[Edit:
Version v5.0 works with latest comfyui (v0.15.0).
If you have any problems, please refer to the FAQ at the bottom of the page or have a look in the comments.
Many thanks to everyone who tested this workflow. Thank you very much for the many inquiries and, of course, for all the knowledge and experience you have contributed. here👍🙂
Special thanks to:
@SeoulSeeker for the "Dead Simple MMAudio" workflow wich are the basis of the audio part here,
@taek75799 for the really well working enhanced models
@Bakazaya pointing to the color issue in version v3.0 and running lots of tests,
@bluntfeather sharing latest experiances with installing Comfyui-Easy-Install,
@nitrovtx for remain persistent in matters of quality and running a lot of tests,
@Icey64 for providing the link to "Comfyui-Easy Install",
@boinobin730 for asking for a First to Last Frame option, running pre tests and responding fast as hell 🙂 and
@SnowShoes311 thank you so much again for all your buzzing 😋]
Features:
Optimized Wan 2.2 workflow, runs perfect on RTX 3060 12 GB VRAM GPU and 32 GB RAM,
"Text to Video", "Image to Video" and "First/Last Frame 2 Video" generation in one workflow, all with easy audio generation,
easy installation/model downloading, all necessary sources are specified,
easy to use workflow, clearly structured, all necessary steps are explained,
easy switches for mode selection,
easy prompt selection for fast prompt creation/testing,
easy switching between "standard" and "enhanced" models,
very fast and smoth high quality outputs up to aprox. 1440 x 960 with 60fps,
2x fast upscaler,
4x fast framerate multiplier,
MMAudio Sampler (generates sound accordingly to the video action),
Triton and Sage Attention option,
A 5 Second long high quality video generation takes about 10 - 15 minutes (see below).
Tested generation times:
As a rough guide value for RTX 3060 GPU: generating a 5 second long high quality 1440 x 960 60 fps video with 6 steps it will take:
t2v: around 10 - 12 minutes,
i2v: around 15 minutes.
Comfyui-Easy-Install with Triton + SageAttention:
This workflow should work with any latest comfyui version >v0.6.0 (Desktop, Embedded, Windows/Linux).
However, comfyui is developing rapidly, and it often happens that some of the custom nodes used are not updated quickly enough or not updated at all. Manual workarounds are sometimes necessary. Furthermore, care must be taken to ensure that there are no conflicts with other nodes.
If you're having difficulties with your existing comfyui system or if you want to run video generation on a separate (parallel) comfyui system, like I do, I would recommend you the following installer: https://github.com/Tavris1/ComfyUI-Easy-Install.
Complete installation of comfyui including manager and some pre configured custom nodes is just one click - really 🙂
Installation of Triton + SageAttention is just a second click - really 🙂 And since it's so easy now, I would definitely recommend it to you for video generation.
Cause it is an embedded version, you can install it parallel to your existing comfyui version without the risk to ruin your working system.
After installation just configure the "extra_model_paths.yaml" file to use your existing models.
After a fresh installation of Comfyui-Easy-Install you might have some issues too, but there are known workarounds - please see the FAQ below.
For testing/understanding/experimenting/changing the workflow:
Click "Toggle Link Visibility" to see the links.
click the Subgraph symbols to open the Subgraphs.
for quick testing you may lower the settings for: steps, clip lenght and video resolution,
be really carefull with modifying Groups or Subgroups (even Titel or Color) cause they are essential for switching,
feel free to try and test other models. Just give me a hint if you find models which deliver better results and fitting the 12 GB VRAM limit.
And as usual: Have Fun 🙂🙂
Short Conclusion:
This workflow is based on elements of a variety of allready published workflows. My "job" was only to put things together, optimize it for a small machine and create a most simple and hopfully user or even "beginner" friendly workflow.
I`m not an "expert" - just a user who wants to get it running on "available" hardware.
There are many things I don't really understand. If you find mistakes or better solutions please give me a hint.
And I really hope that even "beginners" have a chance to go the first steps...
Frequently Asked Questions (FAQ):
For quick and better overview I will try to merge all known issues here - step by step (please be patiant). If your issue is not listed here, please have a look in the comments first. Most issues have been allready discussed.
