chore(main): release 0.2.0

[github-actions]

🤖 I have created a release beep boop

0.2.0 (2026-07-22)

Features

• 'The Last Library' — 2.5 min AI short film, 30 scenes, narrated (b528fbb)
• 0.3 agentic video-editing layer (41 MCP tools + kimi agent) (9dc2dcb)
• 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)
• agent: add run_stream() generator for SSE; run() consumes it (16fd291)
• agent: NVIDIA build client (OpenAI-compatible, default kimi=moonshotai/kimi-k2.6) (35fe3ee)
• agent: real GenerationService (Modal video + edge-tts narration) (82f2907)
• agent: token-streaming loop + create_character tool (774edea)
• agent: token-streaming loop + create_character tool (59ec578)
• agent: tool-calling loop (kimi drives the 41 tools end-to-end) (89a621e)
• agent: wire generation tools to real execution (background jobs) (a0e9219)
• api: agent HTTP API — /agent/chat (SSE), /agent/models, /agent/undo (43494f4)
• cli: add 'flow agent' and 'flow mcp' commands (e236baa)
• config: add [agent], [mcp], [billing] config sections (41901f0)
• deploy: pass Modal credentials as args (per-invocation, not ambient env) (#40) (96d1d8a)
• engine: scene-level generation API (generate_clip / generate_keyframe) (a31a061)
• engine: scene-level generation API (generate_clip / generate_keyframe) (a2452ad)
• 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)
• gpu-backend: unified base64 contract + parameterized deployer for named instances (#38) (1dd0b9d)
• gpu: add flf2v + vace endpoints to Modal backend (c2741c9)
• 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)
• mcp: streamable-HTTP MCP server exposing all 41 tools (65d00a5)
• narrated benchmark — full pipeline (video + TTS + assembly) (3fcd40e)
• publish: Instagram Reels + Facebook publishing via Meta Graph API (#42) (0342a08)
• stitch benchmark scenes into full videos (b4d301f)
• store: audio + text Track/TrackItem models (narration, music, captions) (d982ca1)
• store: Clip + Keyframe + Transform models (scene as video-track element) (1daa4b3)
• store: Clip gains ColorGrade + Effect (color/FX support) (fcd3e97)
• store: frame math helpers (frames as authoritative time unit) (0d99ca8)
• store: import_shotlist shim (pipeline ShotList -> timeline Project) (b331e54)
• store: JSON-backed ProjectStore (atomic save/load/list/delete) (431b665)
• store: make ProjectStore DB-backed (replaces JSON-file store) (6ffa121)
• store: MediaAsset + Folder models (library with provenance + reverse index) (9dcc07f)
• store: Project aggregate (video track + tracks + media + cast + undo) (736979a)
• store: SQLModel engine + projects table (SQLite default, Postgres via env) (46f3277)
• store: UndoEntry model (LIFO reversible-edit record) (794be86)
• tools: advanced timeline tools (keyframes, move, insert, ripple-delete, remove_track) (4d0ebe3)
• tools: analysis-read tools from store (get_transcript, inspect_timeline, inspect_media, search_media) (3941179)
• tools: central tool loader (register all tools on one import) (73a128e)
• tools: clip-editing tools (set_clip_properties, split_clip) (acaeaf9)
• tools: color/FX tools (apply_color, apply_effect, inspect_color) (fe5c98e)
• tools: context/read tools (get_project, get_media, list_characters, list_models) (f966b49)
• tools: Flow-native tools (cast, narration, plan_video, clone_voice, start_generation) (a698705)
• tools: generation tools (generate_video/image/audio, upscale, import) (d8d0705)
• tools: media-management tools (folders + library, reference-aware soft delete) (f5ec89b)
• tools: register advanced timeline tools in the loader (7b61b36)
• tools: register analysis-read tools in the loader (75b942b)
• tools: register Flow-native tools in the loader (5743ffc)
• tools: register generation tools in the loader (eb57be2)
• tools: register media-management tools in the loader (fafdf54)
• tools: register text + color tools in the loader (601c09e)
• tools: result envelope + dispatch chokepoint with guardrails (54a842d)
• tools: scene-management tools (create/update/delete/reorder scenes) (c5f3ba9)
• tools: text tools (add_texts titles, add_captions from narration) (9248b35)
• tools: tool registry — @tool decorator + param DSL -> JSON/OpenAI/MCP schema (19837dc)
• tools: undo tool (pop LIFO + restore snapshot) (a69b5bc)
• tts: add pluggable TTSProvider interface (971111e)
• tts: add TTSConfig.miso_endpoint for a dedicated MisoTTS endpoint (a41c56d)
• tts: EdgeTTSProvider (free Microsoft voices, no cloning) (ce44371)
• tts: MisoTTSProvider (voice cloning; local GPU or HTTP endpoint) (77e1614)
• tts: pluggable TTS providers + voice cloning (dfd6964)
• tts: provider registry — get_tts_provider/register_provider/available_providers (92573c5)

Bug Fixes

• db: normalize asyncpg database URLs to sync psycopg dialect (#44) (28a0ec5)
• db: normalize asyncpg database URLs to sync psycopg dialect (#44) (#45) (b4208b0)
• imports: use 'flow' package imports, not 'src.flow' (cbccfb3)
• release: revert untagged 0.2.0 bump back to 0.1.0 (15d5eea)

Documentation

• 0.3: deep spec — color/FX + media-management tools (ffmpeg filtergraph compiler) (021bb3e)
• 0.3: deep spec — context/read tools (get_project, inspect_media, get_transcript, ...) (33bbaba)
• 0.3: deep spec — Flow-native tools (characters, plan_video orchestrator, voice clone) (51e7a9c)
• 0.3: deep spec — runtime (VPS MCP server + nanocode-style agent loop, kimi via NVIDIA build) (3a234bd)
• 0.3: deep spec — text + generation tools (captions, Wan2.2/VACE, TTS/voice) (1a32656)
• 0.3: deep spec — timeline-edit tools (move/split/ripple/keyframes on scenes-as-clips) (f6f866d)
• 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)
• config: document [agent]/[mcp]/[billing] in config.example.toml (83140e8)
• correct roadmap to reflect shipped pipeline (3d1fc2c)
• correct roadmap to reflect what's actually shipped (97620e2)
• readme: add Agentic Editing (0.3) section + roadmap entry (26aac7c)
• rebrand to Flow with OpenX Flow tagline and comparison table (7c3d7f0)
• research: analyze Palmier Pro + agentic video editing landscape (959fbf9)
• research: catalog Palmier's 35 video tools + map to Flow scene tools (8cf51ed)
• research: refine after Palmier code teardown — MCP-on-VPS is the fit (13b8738)
• research: scope 0.3 to ALL 35 tools; scenes ARE the timeline (3c32906)
• research: teardown nanocode's agent tool-loop (ba0cd56)
• tts: document miso_endpoint + provider options in config.example.toml (023a336)
• tts: README section on the pluggable TTS layer + voice cloning + custom providers (fbbf486)
• update all documentation to reflect Flow branding and current features (e0e1368)

Miscellaneous Chores

• release 0.2.0 (b3aff26)

---

This PR was generated with Release Please. See documentation.