How aigc-apps/videox-fun works
VideoX-Fun is a multi-model image/video generation and model-training toolkit, exposed through interactive UIs, Python/API interfaces, and ComfyUI nodes. The generation workflow accepts text and optional image, video, audio, or control inputs; model-specific pipelines run diffusion models and return generated media. Training and dataset preparation, reward/caption utilities, and ComfyUI control annotators are distinct supporting capabilities. The README documents capabilities broadly; source excerpts are partial, so some cross-layer links are intentionally omitted or marked unverified.
Main entry points: pipeline Generation Pipelines · nodes.py Model Nodes · controller.py UI Controller · ui Web UIs
Components 20
User Interfaces
- Web UIs
videox_fun/ui - Inference API
videox_fun/api - ComfyUI Nodes
comfyui/comfyui_nodes.py - Model Nodes
comfyui/wan2_1/nodes.py - UI Controller
videox_fun/ui/controller.py
Generation Runtime
- Generation Pipelines
videox_fun/pipeline - Diffusion Models
videox_fun/models - Flow Samplers
videox_fun/utils/fm_solvers.py - Memory And LoRA
videox_fun/utils/lora_utils.py - Media Utilities
videox_fun/utils/utils.py
Training Data
- Training Datasets
videox_fun/data/dataset_video.py - Bucket Sampling
videox_fun/data/bucket_sampler.py - Training Scripts
scripts
Controls And Evaluation
- Caption Processing
videox_fun/video_caption - Control Annotators
comfyui/annotator/nodes.py - Reward Evaluation
videox_fun/reward/reward_fn.py - ComfyUI Helpers
comfyui/comfyui_utils.py
Other components
- Creator
- Generated Media
- Model Weights
Connections 21
- Creator → Web UIs: selects generation
README.md - Creator → Inference API: requests inference
README.md - Creator → ComfyUI Nodes: builds workflow
README.md - Web UIs → UI Controller: delegates requests
- Inference API → Generation Pipelines: runs inference
- ComfyUI Nodes → Model Nodes: registers nodes
comfyui/comfyui_nodes.py - Model Nodes → Generation Pipelines: runs generation
comfyui/wan2_1/nodes.py - Model Nodes → ComfyUI Helpers: resolves models
comfyui/wan2_1/nodes.py - Generation Pipelines → Diffusion Models: invokes models
videox_fun/pipeline/pipeline_minimax_h3.py - Generation Pipelines → Flow Samplers: uses solvers
videox_fun/pipeline/pipeline_wan_flex_forcing.py - Generation Pipelines → Memory And LoRA: applies adapters
- UI Controller → Generation Pipelines: selects pipeline
- UI Controller → Model Weights: loads weights
videox_fun/ui/controller.py - Generation Pipelines → Media Utilities: writes media
- Media Utilities → Generated Media: encodes output
- Creator → Training Scripts: trains models
README.md - Training Scripts → Training Datasets: consumes datasets
- Training Datasets → Bucket Sampling: uses bucketing
videox_fun/data/__init__.py - Creator → Caption Processing: prepares captions
README.md - Creator → Control Annotators: creates controls
README.md - Generation Pipelines → Reward Evaluation: scores outputs