Comfyui Nodes 2.0:
Turn off Nodes 2.0 in comfyui (use comfyui menue). Actually not all custom nodes are supported.
Comfyui crashes after generation while vae decode, upscaling or frame rate multiplying (Rife VFI) without any error report:
This is a RAM problem (not VRAM). Increase your swap file (min. 64 to 128 GB) or set it to automatic management on a fast drive with at least 100 GB free space.
JW Nodes (JWFloatToInteger, JWIntergerDiv, JWImageResizeByLongerSide), soundfile missing:
For the workaround look here and here:
python -m pip install soundfileFresh Comfyui-Easy_Install Installation (missing soundfile and Pytorch v2.9.0 issue with SageAttention on Windows:
For full conversation look here.
Open cmd in python_embedded folder:
python -m pip install soundfile python -m pip uninstall -y torch torchvision torchaudiopython -m pip install torch==2.8.0 torchvision==0.23.0 torchaudio==2.8.0 --index-url https://download.pytorch.org/whl/cu126Slider Nodes - how can I modify the "default" values:
Right click the slider node, choose Properties and set the values you like 🙂🙃
Description
Added MMAudio Sampler for easy sound generation.
FAQ
Comments (76)
@330am Thank you for buzzing 😋🙂
Which node should the VAE in the "KSampler" node connect to?
@ogg797 It is allready connected of course 🙂 I use the "magic" AnythingEverywhere nodes. Go to the group "00 Clip and VAE Model" and hoover with your mouse over the node to see the links.
@arkinson However, the entire workflow stops when it reaches the "KSampler" node.
@ogg797 Give a complete description of your issue and you might get help here.
@arkinson The workflow stopped at the "KSampler" node. When I double-clicked to open the "KSampler" node, I found that the border of the "VAE Decoding" node had turned red, clearly indicating an error in this node.
@ogg797 Sorry, I was on the wrong workflow - so forget my first comment. Ok, if I got you right, comfyui server crashes without any error message at the vae decode part - right? Please look at my FAQ here on the model description and increase your swap file.
[Edit: Just updated the FAQ: swap file size min 64 GB]
@arkinson I'm very sorry, I'm too busy with work to reply to you promptly. Are you referring to changing the computer's virtual memory to at least 64GB?
@ogg797 No hurry 🙂 Yes (if you have the typical "silent" server crash as a result of low RAM). But look at my FAQ here.
My GPU is 4090,96 GB of RAM, and I've set the virtual memory to 64–128 GB. The problem remains
Good job getting the sound integrated! Still my go to workflow. I just love it!
@SnowShoes311 Hi - thank you so much 🙂 If there is too much "moaning" you might try the "standard" audio models too, linked by boinobin730 here. I did not tested it yet, but I will do. Maybe I will add some options to select the models or bypass the audio part at all.
Actually we are testing ltx-2 workflows running quite fast with 12 gb vram. If you like have a look here too. Still struggling with some basics, but some tests, like v2v are pretty cool and delivering crazy results...
I always use this workflow. And Q8 Enhanced models running pretty good on my 3060Ti with 8GB vram. 32gb RAM + 48gb pagefile + sageattention. It takes 5-7 mins for 5sec vertical videos and 6-10 mins for horizontal. Idk why its longer but its ok I think :D Thank you!
Hi - thank you so much for your feedback. I had never thought you could run Q8 models with just 8 gb vram 👍 Yes, the devil knows why landscape needs longer 🤣
@arkinson Btw, thank you so much for the explanation of almost every node and the adding of MMAudio, especially since it works automatically.
@wemite I stumbled across mmaudio by chance and, since it's so easy to use, I decided to give it a go as a gimmick. I haven`t tested the "standard" models yet, but you may have look here.
I`m just into LTX-2. Video + Audio is really amazing and we got it running with 12 GB VRAM. I am not sure, if it will work with 8 GB too, but the VRAM management seems to be completely changed with the last comfyui releases. If you like, have a look in our discussion here (just scroll down a little bit). I actually prepare a more automated workflow for 12 GB VRAM....
Any tips to reduce the character bouncing?
