Agentic video production · Open source · September 2026 Edition
OpenMontage Course: agentic video production from install to finished film
OpenMontage is not another browser generator. It is an open system that turns your coding assistant into a production studio: it runs the research, writes the script, plans the scenes, orders the shots, adds narration, music and subtitles, assembles the timeline and checks the result before showing it to you. This course teaches you to direct that process, from an empty terminal to a film you send to a client.
Upfront and honestly: this is not a click-and-go tool. You install a repository, run a few terminal commands and drive production in sentences. You will not write a line of code, but you will open a terminal. If you would rather stay in a browser, the Sora 2 course is the better fit. 176-page PDF, 9 modules, 60+ production prompts, 6 projects.
Note on language: the course PDF is written in Polish. Every technical element in it, meaning commands, configuration, pipeline names and prompts, is in English, because that is how the tool works. This page exists so you can judge the scope before buying. The Polish edition lives at kursvideoai.pl/kurs-openmontage.
176-page PDF
9 modules
60+ production prompts
6 projects
Discord for students
Lifetime access
One-time payment, no subscription · PDF by email on release day · Lifetime access · VAT invoice
Definition
What OpenMontage is
OpenMontage is an open-source agentic video production system released under AGPL-3.0. Rather than generating single clips from a prompt, it runs the entire production: research, script, scene plan, assets, editing, narration, music, subtitles and render. Your coding assistant is the orchestrator, not a separate application.
For two years, working with AI video looked the same: you write a prompt, you get a few seconds of footage, and then you assemble the film yourself. Sora, Veo, Kling and Runway keep getting better at it, but they all stop in the same place: a single shot.
OpenMontage moves the line one layer up. It does not generate video itself; it runs the process and orders material wherever it fits best. The project reached the top of GitHub Trending within days of launch, and not because it makes prettier clips, but because it stops being about clips at all.
Seven stages
What one production looks like
Every pipeline runs through the same sequence. Each stage has its own instruction file that teaches the agent how to execute it, and gates along the way where you make the call.
1
research
15-25 web searches: data points, audience questions, trending angles and visual references, all cited in a structured brief.
2
proposal
Concept, an honest tool path and a cost estimate, before anything gets generated.
3
script
Script with voice direction: pacing, pauses, emphasis. First approval gate.
4
scene_plan
Scene plan, shots, durations, camera movement, a provider assigned per scene.
5
assets
Generation and retrieval. The storyboard pauses on a scene-by-scene contact sheet.
Render through Remotion or HyperFrames, followed by a mandatory self-review.
openmontage · project: baltic-01
pipeline: documentary-montage · budget: $0.00 · gates: on> Make a 75-second montage about the Baltic coast in winter.
> Real footage only, no narration, music yes, elegiac tone.
→ research: 18 queries, cited brief saved
→ corpus: 214 shots from Archive.org and Wikimedia, index ready
→ scene plan: 9 sequences, 74.6 s⏸ GATE: scene plan is waiting for your approval>// fixing it before anything renders> Drop the drone shot over the port, too modern for this mood.
→ edit: 23 cuts, royalty-free music matched to pacing
→ self-review: ffprobe ✓ · frame samples ✓ · audio levels ✓
→ delivery promise: „real footage” ✓ (no static frames)
✓ renders/final.mp4 · 74.6 s · 1080p · provider cost: $0.00
A run from module 5. Free path, a gate before the render, zero provider cost.
Reality check
Four numbers that change the bill for video
All of them come from the project itself: from the repository and from productions OpenMontage publishes with the full prompt, the pipeline used and the cost. That is rare in this field, so it is worth using instead of guessing.
$0
the full free path
Offline narration, footage from Archive.org, NASA and Wikimedia Commons, composition in Remotion, subtitles burned in with FFmpeg. A finished film without a single API key. Most people buy three subscriptions before they ever check whether this path is enough for them.
