How deepbeepmeep/wan2gp works
WanGP is a browser-based generative-media application spanning video, image, audio, and speech, with local inference and API/headless use. The main user flow is to prepare a prompt and optional media, submit a generation, and manage the resulting media in the UI; the README also documents queues, reusable outputs, prompt enhancement, and input-preparation tools. Model pipelines perform inference, while shared generation-progress and media utilities support runtime work. Several model families are represented by directories because their implementation was not sampled; those connections are not asserted without evidence.
Main entry points: app.js Web interface · gradio_engine_patch.py Generation queue · models Video model pipelines · TTS Audio and TTS models
Components 20
Creation experience
- Web interface
shared/deepy/web/app.js - Media Flow tools
plugins/media_flow/form_controller.py - Prompt enhancement
models/TTS/prompt_enhancers.py - Generation queue
shared/gradio/gradio_engine_patch.py
Video generation
- Video model pipelines
models - Wan audio-video
models/wan/ovi_fusion_engine.py - MiniMax H3
models/minimax_h3/pipeline.py - LTX Video
models/ltx_video/pipelines/pipeline_ltx_video.py
Image generation
- Image model pipelines
models - Qwen Image
models/qwen/pipeline_qwenimage.py - Z-Image
models/z_image/pipeline_z_image.py
Audio generation
- Audio and TTS models
models/TTS - ACE-Step music
models/TTS/ace_step/pipeline_ace_step.py - ACE-Step 1.5
models/TTS/ace_step15/pipeline_ace_step15.py - Speech synthesis
models/TTS/qwen3/pipeline.py
Shared runtime
- Generation progress
shared/utils/phase_progress.py - Text encoder cache
shared/utils/text_encoder_cache.py - Checkpoint location
shared/utils/files_locator.py - LoRA settings
shared/utils/loras_mutipliers.py
Other components
- Creator
Connections 21
- Creator → Web interface: creates with
README.md - Web interface → Generation queue: queues jobs
README.md - Web interface → Media Flow tools: uses tools
README.md - Web interface → Prompt enhancement: enhances prompts
README.md - Generation queue → Video model pipelines: dispatches generation
- Generation queue → Image model pipelines: dispatches generation
- Generation queue → Audio and TTS models: dispatches generation
- Video model pipelines → Wan audio-video: includes
models/wan/ovi_fusion_engine.py - Video model pipelines → MiniMax H3: includes
models/minimax_h3/pipeline.py - Video model pipelines → LTX Video: includes
models/ltx_video/pipelines/pipeline_ltx_video.py - Image model pipelines → Qwen Image: includes
models/qwen/pipeline_qwenimage.py - Image model pipelines → Z-Image: includes
models/z_image/pipeline_z_image.py - Audio and TTS models → ACE-Step music: includes
models/TTS/ace_step/pipeline_ace_step.py - Audio and TTS models → ACE-Step 1.5: includes
models/TTS/ace_step15/pipeline_ace_step15.py - Audio and TTS models → Speech synthesis: includes
- ACE-Step 1.5 → Text encoder cache: uses cache
models/TTS/ace_step15/pipeline_ace_step15.py - ACE-Step 1.5 → Checkpoint location: locates files
- ACE-Step music → LoRA settings: uses settings
models/TTS/ace_step/pipeline_ace_step.py - MiniMax H3 → Generation progress: reports progress
models/minimax_h3/pipeline.py - Qwen Image → Generation progress: reports progress
models/qwen/pipeline_qwenimage.py - Z-Image → Text encoder cache: uses cache
models/z_image/pipeline_z_image.py