OpenMontage: Agentic Video Production for Claude Code and Cursor

OpenMontage climbed to #1 on GitHub Trending (Repository of the Day) with 23.6k stars and 2.6k forks — not because it generates one clip from a prompt, but because it turns Claude Code, Cursor, Copilot, or Windsurf into a structured video production studio.

Most people landing on the repo ask the same five things: Do I need paid keys? Is this just slideshow Ken Burns? How do I actually run it? What does it cost? How is this different from ViMax or Thariq's Fable + Remotion pipeline?

This post answers those — with a short summary up front, then the questions that matter.

TL;DR

Repo: github.com/calesthio/OpenMontage · Agent contract: AGENT_GUIDE.md

What problem does OpenMontage solve?

Single-shot tools (Runway, Kling, Veo, Sora) give you a clip. Production teams follow research → script → assets → edit → review → render. OpenMontage encodes that flow as:

research → proposal → script → scene_plan → assets → edit → compose

Each stage has a YAML manifest in pipeline_defs/ and a Markdown director skill in skills/pipelines/. The agent calls Python tools from tools/, self-reviews, checkpoints JSON state, and stops for human approval on creative choices.

Important distinction from the README: OpenMontage can animate still images — but it can also assemble real edited video from free stock and open archives (Archive.org, NASA, Wikimedia, Pexels). Prompt explicitly for "documentary montage" or "real footage only" if that is what you want.

How do I install and run it?

Prerequisites: Python 3.10+, FFmpeg, Node.js 18+, an AI coding assistant.

git clone https://github.com/calesthio/OpenMontage.git
cd OpenMontage
make setup
cp .env.example .env   # optional provider keys

Open the folder in Claude Code or Cursor, then prompt in plain language:

Make a 60-second animated explainer about how neural networks learn

Or for the free footage path:

Make a 75-second documentary montage about city life in the rain.
Use real footage only, no narration, elegiac tone, with music.

No make? Manual path from README: pip install -r requirements.txt, cd remotion-composer && npm install, pip install piper-tts.

Windows note: If npm install fails with ERR_INVALID_ARG_TYPE, try npx --yes npm install. Issue #148 tracks Remotion browser setup on restricted networks.

Apple Silicon: Recent merge (#136) added MPS support for local GPU paths — relevant if you add requirements-gpu.txt and run local generation.

What can I make without paying for APIs?

README sample "Afternoon in Candyland" — 12 FLUX images, Remotion motion, music — $0.15 with an image provider configured. With zero keys, you still get Piper + archival montage + Remotion composition — just not FLUX/Veo/Kling generation.

Run make demo for zero-key demo renders without writing a prompt first.

Which pipeline should I pick?

Reference-video workflow: Paste a YouTube Short, Reel, or local clip — the agent analyzes pacing, hook, and structure, then proposes 2–3 differentiated concepts with cost estimates before asset spend.

How much does it actually cost?

OpenMontage ships budget governance: estimate before execution, default $10 cap, $0.50 approval threshold per action, decision log of every provider choice.

README-published examples (with providers configured):

Local vs paid is an open community debate — Issue #70 asks how to optimize the stack. Rule of thumb: documentary montage + Piper = cheapest; Kling/Veo motion clips = dollars, not cents.

OpenMontage vs alternatives people compare it to

vs one-shot video APIs (Runway, Kling, Sora, Veo)

vs ViMax

ViMax is another agentic end-to-end framework (director + screenwriter + generator). OpenMontage is more production-engineering shaped: 52 registered tools, JSON schemas, budget caps, 12 named pipelines, Remotion and HyperFrames runtimes. ViMax centers on narrative shot generation; OpenMontage adds documentary retrieval, screen demos, LMS corporate playbooks, and reference-video cloning.

vs hand-rolled Claude Code + Remotion (Fable launch video)

Thariq's Fable pipeline is a custom one-off: Whisper → JSON EDL → ffmpeg → Remotion → Figma MCP. OpenMontage productizes that pattern — registry-discovered tools, stage skills, slideshow-risk scoring, delivery-promise enforcement so agents do not ship "animated PowerPoint" and call it cinematic.

