Spaces:
Sleeping
Sleeping
Commit History
-
feat: Portuguese story language, English scene field for image prompts, README pitch
ClaudeUnverified signature- engine.py: StoryState.language, system_prompt(language); model now returns a short English 'scene' field per beat which leads the image prompt — CLIP understands English best, so illustrations stay sharp even for PT stories - tts.py: per-language Kokoro voices (en: af_heart, pt: pf_dora) and edge-tts fallback voices (pt-PT-RaquelNeural); modal TTSModel caches one KPipeline per lang_code — redeploy with 'modal deploy modal_inference.py' - stt.py / modal STTModel: whisper language parameter; local fallback picks multilingual whisper-tiny for non-English - app.py: Story language dropdown in setup, threaded through prompts, TTS and voice input - README: rewritten as the Space card / hackathon pitch — tiny-stack story, how-it-works, Build Small framing, TODO placeholders for demo links/GIF https://claude.ai/code/session_0166UGtjNpECJBfRzLotRwBQ46f845b -
refactor: clean dead files, canonical module names, visible error recovery
ClaudeUnverified signature- delete app_v1.py, app_v2.py, old engine.py/styles.css, empty packages.txt, _obsidian/ notes; move STORYFORGE_SPEC.md to docs/ - rename engine_v2.py -> engine.py, image_model_v2.py -> image_model.py, styles_v2.css -> styles.css; update imports, README table and headers - app.py: generation failures now show a friendly .status-error banner and restore the previous screen (setup or current beat + options) so the user can simply retry; image/PDF failures are logged instead of swallowed - theme-card JS clears the bus value before setting it, so re-picking the same theme after an error still fires .change - CLAUDE.md: full rewrite matching the real architecture (Modal backends, streaming, 19-element tuple, CSS cascade rules) https://claude.ai/code/session_0166UGtjNpECJBfRzLotRwBQe83a7fe
-
feat: initial release — StoryForge branching storybook app
Nelson AndréGradio app for children (ages 4-8) with local GGUF inference via llama-cpp-python. No cloud APIs. HuggingFace Build Small hackathon entry. - engine.py: StoryState, prompt builder, JSON parser with Qwen3 think-block stripping - model.py: llama-cpp-python wrapper with lru_cache model load - app.py: Gradio UI, 14-element index-stable output tuple, theme card message bus - styles.css: custom storybook theme, beats Gradio StreamingBar CSS via end-of-file rules4fdb3ad