ComfyUI: High-quality 720p long context generation

Step-by-step

1. Drag and drop the 10-step high-quality workflow file into ComfyUI
2. Upload your source media using the 'Choose file to upload' button
3. Set the frame count to match your source (e.g., 250 frames)
4. Enter your text prompt in the prompt node
5. Adjust the resolution to 1280x720 for HD output
6. Select the '14 billion parameters image to video 720p GGUF Q8' model
7. Set the block swap parameter based on your available VRAM
8. Click Run to start the generation

Options

• Toggle 'Enabled VAE tiling' or 'Tiled VAE' to prevent memory errors
• Switch between Sage Attention, Flash Attention 2, or SDPA
• Use a low VRAM configuration (8GB or 15GB) if needed
• Adjust block swap count (e.g., 20 for speed, 40 for lower VRAM usage)

Watch out for

• Flash Attention 3 and Flex Attention are not supported by the current auto-installer
• 720p generation takes significantly longer than 480p workflows
• High-quality settings can use up to 28 GB of VRAM
• Quantization settings are disabled when using GGUF models

Tips

• Use Sage Attention for the best balance of speed and performance
• Reduce the number of block swaps to speed up generation if you have sufficient VRAM
• Monitor the CMD window to see how much of the model is offloaded to the CPU versus the GPU
• Ensure the frame count is extremely precise as this workflow is sensitive to that value

Highlights

“With 10 steps version, you don't need to set anything, everything is set.”