It seems whenever I generate a video with this flow using i2v the characters tend to "bounce" (oddly moving back and forth, jiggling)
@requiemss Best way would be you post an example with prompt and models used.
@maxivh Thank you so much for buzzing 🙂
How can I disable upscaling and audio?
Simply copy the video combine preview node and move it right beside the KSampler subgraph. Connect image output with image input and manually set Group 05 to bypass.
@arkinson Thanks
@arkinson I was a little hasty, I figured out how to turn off audio and upscaling, but I need a 16-frame video to go to 60. For this, I understand I need a "Rife," but it's included in this set of upscaling and audio.
@VlrgRomNS Open the subgraph in 05, copy the rife node and append it next to the KSampler subgraph. Set the right frame rate settings in the rife node as well in the video combine node. This stuff is all really simple - just a chain of image processing.
Where can I download mmaudio models?
Look at the model link note in the workflow - click on the link and you get it 😉
how can i use this in MimicPC?
UnetLoaderGGUF
expected str, bytes or os.PathLike object, not NoneType
I keep getting this error, and the 3 boxes (both mmaudio and upscale rife audio) in box 5 are framed in red.
Any help is appreciated, this is a learning curve for me. thanks!
@ph480parthenon504 Hi - seems you did not selected YOUR local models in the loader nodes. Click in every blue loader node and select YOUR downloaded model.
[edit: stupid question: you have installed the loader nodes???]
Thank you so much for the reply!
I did not get every one loaded, sorry. I'm using the enhanced models listed in the link above and am getting a blurry video, checked all the loader nodes and everything checks out. Thoughts?
@ph480parthenon504 Sorry, but I don`t get you. You have solved your above error message and you have a new issue now??? Please describe step by step what you have done and what you are trying to do and what error messages you get.
You should check this out !
https://github.com/Jasonzzt/ComfyUI-CacheDiT
A new way to speed generation time for both image and video generations without affecting the quality
https://www.youtube.com/watch?v=nbhxqRu21js
and here is a tutorial for it
maybe you will consider updating the workflow after this
thanks for your great work by the way i really appreciate it
Thank you, but did you read your source? 🙄😉
Q: Does it support distilled low step models?
A: Currently, only Z-Image-Turbo (9 steps) has been tested and verified. Other low-step distilled models require further validation.
For extremely low step counts (< 6 steps), the warmup overhead significantly reduces the benefit - sacrificing quality for minimal speed gains is generally not worthwhile in such cases.
@arkinson, MMAudio can make people talk? I only seem to be able to get sound effects?
No, is far as I know MMAudio can`t generate speach. As tested, it can generate very simple noises that might be funny in your videos - maybe like some silly "Kung Fu" cries for example 😂 Btw. the provided nsfw models generating a lot of "moaning". You might try the "standard" models too, linked by boinobin, please look here. I had not the time to test them yet, but I am keen that someone else will do some serious comparison tests and provide the results here 😉
If there are any advantages, I would integrate it in the next workflow version....
Speed up guide that worked for my RTX 3060Ti. Decreasing batch_size for RIFE from 8 to 4 and decreasing from 60 to 25 fps, decreasing motion_amplitude from default 1.3 to 1.15-1.20 (cause my animations are not too complex) in PainterI2VAdvanced, disabling mmaudio - decreased my generation time for 480p on 2 high / 3 low steps from 5-6 minutes to 2-3 minutes. Now it takes 5-6 minutes for 720p!!! With enchanced models and 2 additional loras for sure (1 of them is heavy). It speeds up yes. Not the same impact on each animation, for me it works good with vertical orientation.
Yes, you can keep mmaudio, it doesn't take so much time, but I prefer speed instead of sound, at least with such quality we have now.
This is awesome, I'm so fucking happy ahaha
Of course, to achieve good speed, you must perform several generations to "warm up" as always
I tried CacheDiT, it doesn't work properly, so don't even try it in this workflow guys
@wemite Thank you so much for testing and publishing your experiances for the RTX 3060Ti (with 8 GB VRAM if I remember right) 👍
I will check your hint with final 25 fps for my gpu too. Your generation times refering to a 5 second video?
