Foton's Wan 2.2 Image2Video workflow
Designed for 12GB or lower VRAM.
Can be modified to work with more demanding models, but there's a lot of workflows for that already.
Features
Easy generation config
Lora Support
Advanced Cropping
Built-in Upscaling and Interpolation
Add audio track (not audio sync)
v2.0 update
Tested with ComfyUI 0.43.0
Added new nodes for cropping input images. (Be aware that If you create multiple crops it will generate multiple videos in sequence). Examples:
The downside of a new crop tool is that it doesn't save crop settings inside the workflow. As a workaround I added a Save Image node for cropped outputs if you want to use them later.
Removed deprecated (outdated) nodes.
Included Seed VR2 upscaler as an alternative.
(Advanced/experimentall) Added a Torch Compile switch that is turned off by default. In short, it optimizes generation speed for consecutive videos by 15-20%, but your first generation will be much slower. When you change model/loras, output resolution of number of frames it will recompile again.
Easy Generation Config
Uses Lightning Lora by default to speed up generation considerably (3 min vs 15+mins on my 4070) at the cost of motion dynamics. Can be disabled with 1 click but will require 20-30 steps.
Lora Support
Support up to 2 Loras, you can add more manually with higher VRAM.
Advanced Cropping
Lets you crop images to fit landscape image into portrait and vice versa. Use offset to adjust crop center. Example:
Built-in Upscaling and Interpolation
Uses RIFE interpolation to generate frames. Use rife49.pth or rife47.pth for best results. You can replace Remacri upscaler with any other ESRGAN model like UltraSharp, R-ESRGAN4x, etc.
Blend_factor does some initial color correction for Remacri - it tends to desaturate images. Set it to 0 if you're using another upscale model and the result is too saturated.
Save Last Frames
Saves a number of last frames if you want to use them to generate a new video and then stitch them together. (Sometimes last frame has motion blur, so I added an option to save as many as you want).
Add audio
Audio is not syncronised, this requires another workflow entirely with SI2V models.
