

Hi, it's amazing that you made this. but on the runpod, even after upgrading everything, it doesn't solve this time. Here are the missing nodes that doesn't fix : OnnxDetectionModelLoader PoseAndFaceDetection DrawViTPose

Ricky Lawrence

Did you know that in the video, the mask is connected to...? but in the workflow json file, it is not? where does it connect to?

Banloss

check the description here again, he has notes on that error, just below the first error listed

Vincent Nestler

Here is my current issue - which may be that I am on a 5090 laptop: PoseAndFaceDetection [ONNXRuntimeError] : 1 : FAIL : Non-zero status code returned while running QuickGelu node. Name:'/model.0/act/Mul/QuickGeluFusion/' Status Message: CUDA error cudaErrorNoKernelImageForDevice:no kernel image is available for execution on the device Thoughts?

Sequoia

Thanks a lot! is it normal for 1 second of Video to take 1.5 minutes to be generated (25 frames)? (RTX3090)

Vincent Nestler

did not work for me either.

Oziel Leal Salinas

Thank you!

Phillip Gimmi

the workflow you reference does not work in runpod

Chris Russell

this is amazing! But having an issue with some nodes. I tried updating and I cant seem to get them: PoseAndFaceDetection DrawViTPose OnnxDetectionModelLoader

Travis Hightower

ok - so everything seems to work except when I try to run a template, a bunch of models are missing. I can download them, but don't know where to put the files once I do.

Aitrepreneur

it is a wanvideowrapper node update issue that is not up to date go into the ComfyUI_windows_portable\ComfyUI\custom_nodes folder and delete the ComfyUI-WanVideoWrapper folder then if you click on the folder path, type cmd press enter, it will open a cmd window in that folder and inside you can type: git clone https://github.com/kijai/ComfyUI-WanVideoWrapper ..\..\python_embeded\python.exe -m pip install -r ComfyUI-WanVideoWrapper\requirements.txt it will clone the latest version of the node and install the requirements, then you can reopen comfyui and redrag and drop the workflow in again.

Aitrepreneur

what's the error exactly?

Aitrepreneur

Bro I cannot agree more unfortunately, believe me I wish there was an easier way, I spend so much of my own time troubleshooting shit on my side that I also get annoyed with it but unfortunately comfyui is the best more versatile open source tool there is right now, so yeah this all come to a cost....if you got any issues you can always dm me, we'll try to find a solution

Aitrepreneur

never had this issue during testing, are you sure that you are using my comfyui template?

Aitrepreneur

Image gen? you mean video gen, what parameters are you using?

Aitrepreneur

try restarting comfyui and open in another browser

Aitrepreneur

it's the install_triton_and_sageattention_auto.bat file

Boxertwin

Hi, first attempt I got this message "self and mat2 must have the same dtype, but got Half and Float8_e4m3fn"

jon rich

i think new windows + nvdia update is breaking it

jon rich

i got it 2. its only stable diffusion that is causing the crash

jon rich

weird it says my 64 bit cant run this because its not up to date? Is all of stable diffusions crashing this month?

Nick Sarkisov

guys, how fix this type of error: Failed to update ComfyUI or several extensions / TypeError: Failed to fetch. When I'm updating comfy.

evil chucky69

you know i must have installed comfyui, both portable and std versions, on my pc, at least 12 times. i have to say not once, not ONCE - have i been able to get it operating without some kind of crap that needs updating, installing, re-installing, downloading....blah, blah, blah. missing nodes, missing this and that, outdated this, change the version to this...blah, blah, blah...what a piece of shit program that promises sooooo much, yet delivers sweet fk all! the day this program gets nuked i will be dancing in the fken street with no clothes on!

Sense Departed

download git from git-scm.com, the one click downloader didnt help me with that one, had to do it manually

Emilien-Edmond D'aramitz (LeonceTheLion)

In case anyone is getting this error in runpod when trying to install the .sh file: "git: 'lfs' is not a git command. See 'git --help'. The most similar command is log" here is a solution that Gemini helped me fix: Even though you are on a specialized RunPod template (like the ComfyUI one), sometimes these base images do not include Git Large File Storage (LFS) by default, especially if they are trying to keep the initial image size small. The script you are running, WAN-ANIMATE-AUTO_INSTALL-RUNPOD.sh, is clearly trying to clone a repository that uses Git LFS to manage the large model files, and it's failing because the lfs command is missing. Since you are running as root in a Linux environment (likely Debian/Ubuntu-based), the fix is simple: install Git LFS. Here are the commands you should run in your RunPod terminal (JupyterLab terminal or web terminal): Update your package lists: Bash apt update Install Git LFS: Bash apt install git-lfs -y Initialize Git LFS (Good practice): Bash git lfs install Re-run your installation script: Bash ./WAN-ANIMATE-AUTO_INSTALL-RUNPOD.sh After running these steps, the script should execute without the git: 'lfs' is not a git command error, and the necessary large files for WAN-ANIMATE should be downloaded successfully.

troy X

Nevermind. everything worked after turning off triton.

