Create Realistic Talking AI Avatar Videos | InfiniteTalk ComfyUI Tutorial
Info Page: https://github.com/MeiGen-AI/InfiniteTalk/tree/main
Tutorial: https://youtu.be/2yeo3D76a4s
1οΈβ£Step 1: Install ComfyUI
π Download: https://www.comfy.org/
π Follow the installation steps for your operating system. Once installed, launch ComfyUI.
*or you can install portable version if you want to unlock more customizability: https://youtu.be/E_D7y0YjE88
2οΈβ£ Step 2: Get Workflow
πGREAT News!!
You can run the workflow online with ZERO setupπ https://bit.ly/4iKAprU
OR
Get the Workflow file attached below
3οΈβ£ Step 3: Download Required Models
Wa2.1 models
π Place in: ComfyUI\models\diffusion_models\WanVideo
GGUF 480p Q8:πhttps://huggingface.co/city96/Wan2.1-I2V-14B-480P-gguf/blob/main/wan2.1-i2v-14b-480p-Q8_0.gguf
InfiniteTalk models
π Place in: ComfyUI\models\diffusion_models\WanVideo\InfiniteTalk
GGUF 480p Q8:πhttps://huggingface.co/Kijai/WanVideo_comfy_GGUF/blob/main/InfiniteTalk/Wan2_1-InfiniteTalk_Single_Q8.gguf
VAE
π Place in: ComfyUI\models\vae
π https://huggingface.co/Kijai/WanVideo_comfy/blob/main/Wan2_1_VAE_bf16.safetensors
Text Encoder
π Place in: ComfyUI\models\text_encoders
πhttps://huggingface.co/Kijai/WanVideo_comfy/blob/main/umt5-xxl-enc-bf16.safetensors
Clip Vision
π Place in: ComfyUI\models\clip_vision
πhttps://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/blob/main/split_files/clip_vision/clip_vision_h.safetensors
Vocal separator
π Place in: ComfyUI/models/diffusion_models/MelBandRoformer
πhttps://huggingface.co/Kijai/MelBandRoFormer_comfy/blob/main/MelBandRoformer_fp16.safetensors
Lightning Lora
π Place in: ComfyUI\models\loras
πhttps://huggingface.co/Kijai/WanVideo_comfy/blob/main/Lightx2v/lightx2v_I2V_14B_480p_cfg_step_distill_rank64_bf16.safetensors
4οΈβ£ Final Step
Load the WAN 2.2 Animate workflow
Install missing nodes
Restart ComfyUI
Edit + run the workflow
Comments (56)
I tried, it's pretty good, but why head movement is too much? any config we can change to move head stop moving around?
It's been so long since I used this but if am not mistaken there's a strength/exaggeration setting that i went through if the video, you can also push for "calm, natural" style in the prompt just using keywordsnto keep it from moving too much
Which graphics card is optimal for this?
I recommend atleast an RTX with 24GB VRAM, higher is better
safetensors._safetensors_rust.SafetensorError: Error while deserializing header: header too large This error is coming while running the video
i have apple studio can i make 1080p with that i like his natural movemnet
Hi Rayan, unfortunately I won't be able to tell you since I don't have that, why don't you try ? It's free
Crazy, I know... but has anyone tried this on a mac?
hmm, depends on which generation u have, but I assume it's gonna be painfully slow :/
I try this workflow work good for me but way of your video look nature my video not look nature how can I fix it ?
what did you use for image generation ?
I got an error it said: KeyError: 'blocks.0.norm_x.weight what do I have to change?
bro I was trying to make it all my own, you're a life saver lol
happy to help
Where to Upload the json on file ?
you drag and drop it into a empty comfy workflow
I just updated the workflow and model locations to solve the "I run but nothing happens" issue
Thanks for the awesome video and detailed guide, I was able to install comfy comfortably, but I kept getting this feedback: torch.OutOfMemoryError: Allocation on device This error means you ran out of memory on your GPU. I used a HP Zbook Power G7
yeah this means the workflow requires more VRAM than your GPU is able to offer
I can't see the ouptut video
hey, check the terminal for any errors, and try different inputs
I've 3060 with 12 GB Vram and 16GB PC RAM. and i've generated 30 sec video. it's already 16 hours killed but still running. why it's happaning and what should do now? and i've followed your settings 480p.
the more VRAM the better, and 16GB RAM is way too low for this
Can you make a video on LongCat Avatar video with wan2.2
I will look into it
I can't see the ouptut video when execution ends.
check the terminal for error logs to see what's happening
You are the BEST!!!
once you setup the workflow. how long it take you the generate a lip sync video of 20 secinds? cuz i tried one and it took me almost 2 hours even though i have an rtx 5060 ti 16 gb
I shared the generation times in the tutorial, I hope that works as a good reference, but how long it takes for you will depend on so many factors
Thanks bro, Iβve successfully installed it on my PC. Iβm using the model wan2.1-i2v-14b-480p-Q8_0.gguf β once again, thank you.
Nice! glad it worked
Hi, I have this error when i want to run the workflow : "Cannot read properties of null (reading 'replace')". What could be the problem?
sorry, I did not encounter this one before
hi, I tried installing the missing nodes from the manager, and it did, but I still have those nodes as missing, any idea why that might be the case? Thanks in advance
can u tell which nodes are those ? and is yoru comfyui up to date ?
Is it useful on mac os m4 chips
a little tricky, better to run on NVIDIA, AMD graphics, or online
what system requirements are required?
is possible to create different scenes into the same video without need different videos? to make more entertaining videos
not sure what you mean, you can only upload an image and turn it into one single video
Any 4K or 1080 versions?
you can generate 4k with this, if you have the right hardware
have you tried wan AIO model?
brother, I don't know how to add the downloaded files on Mac from hugging face any ideas?
I haven't used ComfyUI on Mac myself, but it should have a models folder just like on windows, did you look for that ?
When I run the workflow, the MelBandRoFormerSampler shows the error "string indices must be integers not 'str'", but I ran this workflow perfectly well at the same day. Plz help
weird, not sure what that error means, did you try googling it ? or maybe use a different audio track
i have a problem with a cuda, how can i fix that.?
what's the exact problem ?
Where is this WAN 2.2 Animate workflow?
Sorry, found it. Didn't realize you were referring to the workflow you provided.
Hi MDMZ, I got the error "No patch_embedding weight found, is the selected model a full WanVideo model?" can you help?
Same issue. Could we get a response MDMZ?
Thanks!
u r welcome