coco-research/Mira

Latest recorded source activity: October 8, 2026 at 19:25

SkillsMP catalog refreshed: October 29, 2026 at 18:30

skills collected: 17

GitHub stars: 6
GitHub forks: 0

Install with an AI assistant

Copy this prompt into the AI assistant you're using.

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/coco-research/Mira.

Repository README

Repository README source: README.md

Read the source README

Expand full README

<img src="assets/banner.svg" alt="Mira — the local-first AI engine for short-form video" width="100%"> <img src="https://img.shields.io/badge/license-AGPL--3.0-8B5CF6?style=for-the-badge" alt="AGPL-3.0"> <img src="https://img.shields.io/badge/python-3.11%2B-4C8DFF?style=for-the-badge&logo=python&logoColor=white" alt="Python 3.11+"> <img src="https://img.shields.io/badge/tests-400%20green-22C55E?style=for-the-badge" alt="400 tests"> <img src="https://img.shields.io/badge/languages-EN%20%2B%20%E0%A4%B9%E0%A4%BF%E0%A4%A8%E0%A5%8D%E0%A4%A6%E0%A5%80-EC4899?style=for-the-badge" alt="English + Hindi"> <img src="https://img.shields.io/badge/runs-100%25%20local-F59E0B?style=for-the-badge" alt="100% local"> Generate consistent, high-quality Instagram Reels & YouTube Shorts — in English and Hindi — from one local command. Faceless channels and personal-brand content · automate the busywork · you keep the taste and the final say <a href="#why-mira">Why</a> · <a href="#how-it-works">How it works</a> · <a href="#production-lanes">Lanes</a> · <a href="#quick-start">Quick start</a> · <a href="#architecture">Architecture</a> · <a href="#license">License</a> --- ## Why Mira Short-form video is a volume game, but volume kills quality — and most "AI video" tools are cloud black boxes that take your keys, your footage, and your money. **Mira flips that:** it runs the whole pipeline **on your machine**, generates a real `.mp4` with **zero API keys** to start, and keeps a **human in the loop** on the things that matter — so you get consistency *and* taste, not slop at scale. - 🖥️ **Local-first** — keys, media, and rendering stay on your machine. No telemetry, no SaaS lock-in. - 🎬 **Two engines, one router** — a quality engine for hero content, a headless engine for trusted auto-volume; the router picks the path by channel trust. - 🗣️ **English & हिन्दी** — script and voice in both, with localization-dub from an English source. - 🆓 **Zero keys to start** — no key → LM Studio + edge-tts + local Remotion/HyperFrames fallback. Add keys only when you want more. - ✅ **Tested** — a stdlib-only command layer with **400 passing tests** as the seam under every feature. ## How it works <img src="assets/pipeline.svg" alt="Mira pipeline: topic → script → voice → visuals → assemble → you review" width="100%"> One command runs the chain end-to-end and hands you a finished vertical video plus captions and metadata — then stops for your review before anything ships: ```bash mira generate --topic "three free AI tools every student should know" ``` ## Production lanes Mira isn't one template — it routes a topic down the right lane for the channel: | Lane | What it makes | For | |------|---------------|-----| | 🎞️ **Faceless** | Designed motion graphics or stock B-roll + TTS | High-volume faceless channels | | 🎥 **Cinematic / story** | Veo-interpolation chain (StoryGen-style) | Hero pieces | | 👤 **Personal brand** | Your footage → clip-factory + captions | Founders, creators | | ✨ **Premium AI** | Motion transfer / lip-sync (metered, gated) | Standout drops | | 🌐 **Hindi** | Localization-dub from an English source | Bilingual reach | --- ## Quick start **Prerequisites:** Python 3.11+ (developed on 3.14) · `ffmpeg`/`ffprobe` on `PATH` · optional: LM Studio (local scripts), Node (Remotion/HyperFrames renders). ```bash cd mira python3 -m venv .venv && source .venv/bin/activate pip install -e ".