Prepare VoiceForge for open-source launch

[mohitkale]

Summary

• Product positioning: README leads with “Clone a voice locally. Generate speech privately,” multi-engine Studio/API value, self-hosted (not overstated “100% laptop-only”) wording, and early-public-release status instead of M0–M8 milestone language.
• Responsible use & consent: Added RESPONSIBLE_USE.md, docs/CONSENT.md, Studio consent checkbox (API still requires consent=true), honest “declaration ≠ verification” language.
• Engine readiness & licensing: docs/ENGINES.md + docs/ENGINE_LICENSING.md with categorized engines and licence separation (MIT app ≠ model terms).
• Onboarding & ops: docs/FIRST_CLONE.md, docs/SELF_HOSTING.md, docs/REEL_STUDIO.md, Makefile helpers (start-cpu / stop / logs / smoke-openvoice).
• Watermarking: docs/WATERMARKING.md — optional weak fingerprint, not forensic proof.
• Community: ROADMAP, CHANGELOG, CoC, CONTRIBUTING, issue/PR templates (block private audio/tokens/weights), SECURITY updates.
• Assets: Authentic Studio screenshot + engine-comparison SVG; audio demos intentionally not fabricated (engines not ready in lightweight local venv) — see docs/DEMO_CAPTURE.md / docs/assets/AUDIO_SOURCES.md.
• CI fix: Ruff S110 / import-order failures that were failing lint-and-test on main.
• Studio UX (safe): Token-optional engine loading when API has no token; default highlight openvoice-v2.

README before → after

• Before: Engine/milestone/architecture-heavy top; “100% local-first”; consent buried; no Studio screenshot near top.
• After: Product promise → screenshot → responsible use → engine categories/matrix → quick start / first clone → API / Reel Studio / security → architecture last.

Authentic screenshots and audio assets

Asset	Status
docs/assets/voiceforge-studio.png	Captured from running Studio
docs/assets/engine-comparison.svg	Illustrative (not a quality ranking)
docs/assets/clone-flow.gif	Not yet recorded — capture steps in DEMO_CAPTURE.md; existing docs/clone-flow.svg linked
docs/assets/audio/*.mp3	Not committed — would require Docker+models; do not fabricate

Engine support matrix

See docs/ENGINES.md. CPU e2e verified claims retained only for openvoice-v2, xtts-v2, f5-tts. GPU still marked verification needed.

Responsible-use and consent changes

Docs + Studio checkbox. No identity-verification infrastructure added. Optional consent metadata fields documented as future/not implemented.

Licensing review

docs/ENGINE_LICENSING.md + README/NOTICE pointers. XTTS CPML non-commercial preserved.

Security review

No telemetry/cloud TTS added. Auth/CORS/resource limits unchanged. SECURITY.md expanded for voice-data incidents. Studio still works with bearer when configured.

Tests executed

• ruff check app/ scripts/ tests/ — pass
• pytest tests/ -q — 49 passed
• docker compose -f docker/docker-compose.yml config — OK
• pip-audit — not re-run in this pass (available via pip install -e ".[dev]")

Engines not tested / why

Engine	Why no new demo audio here
openvoice / f5 / xtts	Local uvicorn .venv had 0/9 engines ready (ML stacks not installed); Docker model run not executed in this PR to avoid long downloads
qwen3 / chatterbox / rvc / fish / cosyvoice / indextts	Same + GPU/sidecar/worker requirements

Remaining manual GitHub settings

Follow docs/GITHUB_METADATA.md: description, topics, social preview image/text.

Test plan

• CI green on this PR
• Open Studio after make start-cpu; confirm consent checkbox blocks clone when unchecked
• Localhost without token still lists engines
• With token set, invalid token still 401s
• Optional: capture audio demos per DEMO_CAPTURE.md and follow-up commit
• Set GitHub About / topics / social preview manually

Do not merge until you are satisfied with docs + CI.

Made with Cursor