How choi403/alg works
This repository is a command-line image-to-video generator implementing Adaptive Low-pass Guidance (ALG) for CogVideoX, HunyuanVideo, and Wan. The main workflow is: load a YAML configuration and input image, select and configure a model pipeline, guide its denoising with scheduled low-pass filtering, then encode and save the generated frames as a video. The shared utilities implement filtering, schedules, and HunyuanVideo sizing. Edges reflect the shown runner and pipeline references; model-family capabilities and configuration roles are also documented in the README.
Main entry points: run.py CLI runner · wan_alg.yaml Generation settings · pipeline_hunyuan_video_image2video_lowpass.py HunyuanVideo pipeline · pipeline_wan_image2video_lowpass.py Wan pipeline
Components 15
Generation runner
- CLI runner
run.py
ALG guidance
- Low-pass filters
lp_utils.py - Strength schedules
lp_utils.py - Hunyuan sizing
lp_utils.py
I2V pipelines
- Wan pipeline
pipeline_wan_image2video_lowpass.py - CogVideoX pipeline
pipeline_cogvideox_image2video_lowpass.py - HunyuanVideo pipeline
pipeline_hunyuan_video_image2video_lowpass.py
Model configuration
- Generation settings
configs/wan_alg.yaml - Wan configs
configs/wan_default.yaml - CogVideoX configs
configs/cogvideox_default.yaml - Hunyuan configs
configs/hunyuan_video_default.yaml
Other components
- User
- Input image
- Pretrained models
- Output video
Connections 21
- User → CLI runner: invokes
readme.md - User → Input image: provides
readme.md - User → Generation settings: selects
readme.md - CLI runner → Generation settings: reads
run.py - CLI runner → Input image: loads
run.py - CLI runner → Wan pipeline: selects and calls
run.py - CLI runner → CogVideoX pipeline: selects and calls
run.py - CLI runner → HunyuanVideo pipeline: selects and calls
run.py - CLI runner → Hunyuan sizing: gets dimensions
run.py - CLI runner → Output video: writes frames
run.py - Wan pipeline → Low-pass filters: uses filtering
pipeline_wan_image2video_lowpass.py - Wan pipeline → Strength schedules: uses schedules
pipeline_wan_image2video_lowpass.py - CogVideoX pipeline → Low-pass filters: uses filtering
pipeline_cogvideox_image2video_lowpass.py - CogVideoX pipeline → Strength schedules: uses schedules
pipeline_cogvideox_image2video_lowpass.py - HunyuanVideo pipeline → Low-pass filters: uses filtering
pipeline_hunyuan_video_image2video_lowpass.py - HunyuanVideo pipeline → Strength schedules: uses schedules
pipeline_hunyuan_video_image2video_lowpass.py - CLI runner → Pretrained models: loads weights
run.py - Wan configs → Generation settings: configures
readme.md - CogVideoX configs → Generation settings: configures
readme.md - Hunyuan configs → Generation settings: configures
readme.md - HunyuanVideo pipeline → Hunyuan sizing: uses sizing
pipeline_hunyuan_video_image2video_lowpass.py