If you already run /goal loops for creative work, OpenMontage is the repo-shaped version with tests and manifests.

vs OpenCut

OpenCut targets NLE-style editing with plugins and MCP. OpenMontage targets greenfield generation from a text brief — closer to "production team in a repo" than "CapCut replacement."

Does it work with my coding agent?

Dedicated entry points all redirect to shared docs:

OpenClaw-style agents: README includes explicit instructions — read AGENT_GUIDE.md, discover tools via registry, never improvise the pipeline:

python -c "from tools.tool_registry import registry; import json; registry.discover(); print(json.dumps(registry.support_envelope(), indent=2))"

MCP support is requested but not shipped — Issue #164. Today the agent runs Python tools directly, not via an MCP server layer.

What are people struggling with? (GitHub issues)

Worth knowing before you bet production on it:

Maintainers posted #149 Call for PR reviewers with video production experience — signal the project is hot but still consolidating.

AGPL-3.0 — can I use this commercially?

OpenMontage is AGPL-3.0, not MIT. If you modify it and run it as a network service users interact with over a network, AGPL typically requires source availability for those modifications. Internal use on your machine with Claude Code is a different calculus than shipping "OpenMontage-as-a-Service."

Not legal advice — read the license and consult counsel before wrapping it in a paid product.

Remotion vs HyperFrames — which renderer does the agent pick?

OpenMontage ships two composition runtimes. The agent locks render_runtime at proposal time — you do not manually flip between them mid-pipeline.

Rule of thumb from the README: Remotion is default for data-driven explainers and anything using the existing React scene stack. HyperFrames is default when the brief reads like motion graphics — heavy type animation, registry blocks, website-to-video, or the character-animation pipeline's GSAP output.

If you prompt for a Ghibli still-image animation, you almost always land on Remotion. If you prompt for a product launch reel with bold typography, HyperFrames is the natural fit. When in doubt, ask the agent to state the chosen runtime in the proposal stage before asset spend — that is when the decision log records alternatives considered.

Both paths still run the same upstream stages (research, script, scene plan). The split is compose only. That matters for developers already invested in Remotion from the Fable launch pipeline: your mental model transfers directly to OpenMontage's default explainer path.

Prompts that work (copy-paste)

Zero keys:

Make a 45-second animated explainer about why the sky is blue, with narration and captions.

Real footage:

Make a 90-second documentary montage about what a city feels like at 4am.
Use real footage only, no narration, elegiac tone.

From a reference:

Here's a YouTube short I love. Make me something like this, but about CRISPR for high school students.
Keep the pacing and hook; change the topic and visuals.

Budget-conscious Ghibli-style:

Create a 30-second Ghibli-style animated video of a magical floating library at golden hour.
Prefer FLUX stills + Remotion motion, not paid video generation.

More in repo PROMPT_GALLERY.md.

How OpenMontage fits the agentic stack we cover

OpenMontage sits at the intersection of trends explainx.ai has tracked all month:

• Loop engineering — long-horizon agent stages with checkpoints
• Token economics — multi-provider pipelines burn tokens and API dollars; budget caps matter
• Remotion as composition layer — same engine as Fable's launch edit
• Open weights vs gated APIs — documentary path works while Fable 5 stays offline

If you are choosing between learning OpenMontage and waiting for frontier video APIs to get cheaper, the decision hinge is: do you need one clip or a repeatable production pipeline you own in git?

Related reading

• ViMax agentic video guide — alternative end-to-end framework
• Fable 5 edited its own launch video — custom Claude Code + Remotion pipeline
• OpenCut plugins and MCP — editor-first open source
• Claude Code /goal command — long-running agent patterns
• What are AI agents? — definitions for agentic production

Official: OpenMontage GitHub · Providers guide · Architecture · Discussions

Star counts, issue counts, and demo costs reflect the OpenMontage repository as of June 27, 2026. The project moves quickly — verify main README and open issues before production use.