- Format endpoint now calls ACE-Step /format_input REST API directly instead of
spawning Python, fixing ENOENT errors on Windows (#44, #27, #34)
- isGradioAvailable() tries /gradio_api/info, /info, / in sequence to handle
Gradio 4.x/5.x/6.x version differences, fixing generation fallback (#53, #20)
- Storage getUrl/getPublicUrl normalize /audio/ prefix to prevent double-prefix
URLs when reference tracks are used for cover generation (#10)
- Gradio args: fix normalization_db default from 0.0 to -1.0 (Gradio default)
- Volume popover: add 400ms delay before hiding to prevent accidental dismissal (#51)
- Polling: skip setSongs state update when nothing changed to reduce re-renders (#51)
- FFmpeg: add jsdelivr CDN fallback when unpkg fails (#30)
- Model switching: add switchModelIfNeeded() via /v1/init REST API (#45)
When enabled, uses the LLM to enrich genre/style tags into detailed
music descriptions and generate proper BPM, key, and time signature
metadata (CoT features). Fixes genre tags like "pop, rock" producing
ballad-like output by matching Gradio UI default behavior.
- Add enhance toggle in Style of Music card header
- Gate CoT metas/caption/language by enhance OR thinking flag
- Remove unsupported --lm-backend/--lm-model from Python fallback
- Add i18n translations (en/zh/ja/ko) with tooltip
- Migrate backend from REST API to Gradio @gradio/client for generation
- Fix Gradio parameter alignment (positions 36-49) for reference/cover audio
- Add LoRA training pipeline with dataset upload, preprocessing, and export
- Add News page with dismiss/restore and GitHub star button
- Add localization info icon in Settings language section
- Fix upload audio URL prefix, add missing MIME types
- Add training API routes and Python preprocess script
- Update i18n with news keys for all languages
- Add lmModel parameter through full chain (types, API, routes, service)
- Add LM Model dropdown in Advanced Settings (defaults to 0.6B)
- Pass lmModel and lmBackend to format/enhance endpoint
- Update format_sample.py to accept --lm-model and --lm-backend args
- Auto-download model from HuggingFace if not present locally
- Persist model selection in localStorage
- Improve format route error logging with exit code and stdout/stderr
Fixes#9
- Add Windows Portable Package support with auto-detection
- Fix API health check to support multiple response formats
- Fix task ID parsing for ACE-Step API integration
- Change thinking mode default to false for 4GB GPU compatibility
- Add explicit CoT parameter control for LLM features
- Fix audio download path encoding for portable package
- Improve file upload validation with extension fallback
- Add start-all.bat/sh scripts for one-click startup
- Update README with Windows Portable Package instructions
- Add comprehensive troubleshooting section
- Improve cross-platform compatibility
- Check if ACE-Step API is running at startup (GET /health)
- If API available: use REST endpoints (POST /release_task, POST /query_result, GET /v1/audio)
- Model is already loaded by the API server, no double initialization
- If API not available: fall back to Python spawn (existing behavior)
- Model loads once via Python script
This prevents the issue where running `acestep-api --port 8001` loads the model,
then the UI spawns Python which loads it again, consuming double VRAM.
API response is cached for the session to avoid repeated health checks.
- Add resolvePythonPath() function that handles:
- PYTHON_PATH env var override (highest priority)
- Portable installations (python_embeded/python.exe)
- Standard venv with correct paths per OS:
- Windows: .venv/Scripts/python.exe
- Unix: .venv/bin/python
- Remove hardcoded Unix paths from acestep.ts and generate.ts
- Remove hardcoded /home/ambsd default path
Fixes "ENOENT" errors on Windows when spawning Python processes.
fspecii