[dev]" cp .env.example .env # optional keys — blank = free fallback mira up # API + dashboard + local engine supervisor ``` Open the dashboard at `http://localhost:8765`, then generate from **Studio** or the CLI: ```bash mira generate --topic "three free AI tools every student should know" mira bench --topic "three free AI tools every student should know" # Remotion vs HyperFrames ``` **Zero API keys required** — with no keys, Mira uses LM Studio (if running), edge-tts voice, and local Remotion/HyperFrames visuals. Full setup tiers (keys, engines, posting): **[`mira/SETUP.md`](mira/SETUP.md)**. ```bash cd mira && .venv/bin/python -m pytest # 400 tests — stdlib-only core, no wheel risk on 3.14 ``` --- ## Architecture ```mermaid flowchart TD subgraph You D[Dashboard / Cursor / CLI] end subgraph Mira C[mira command layer] R[Router + cost + QA gates] end subgraph Engines OM[OpenMontage — primary] MPT[MoneyPrinterTurbo — headless volume] HF[HyperFrames / Remotion] end subgraph Auto N[n8n schedules] end D --> C N --> C C --> R R --> OM R --> MPT OM --> HF C --> OUT[.mp4 + captions + metadata] ``` **Two engines, one router.** OpenMontage is the in-the-loop quality engine (pipelines, provider selection, checkpoints). MoneyPrinterTurbo handles trusted faceless auto-volume. Channel trust level picks the path. **Design principles:** local-first execution · the command layer is the test seam (JSON in/out before UI) · graceful degradation (missing key → free fallback, never a crash) · cost-aware (budget caps + live meter) · Mac-primary, Windows-supported (RAM/VRAM tier probe). <details> <summary>Repo layout & documentation map</summary> ``` . ├── mira/ ← Python package (mira CLI + providers + pipeline) │ ├── SETUP.md ← step-by-step setup (start here after clone) │ ├── BUILD_LOG.md ← append-only build history │ └── tests/ ← 400 pytest tests ├── dashboard/ ← UI prototype + partials ├── plan/ ← strategy, specs, ADRs (PLAN, DASHBOARD_SPEC, DEV_PLAN, DECISIONS, SECURITY, LAUNCH) ├── repos/ ← bundled engine source (see THIRD-PARTY.md) ├── n8n/ ← workflow exports (mira_daily.json) └── scripts/ ← setup-engines.sh (fetch external engines) ``` Read in order: [`mira/SETUP.md`](mira/SETUP.md) → [`plan/PLAN.md`](plan/PLAN.md) → [`plan/DASHBOARD_SPEC.md`](plan/DASHBOARD_SPEC.md) → [`plan/DEV_PLAN.md`](plan/DEV_PLAN.md) → [`plan/DECISIONS.md`](plan/DECISIONS.md) → [`plan/SECURITY.md`](plan/SECURITY.md) → [`plan/LAUNCH.md`](plan/LAUNCH.md). </details> --- ## License **Mira is licensed under [AGPL-3.0](LICENSE).** Copyright © 2026 Coco Inc. If you run a modified Mira as a network service, AGPL §13 requires you to offer users the complete corresponding source. Mira bundles several upstream engines under their own (compatible) licenses and references two more externally. Full attribution + pinned upstreams: [`THIRD-PARTY.md`](THIRD-PARTY.md). | Engine | License | Status | |--------|---------|--------| | OpenMontage, locally-uncensored | AGPL-3.0 | Bundled (`repos/`) | | MoneyPrinterTurbo, Open-Generative-AI, Free-ai-video-generator | MIT | Bundled (`repos/`) | | Wan2.2, hyperframes, StoryGen-Atelier | Apache-2.0 | Bundled (`repos/`) | | n8n | Sustainable Use License (fair-code) | External — `bash scripts/setup-engines.sh` | | SkyReels-V2 | Skywork Community License | External — `bash scripts/setup-engines.sh` | | AVTR-1 | Goodsize model license | Dropped (AGPL-incompatible + commercial trigger) | --- ## Status | Area | State | |------|-------| | `mira` command layer | Phase 0 complete — 400 tests green | | Offline generate pipeline | Working — real `.mp4` with no keys | | Stock B-roll + caption burn | Working with a Pexels key | | Dashboard prototype | Shell + partials; wiring in progress | | Publish / pod / cloud GPU | Planned (DEV_PLAN phases 5–8) | **Known issue (fonts):** compiled output may embed web fonts that aren't licensed for redistribution: commercial faces that Google Fonts serves under a licence to Google only, and subsets of open fonts that keep a Reserved Font Name. Until [#3](https://github.com/coco-research/Mira/issues/3) is fixed, Coco doesn't publish or distribute Mira-compiled output that embeds web fonts; if you use Mira, check the fonts in your output before sharing it. ## Contributing Open-source under AGPL-3.0; contributions welcome via PR to `main`. Please don't commit `.env`, `media/`, or generated video artifacts, and append context to [`mira/BUILD_LOG.md`](mira/BUILD_LOG.md) after meaningful changes. Part of <a href="https://github.com/coco-research">CoCo Research</a> — open-core AI orchestration & agentic developer tooling.

Loading the repository README…

Skills in this repository