I have no experiance with the RTX 3060Ti, but I assume this gpu should be a little bit faster then the RTX 3060 (except the "bottleneck" of 8 gb vram in your case)?
MMaudio: I believe I better add a switch option for easy turning on/off in the next update.
Thank you for testing CacheDiT too. I never tried it, since I was already skeptical about the published description.
I´m really glad that the workflow is usefull and you are fu**king happy 😂🙂 I`m really learned a lot here too with all the user input and experiance.
@arkinson yes, I'm talking about 5 sec video. I ended up generating videos longer than 5 seconds because looping artifacts are inevitable.
thanks for crediting me in your description mate. looks like an awesome workflow
Hi - thank you so much for "passing by" here 🙂 MMaudio is a lot of fun. I had never thought, that it would be so "deadly" simple to spice up some stupid videos with some funny sounds. Your workflow has enlightened me 😅 Thanky you bro.
@kAIbold Hi -thank you for buzzing 🙂
Can this workflow be used with a 4060ti graphics card?
Why not??
does it work with a 5070 12gb? lol all these files are huge to download..gotta ask before i download
@akatakeoverz126 I don`t now the specs of your card. 12 gb vram??? This will definately work, but see my desription about ram + swap file too.
I think I downloaded your workflow about a month or two ago, it was working well, now my character lora doesn't seem to work at all. I don't recall it having the lightning loras added or sound added previously, just wondering if you have any of the old versions (without sound?) Or am I thinking of a different workflow.
@calepornhub2895 I don`t know what you did nor I know if you are thinking of the right workflow 🙄 but feel free to download any older version here you like 😉
@arkinson thanks, did they always have sound? I went back to the earliest versions and they still have sound incorporated? I remember a workflow with that same video of the cat that was working well for me and the structure seemed very similar outside of no lightning loras and no sound, but for some reason it isn't working for me now, at least not with my character lora. Thanks for the workflow though!
@calepornhub2895 I really don`t get what you are doing.
Just click here on the menue over the cat image at v3.1 and download the zip file from this v3.1 page as usual. This is the last version without sound. And read the version descriptions if you are not sure wich version you want to load.
But I believe you have a completely other issue, maybe something wrong with your comfyui system after the last comfyui updates? Cause my version v4.0 with MMAudio did not change anything in the video generation part and Lightning Loras did not changed over the last versions.
@arkinson TY!
I installed every Custom Node package. But the "RIFEInterpolation" is still market in red. Any tips?
@leylalovegood855 Wich comfyui version? Search for "Comfyui Frame-Interpolation" in the Manger or use the github page for manual installation.
@arkinson I use the newest ComfyUI-Desktop Version (v0.8.6). I have already installed "Comfyui Frame-Interpolation" but the "RIFEInterpolation" Node is still marked in red. I can use the workflow right now only without Group 05.
Ok, I often see user issues with the Desktop version especially after the major comfyui updates. Without any detailed informations (what means "node is marked red" - not installed? execution issues? error logs? etc.) it is hard to guess what your issue is, but I would assume it is just a node conflict.
If you can`t fix it by yourself one consideration would be to install a "clean" Comfyui-Easy-Install just for video generation (see my FAQ here).
Help please, when i press run on comfy ui it says Error Cannot read properties of undefined (reading 'workflow'). How do i fix this T-T
nevermind, just update the comfy ui to the latest ver. which is weird since runpod said that it was the latest ver
@rupertmurdochtown816 Actually I get the same error for myself with Comfyui-Easy-Install, release v0.14.1 (release v0.12.x is still working). This error is really strange, cause i could not identify any conflicting or unsupported node yet. If any one else has this error or find a solution, please give me a hint.
Hi im getting this error when running:
[Errno 2] No such file or directory: 'C:\\Games\\ComfyUI-Easy-Install\\ComfyUI-Easy-Install\\ComfyUI\\models\\mmaudio\\nvidia\\bigvgan_v2_44khz_128band_512x\\bigvgan_generator.pt'
I'm using the latest comfy ui easy install
@wobmleyplyontek220 Strange error. Do you loaded the right models to ...\models\mmaudio and selected them in the loader nodes?
is there a way to use PainterI2V with VACE?