$1.33
a one-minute animated short
That is what „The Last Banana” cost: six Kling v3 clips through fal.ai, Google Chirp3-HD narration, royalty-free music, word-level captions and a Remotion composition. The whole production, not a single shot.
60+
provider integrations
Cloud APIs, local models, stock libraries and open archives behind one selection layer that scores every provider across seven dimensions and logs the reasoning. Without understanding that layer you pick at random and pay for the randomness.
700+
knowledge files for the agent
Stage directors, creative techniques, quality checklists and deep production knowledge packs. They are what makes the agent work like a crew rather than another prompt box. Nobody reads them, because nobody knows they are there.
Before you start
What you need to run OpenMontage
OpenMontage system requirements
Component
Requirement
What it is for
Python
3.10 or newer
Virtual environment, production dependencies
Node.js
18+, 22+ for motion graphics
Remotion and HyperFrames composition engines
FFmpeg
Any current build
Encoding, audio mixing, subtitle burn-in
Coding assistant
Claude Code, Cursor, Copilot, Windsurf or Codex
It is the orchestrator; there is no separate control engine in the system
OS
macOS, Linux or Windows
Windows is the bumpiest path, and the project ships no Docker image
API keys
Optional
Without them the free path works; every key unlocks more tools
GPU
Optional
Unlocks local video generation instead of a cloud bill
Getting started
Installing OpenMontage in five steps
A condensed version of module one. In the course each step has a macOS, Linux and Windows variant, along with the errors that actually come up and how to get around them.
1
Prepare the environment
Python 3.10+, Node.js 18+ and FFmpeg. On macOS and Linux that is three package manager commands. On Windows, expect to work around a Node dependency install error, or pick the WSL route, which usually saves you an evening.
2
Clone the repository and run setup
The repository is public and free. The setup script creates a virtual environment, installs Python and Node dependencies, adds the free speech engine and creates the config file for your keys.
3
Verify the install on demo renders
The project renders demos without any keys. This is the right moment to confirm the whole chain works: Python, Node, FFmpeg and the composition engine. Before you spend anything, not after.
4
Connect your coding assistant
Open the project in Claude Code, Cursor, Copilot, Windsurf or Codex and describe the film you want. The agent reads the pipeline manifest, then the stage skill, then reaches for tools. From here on you work in sentences, not commands.
5
Only now add provider keys
Set cost estimation and a spend cap first, then paste your first video generation key. The reverse order is the most expensive beginner mistake and the most common reason people bounce off this tool.
Decision table
Eleven pipelines and what comes out of each
Choosing the pipeline is the first and most consequential production decision. It sets the cost, the time and whether the result is something you can show a client at all. The agent will choose on its own if you do not, and that is where most disappointment starts.
OpenMontage production pipelines, output and cost
Pipeline
What it produces
Best for
Cost
Animated Explainer
Explainer with research, narration, visuals and music
Education, tutorials, topic breakdowns
$0 – $2
Documentary Montage
Montage cut from real footage, corpus built from open archives
Video essays, mood pieces, real-footage films without paid models
A batch of ranked short clips cut from one long source
Repurposing long content for social
$0
Talking Head
Footage-led speaker videos
Presentations, vlogs, interviews
$0 – $1
Avatar Spokesperson
Avatar-driven presenter videos
Corporate comms, training, announcements
$1 – $4
Screen Demo
Polished screen recordings and walkthroughs
Product demos, tutorials, documentation
$0
Podcast Repurpose
Podcast highlights turned into video
Podcast marketing, audiograms
$0
Localization & Dub
Subtitles, dubbing and translation of an existing film
Multi-language distribution
$0.50 – $3
Hybrid
Your own footage plus generated support visuals
Enhancing existing footage with graphics
$1 – $3
Indicative costs for material of typical length in each format. They depend on how many scenes you generate and which providers you use.