E

flow worked for me but my image generation is super grainy. Any suggestions?

Alpha Ghost 47

For some reason once the ui loads up i can move around, the small key at bottom right is telling me what I am doing but the actualy comfy ui isnt moving??? (So I can't navigate or zoom in)

Larry1710

Not seen this one before - running Wan Animate 2.2 Auto_Install. Errors on git-setup.exe - "cannot start or run due to incompatibility with 64 bit version of windows". I've run other Installers b4 with no problems on another system that was WIN 11 Pro 64 - what did i break? 🙁

troy X

You must have an older GPU. This is a similar error on my 2080Ti

Travis Hightower

Is there a pre-required step I am missing before I try to run the WAN-ANIMATE-COMFYUI-MANAGER_AUTO_INSTALL.bat? I think I saw someone else with this problem last time, but not sure if there was an answer. I get this error pop up. Unsupported 16-Bit Application × The program or feature "\ ?? \C:\Users\Travis and Jennifer\Downloads\git-setup.exe" cannot start or run due to incompatibity with 64-bit versions of Windows. Please contact the software vendor to ask if a 64-bit Windows compatible version is available.

Travis Hightower

Hi there. Trying again to get this installed for the first time.. I am still getting this error when I try to run the installer. I apologize if it has already been addressed elsewhere

troy X

Does anyone here have an issue with the wan video sampler? it runs, then faults out with this.Command '['C:\\ComfyUI\\Wan 2.2 animate\\ComfyUI_windows_portable\\python_embeded\\Lib\\site-packages\\triton\\runtime\\tcc\\tcc.exe', 'C:\\Users\\JORGEN~1\\AppData\\Local\\Temp\\tmp96s1hq1k\\cuda_utils.c', '-O3', '-shared', '-Wno-psabi', '-o', 'C:\\Users\\JORGEN~1\\AppData\\Local\\Temp\\tmp96s1hq1k\\cuda_utils.cp313-win_amd64.pyd', '-fPIC', '-D_Py_USE_GCC_BUILTIN_ATOMICS', '-lcuda', '-lpython313', '-LC:\\ComfyUI\\Wan 2.2 animate\\ComfyUI_windows_portable\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\lib', '-LC:\\ComfyUI\\Wan 2.2 animate\\ComfyUI_windows_portable\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\lib\\x64', '-IC:\\ComfyUI\\Wan 2.2 animate\\ComfyUI_windows_portable\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IC:\\ComfyUI\\Wan 2.2 animate\\ComfyUI_windows_portable\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IC:\\Users\\JORGEN~1\\AppData\\Local\\Temp\\tmp96s1hq1k', '-IC:\\ComfyUI\\Wan 2.2 animate\\ComfyUI_windows_portable\\python_embeded\\Include']' returned non-zero exit status 1.

SVT-Fashion-ART

I can't find where you have the installer for TRITON?

Aitrepreneur

Yeah it happens, it's apparently a browser extension issue, open in another browser or in incognito mode with the same url and it will work

Leafar

Can you help me ? I'm trying to install on runpod, but, for some reason, the manager button doesn't show after the update and restart, i'm using the 4090

Aitrepreneur

have you tried the installer to install sage attention and triton first? also make sure that you have the latest nvidia drivers and that you have sysmem fallback enabled in the nvidia parameters. also how much ram you got (not VRAM, ram)? cause you usually need a huge amount of Ram to compensate for the lack of vram

Aitrepreneur

hmm...try uninstalling any 7zip install you got, it should be located in C:\Program Files\7-Zip folder then try relaunching the installer again or install 7zip manually with this link: https://www.7-zip.org/a/7z2501-x64.exe

Malachi Dolan

Hi I have a 4070 TI with 16 gigs of vram, I did a fresh comfy UI install using the automatic installer and I didn't get any errors, but when I try to do a run at 480×832×81f I run out of memory right away even with block skip set to 40, dumping my log in chat GPT it says that I don't actually have Triton or Sage attention running and that it's falling back to PyTorch attention, my question is should I be able to run this with 16 gigs of vram and the problem is that I don't have Sage attention running or is that just impossible and I need to get the quantized version of the model? editi just tried 16 frames at 240x416 and it still ran out of mem

8696969

Same issue. 5090.

