Hunyuan Video
Kijai marked files only for use with Kijai Nodes You do not need them for Comfy Native
Full Guide to picking the correct file above
Workflow for 8GB Card users
Uncensored llama will work with COMFY Native
Using the Kajai marked models on COMFY native will cause rainbow or black output.
I do not recommend the the FP8 VAE unless you are trying to fit all models into GPU, see the guide for 4090 full GPU launch commands.
Technical details regarding "Uncensored"
The model used for Hunyuan was based on llava-llama-3 8 billion parameter LLM. The Intel vision tuned model was used to refine the tokenized model restoring over 5 million values.
Description
FAQ
Comments (57)
各位，秋叶sd包可以使用这个吗，我现在还没显卡，先下模型，下了6T模型 ，几千个LORA，都是图片的
不行的孩子，只能用于视频的lora
Hello, i see that xtuner-llava-llama-3-BF16 file just uploaded. what is this used for?
It can be used in the dual clip loader along with clip-l
Anyone get this working on a 4060 8gb potatoe or similar card?
Yes, I'm using a 3070Ti 8GB laptop GPU and 64GB of RAM, running it with native nodes in ComfyUI. I also start ComfyUI with --lowvram and --disable-smart-memory arguments to help offloading on RAM. Hope that helps!
@LG89 Thanks for that!
@LG89 how long did it take to generate a 5s-video?
@liwuran84812 Since I've installed sage attention in my python environment, it takes about 10-12 minutes for 20 steps (~34s/it) 640x480 121 frames (without sage attention it's about 20 minutes)
@LG89 thanks bro, maybe I should replace my 2060s firstly, or it will take too much time ,lol.
Am I wrong or was there a 16GB file hunyuanVideoSafetensors_llavaLlamaTE.safetensors before ? Is it obsolete and to be deleted on my drive ? Thank you
It was a mixed precision TE based on autocasting but only 4 blocks ended up converting to FP8, I thought that might have caused problems for some, so I added the BF16 and FP8 models separately.
Can I use this model for sd-webui?
it's not on their open source plan but I'm sure somebody else will eventually integrate it
【EmptyHunyuanLatentVideo】 This node is no longer updated and cannot be downloaded.
I had the same problem, regular update comfyui button in manager didn't fix it. I had to update my comfyui from the update.bat in update folder in comfyui. I had several other problems after that but figured them out one by one
I am going to try it out on Forge, hopefully it works...🫤
maybe a stupid question but i've been trying to get to run in Forge. If you get it to work could you please help me because it seems that i'm doing something wrong
@Poposan Haven't gotten it to work yet, everyone seems to be using ComfyUI 🤷
@ultratenz10362 Replying incase there's an update
I'm using llava_llama3_fp8_scaled. Is there any benefit to using the one you posted here? thanks
Ideally for comfy a FP32 version would be available, I would use the version that fits in your system ram with no issue
llava_llama3_fp8_scaled Where did you download it?
I get only noise, tried so many workflows and it takes 5 hours on my 12gb 3060 with 64gb RAM. using llava_llama_fp8_scaled as above. any ideas??
唉，我显卡太垃了，生成 848*480 20fps 3s 要2小时
I have black output all the time,anyone meet encounter this issue before?
I think that is do to the FP32 VAE running out of memory.
I get black output as well. 12GB VRAM 3060
@TecKnight chech if you have the correct clips loaded
@K3NK Thank you,problem solved!
@K3NK which are the correct clips?
someone help. all my gens are greyed or blacked out or nonsense. what are the right clips. im using the llama and clip-l and it's not working
llava_llama3_fp8_scaled Where did you download it?
找到giehub的comfy官网，找到混元视频，点进去有指引
Hey, this was what I found when I googled it:
does anyone have a workflow model that I can download for 8gb card which works with loras and can take an image as baseline?
我发现我使用官方的fp8模型能正常使用，但你的这个fp8模型跑出来的却是纯噪声图。
我想问之后会解决这个问题吗？
我跑出来也是噪声
一样的问题
@Felldude 哦哦
any workflows for this?
Below is the link.
to those that don't know, you can download any of these videos and just drag them into comfyui and you'll get the workflow
@tomazxzas143 How?? that dosn not work for me
@rule34user you just drag the downloaded video file anywhere onto comyUI's interface, try a different video if it doesn't work
@tomazxzas143 I've seen people make this comment many times. I've seen it work with webp (animated image) but never with an mp4 (video container). Do you have an example of an mp4 that works OR is there something I'm missing about converting an mp4 before drag and drop? Thank you!
@sarashinai same question here! never works for me with .mp4
@sarashinai @ermbaby no idea tbh lads, I'm no expert at this, chances are that you might be missing some sort of custom node maybe, all I know is that 90% of these vids do actually work for me, when I just right click and "save as" and then just drag them into comfyui
@tomazxzas143 Fair enough. Could you tell us one of the vids you're SURE works and we'll experiment on that.
@sarashinai https://civitai.com/images/55477992 this one works. What I've noticed is that it basically just depends on how the user uploaded the vid file, if it's named something other than vid_*number* or video_*number* it usually won't work
hello guys, is this possible to use this on your own computer and how please?
With the Comfy Native nodes I am able to use this on a 8GB video card
@Felldude how?? Pls
@Marvelx You can use the Comfy native nodes to load the FP8 or BF16 model, if you have enough system ram you can run the model at 360x240 - I do not recommend the full size video