Generate 480p talking video in ComfyUI

Step-by-step

1. Drag and drop the 480p 10-second workflow into ComfyUI
2. Load your input image and audio file into the respective nodes
3. Set the width and height parameters to match your input image resolution
4. Update the positive and negative prompts to describe your specific image and desired performance
5. Calculate the required number of frames by multiplying the audio duration in seconds by 25
6. Enter the calculated frame count into the workflow
7. Click Queue Prompt to start the generation

Options

• Enable VAE tiling or Tiled VAE if you have low VRAM
• Increase 'Number of blocks to swap' up to 40 if you are low on GPU
• Use MediaInfo to get the exact millisecond duration of your audio file

Watch out for

• This specific workflow degrades in quality for videos longer than 10 seconds
• The sampler is locked to four steps and cannot be changed
• The model is natively made for 81 frames and uses an embedding system to reach 250 frames
• High RAM is required; at least 100 GB of virtual RAM (page file) is recommended if physical RAM is insufficient

Tips

• Start with the provided test.jpeg and test.mp3 to verify the workflow is working
• Use 'nvitop' in the command line to monitor GPU watt usage; if usage is low, you may be using shared VRAM instead of dedicated VRAM
• If you get out of memory errors, check the CMD terminal window to confirm VRAM issues
• For videos longer than 10 seconds, use a long context generation workflow instead

Highlights

“once you set frame count, your prompt, your input image and its resolution, actually you are ready. You don't need to change anything else.”