chore(main): release 0.3.0

[github-actions]

🤖 I have created a release beep boop

0.3.0 (2026-06-15)

Features

• 'The Last Library' — 2.5 min AI short film, 30 scenes, narrated (b528fbb)
• 20-clip benchmark batch — underwater, space, nature, urban, abstract (2c23f4d)
• add --dry-run mode to generate command (script only, no GPU) (e0fa24b)
• add autonomous scheduler with topic generation (1b64f9b)
• add character bank for cross-scene and cross-video consistency (0179c24)
• add core module scaffold (config, schemas, CLI entry point) (41adc18)
• add DeepSeek and Ollama LLM provider support (7fec793)
• add generated video sample (242KB, 5s @ 16fps) (190f812)
• add GPU backend client abstraction (Modal, RunPod, self-hosted) (f228256)
• add GPU backend package scaffold (4320b3b)
• add keyframe generator for scene boundary images (Pass 1) (22aaef3)
• add LLM-powered writer for script and shot list generation (19f9a1a)
• add MisoTTS 8B with one-shot voice cloning support (0f0760e)
• add Modal deployment for Wan 2.2 GPU backend (A100 80GB) (78dc023)
• add parallel FLF2V generator for concurrent scene generation (Pass 2) (1ad810e)
• add pipeline orchestrator (c01affc)
• add pipelined_flf2v mode (overlaps keyframe + video generation) (e7946b0)
• add post-production (TTS, subtitles, FFmpeg assembly) (c91cfee)
• add publisher module (TikTok, YouTube, Instagram stubs) (8bb0d23)
• add quality validation with retry on failed clip generation (bded333)
• add standalone FastAPI GPU backend server (RunPod/self-hosted) (36a465c)
• add video generator with scene chaining (first-frame conditioning) (e6039a0)
• add xDiT multi-GPU sequence parallelism for MI300X (a29da98)
• first benchmark — Wan 2.2 I2V generation on Modal A100 (5a08ea1)
• first benchmark — Wan 2.2 I2V on Modal A100 (6c93a77)
• full benchmark suite — 9 videos, 2 narratives, cost analysis (88a7770)
• implement TikTok upload via Content Posting API (af5dc38)
• implement YouTube Shorts upload with resumable upload protocol (31de3d1)
• integrate two-pass parallel generation into pipeline (generation_mode config) (6feac98)
• narrated benchmark — full pipeline (video + TTS + assembly) (3fcd40e)
• stitch benchmark scenes into full videos (b4d301f)

Documentation

• add CI badge and dry-run usage to README (3b2b796)
• add contributing guidelines (c511ccf)
• add cost projections for video generation at scale (c4fdeff)
• add example configuration file (1c99b03)
• add Google Flow architecture analysis (7c54b68)
• add GPU infrastructure research (MI300X, A100, pricing, throughput) (27417e0)
• add MI300X multi-instance benchmarking plan (9dd9435)
• add publishing & distribution research (TikTok, YouTube, Instagram APIs) (7e2d6a5)
• add system architecture design (b601d30)
• add technology stack decisions (2a96150)
• add video generation models research (Wan 2.2, HunyuanVideo, LTX, CogVideoX) (4acc051)
• rebrand to Flow with OpenX Flow tagline and comparison table (7c3d7f0)
• update all documentation to reflect Flow branding and current features (e0e1368)

---

This PR was generated with Release Please. See documentation.