So oN

Runpod seemingly has issues atm saying got no cuda device no matter if 4090 rtx 6000 or 5090 websearch for troubleshooting yielded a couple of posts of people running into the same issue it seems hope they solve it quickly can't wait to test this new workflow

ti

PoseAndFaceDetection [ONNXRuntimeError] : 1 : FAIL : Non-zero status code returned while running QuickGelu node. Name:'/model.0/act/Mul/QuickGeluFusion/' Status Message: CUDA error cudaErrorInvalidPtx:a PTX JIT compilation failed Any idea?

R4

-------- Checking prerequisites -------- 7-Zip not found. Downloading... % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 162 100 162 0 0 271 0 --:--:-- --:--:-- --:--:-- 271 The system cannot execute the specified program. 7-Zip install failed. Install it manually then rerun this script. Press any key to continue . . .

Aitrepreneur

can you send screenshots of the parameters you're using?

Nicolas

After installing node dependencies, adding the models and running default workflow settings I had: WanVideoSampler self and mat2 must have the same dtype, but got Half and Float8_e4m3fn Wan model is Fp8 and Lora BF16. Are the models on the workflow ok ?

Aitrepreneur

not sure, send me screenshots maybe

Aitrepreneur

Well perfect then, make sure you got all of the models installed though

Aitrepreneur

I've already updated the installer, people shouldn't have this issue anymore

_i_

I had a lot of issues with the install_triton_and_sageattention_auto.bat script. All the issues were related to the syntax in the for loop on line20. It got too complex for me to figure out but I knew what the script was needing so I hacked it to get it to work. If you are getting errors saying you don't have Torch installed but you do, here is how i fixed it: replace the entire for loop on line 20 of the script with this: for /f "tokens=*" %%A in ('"%PY%" detect_torch_cuda.py') do ( then create a file in the python_embeded folder named detect_torch_cuda.py inside that detect_torch_cuda.py file include this bit of python import torch print(torch.__version__) print(getattr(torch.version, "cuda", None) or "unknown") save the file and run the original install_triton_and_sageattention_auto.bat as normal - it should correctly pick up on your existing Torch installation - after that I had no issues. Hope this helps someone else :)

Jollyboy

unlike qwen image edit, here i just dropped the workflow in to comfy and everything works out of the box, nice. i have no idea what im doing right now

HealingPaint

Using SDPA first attempt, just based two things loaded on top (video clip on left, image on right), and get this "self and mat2 must have the same dtype, but got Half and Float8_e4m3fn" -- ideas?

Emmanuel Bronson

Grok helped me fix I had to drop down a version to 1.20 I think.

Aitrepreneur

try this one: https://we.tl/t-aSxYx5LWZ4

Aitrepreneur

it is for paying members, for the images you could technically use it by maybe inputting a frame cap of 1 so that it only takes the first frame of the video to transfer it to the image

Salihz

I am having this error with the auto installer PoseAndFaceDetection [ONNXRuntimeError] : 1 : FAIL : Non-zero status code returned while running QuickGelu node. Name:'/model.0/act/Mul/QuickGeluFusion/' Status Message: CUDA error cudaErrorInvalidPtx:a PTX JIT compilation failed Any idea where this could come from ?

newtoai

Hello. I am having this error message when using install_triton_and_sageattention_auto.bat to install them === Detecting Torch and CUDA === 'C:\Users\marco\Downloads\ComfyUI_windows_portable\python_embeded\python.exe" -c "import' is not recognized as an internal or external command, operable program or batch file. [ERROR] Could not import torch. Please launch ComfyUI once or install torch in this python. Press any key to continue . . .

Daniel

Is this part supposed to be open for everyone? Shoudn't this just be for paying members? Also, is there a way to make it work for just images? Love the pose transfer and mask for images.

Daniel

I do that all the time with no issue, I even do the same drive, just another folder.

Emmanuel Bronson

Can I do a second install of comfyui and would you recommend that? I'm wondering if i leave my current one just in case and install this in another drive