perf: cut Modal idle spend — 120s scaledown, IP-Adapter cached per container, TTS on 2 CPUs

Merge branch 'claude/huggingface-hackathon-improvements-oem6d9'

feat: Portuguese story language, English scene field for image prompts, README pitch

refactor: clean dead files, canonical module names, visible error recovery

feat: Kokoro-82M TTS on Modal, keyboard-accessible theme cards, focus styles

Merge pull request #2 from nelsonandreproton/claude/huggingface-hackathon-improvements-oem6d9

feat: stream beat text live, image shimmer, sound toggle, mobile CSS, shorter default story

docs: update README — fix fallback note, Python 3.13 + Gradio 4.44 deploy reality

feat: scene-first image prompts, negative prompt for no borders/boxes, higher guidance

feat: meaningful PDF filename from theme+hero instead of random tempfile name

fix: unload IP-Adapter on container start, load per-call only when reference given

fix: load IP-Adapter lazily only when reference image exists; drop ambient volume to 0.12

fix: lower ambient volume 0.72→0.25; log image Modal failures, return None instead of crashing local fallback

fix: pass dummy image to IP-Adapter when no reference — None causes TypeError in diffusers unet

debug: log Modal failure reason before fallback

fix: pin starlette<1.0 and jinja2<3.2 — newer versions break gradio 4.44.0 TemplateResponse

fix: patch Blocks.get_api_info to swallow APIInfoParseError — recursive fn not patchable via module attr

fix: patch _json_schema_to_python_type instead of get_type — bool additionalProperties crashes gradio-client 1.3.0

fix: remove gradio-client pin (conflicts with gradio 4.44); patch bool-schema bug inline

fix: server_name=0.0.0.0 for HF Spaces proxy; bump gradio-client for Python 3.13 bool schema bug

fix: move css= to gr.Blocks() — launch() never accepted css kwarg

fix: pin huggingface_hub<1.0 — HfFolder removed in 1.x, gradio 4.44 oauth needs it

fix: replace pyaudioop with audioop-lts — correct PyPI shim for Python 3.13

fix: add pyaudioop shim — audioop removed in Python 3.13, pydub needs it

fix: bump gradio to 5.x — 4.x pydub/audioop broken on Python 3.13

fix: move extra-index-url to top of requirements so pip resolves pre-built wheel

fix: use pre-built llama-cpp-python wheel to fix OOMKilled HF Space build

Add Obsidian second brain notes for StoryForge project

Update README: sdk_version quoted, add upload + Python version notes

Fix Gradio 6.0 css deprecation and ambient audio int16 format

Push local master to HF Space main branch

Fix HF Space repo ID to nelsondiasandre/StoryForge

Switch to git push for HF Space deployment

Fix upload_large_folder — use login() instead of token param

Empty packages.txt — HF Spaces base image already includes ffmpeg

Switch to upload_large_folder with tighter ignore list

Remove create_repo call — Space must exist first

Add upload_space.py helper script

Fix HF Space colorFrom to valid value (yellow)

Merge pull request #1 from nelsonandreproton/claude/hugging-face-credits-plan-PzCSQ

Add Modal GPU backend + 6 new features

fix: defer torch/diffusers imports to avoid OOMKilled on HF free tier

chore: promote app_v2.py to app.py for HF Spaces compatibility

feat: add phase 2 — per-beat Ghibli-Diffusion illustrations

feat: initial release — StoryForge branching storybook app