I run out of RAM during the Low-Noise KSampler process using the Enhanced Q8 GGUFs :( Is torturing my SSD really the only option to solve this? RTX 3060 12 GB + 32 GB RAM
@BoredJake Not sure if it is your issue, but actually I get a lot of reports about issues with the latest major comfyui updates (v0.14.x) regarding all my workflows. Please have a look here.
@arkinson Thanks, will check it out.
@BoredJake Just some questions: you mean vram or ram oom??? Wich comfyui version and release do you use? Enhanced models - so you talk about i2v?? Wich aspect ratio has your start image - square? Did you reduced resolution allready?
@arkinson RAM oom, it goes to 99% then crashes. I'm on ComfyUI 0.14.1, I think latest stable. I2V with just a start frame, 16:9. I just tested 480p and that barely works with 97% RAM and 92% VRAM... but higher resolutions give oom. I'm on Linux btw.
@BoredJake Ok, thank you. Comfyui v0.15.0, just some hours ago :-)
The ram issue with server crash is a known problem, see my FAQ here. I can`t speak for Linux, but you might set similar swap file size - min: 64 - 128 gb. Video generation needs much more then 32 gb ram. Your aspect ratio should work with highest resolution.
Btw. actually are a lot of issues with the latest comfyui updates. For myself, I can`t run the Wan workflows with comfyui 0.14.1 on a fresh Comfyui-Easy-Install. Unfortunetely I did not got the error source yet.
@arkinson Thank you, I will try to increase the swap file. :) I am also using Wan2GP where video gen even with my specs is smooth, it's a bummer that Comfy isn't optimized in a similar way. Appreciate your work and support!
When selecting the first frame of a video (PainterI2VAdvanced), comfyui returns an error: Unable to read undefined properties (reading "output").
@Mypm9y Yes, it happens after the latest comfyui updates. Actually the whole workflow is not useable, even with a fresh comfyui installation. I`m still trying to catch the source of this issue.
Please, give me some informations: your OS, comfyui version and comfyui release version. All custom nodes up to date?
Did you get any output in the console window when you press start - or just like in my case, the error message only pops up in comfyui window? Do you get any relating error messages in the logs?
@arkinson win 10, comfyui v0.15.0 (stbility matrix), python 3.10.11, triton 3.6.0, workflow v3.1, all nodes have been updated to the latest version. The error appears only on the top right. There may be a bug in Painteri2v Advanced. I also noticed that the eyes are drawn very poorly in the finished videos, how can I fix this? Thx
@Mypm9y Thank you. There seems to be just something wrong with my logic wich is not excepted after the updates anymore.
As a quick workaround: please completely delete the "Resize + WanFirstLastFrameTo Image" subgraph and don`t activate this option. Let me know if this works for you.
Image quality issues: generally try other start images and prompts at first. And of course, do not expect to much with 12 gb vram. In some of my sample clips you find the same issue with poor drawn eyes.
@arkinson i get 'can't access property "output", res is undefined' assuming this is related?
win 11, comfyui zluda version (2026-02-25 latest), python 3.12.10, triton 3.4.0
@arkinson Oh, thank you! this method worked. My memory is 24 GB, but the problem is different now, I can't make the people in the frame not move, even though I set the motion_amplitude to 1.0. I literally tell them to look at the camera, but by the end of the video, movement begins.
@Mypm9y Thank you for the feedback.
With 24 gb vram and maybe a fast gpu you're playing in a completely different league. You could use my workflow just for fast generations or optimize it to your hardware to get better quality or just look for a workflow that is allready optimized and pre configurated to your hardware. Things in mind: higher models, generation resolution and framerate, generation without lightning Loras, etc... But in my experiance, every video model we actually use (even LTX-2) is limited. And as mentioned, don`t try to force the AI in one direction. If your concept (start image, prompt) do not work well, try somethining completely different.
@b0mbays I`m not sure if your issue is related. Did you tried the workaround by deleting the subgraph?
I actually try to completely redesign the workflow to get rid of the error caused by the comfyui updates. If the workaround do not help, please be patiant. I will try to publish the update soon.
@Mypm9y and @b0mbays I published workflow version 5.0. This should work with latest comfyui version now . Please let me now if you still have any issues.