Zero API keys
A full production for nothing, and why it is not a slideshow
Most free AI video stacks quietly mean „we animate stills”. This one is different: the documentary montage pipeline builds a searchable corpus of real footage from open archives, matches shots to scenes by meaning and cuts them into a timeline. You just have to ask for it explicitly.
Narration
Piper TTS
Free offline speech synthesis. No account, no quota, nothing leaves your machine.
Footage
Archive.org, NASA, Wikimedia
Open archives with real motion footage, indexed semantically so it can be edited.
Stock
Pexels, Unsplash, Pixabay
Developer keys are free and worth getting, because they widen the corpus considerably.
Composition
Remotion
React-based rendering: scenes, text cards, charts, word-level captions, talking head.
Motion graphics
HyperFrames
HTML, CSS and GSAP rendering: kinetic typography, product promos, rigged SVG character animation.
Post
FFmpeg
Encoding, audio mixing, color grading, subtitle burn-in.
Money
What one film actually costs
The figures below are the costs of productions the project published together with their prompt and pipeline. These are not marketing estimates; they are bills for specific films.
$0
Documentary montage from open archives
Real footage, offline narration, royalty-free music. Zero paid generations.
$0.15
30-second animated scene
Short form in a single style, one image model and one video model.
$1.33
60-second short with narration and music
Six Kling v3 clips, Chirp3-HD narration, word-level captions, Remotion composition.
~$4
50-second vertical film
Five generated scenes, sparse narration, a licensed score, a bespoke composition.
~$5
Showcase across seven generated worlds
Three image models and four video models in one music-driven piece.
Quality control
Six gates standing between you and a burned budget
This is the part no browser generator has, and the reason agentic production is usable on client work at all.
Script gate
The agent stops on the finished script and waits for you. A change here costs nothing; the same change after the render costs the entire scene budget.
Storyboard gate
Asset generation pauses on a scene-by-scene contact sheet: takes, prompts, per-asset cost, quality scores. You approve the visuals before the money is spent.
Pre-compose validation
The plan is checked before the render starts, so a broken configuration does not eat GPU time and your evening.
Post-render self-review
Mandatory checks: ffprobe validation, frame sampling, audio level analysis and subtitle verification. The agent never presents material it has not inspected.
Delivery promise enforcement
If you asked for a film cut from real footage and got a slideshow, the render is blocked. No browser generator has anything like this.
Decision log
Every provider choice, style decision and fallback is recorded and replayable. Useful the moment a client asks where a particular shot came from.
Comparison
OpenMontage versus the generators you already use
These are not competitors but different layers of the same job. The table shows where each of them stops.
OpenMontage compared with Sora 2, Runway and CapCut
I am putting this before the pricing section rather than after it, because half the refunds in this field come from promises the tool never made.
It will not make a film in one click
This is a production system with approval gates, so your role is director, not audience. On a longer piece you will say what to fix several times. If you want a single button, this system will wear you out.
It will not replace a camera operator or an actor
Material comes from generative models or from archives. For shots of a real product in a real person's hands you still need a camera.
It is not a browser app
There is no sign-up, no dashboard, no cloud. There is a repository, a terminal and your computer. For some people that is the appeal; for others it is a wall.
There is no Docker image
Installation goes through system dependencies, which can hurt on Windows. The project is young and this is exactly where you feel it.
It will not free you from model bills
The free path is real but has a quality ceiling. The moment you want high-grade generated footage, you pay providers their normal rates.
Curriculum
Nine modules, from an empty terminal to a client-ready film
Every module has theory (what it is), practice (how to do it), anti-patterns (what to avoid) and an exercise with a worked solution to compare against. 176 pages in total.
01
Painless install on macOS, Linux and Windows
18 pages
Python, Node and FFmpeg step by step, virtual environment, config file
Windows: working around the Node dependency install error, with and without WSL
Why there is no Docker image and what that means for your setup
Which coding assistant to connect and how they differ over a long session
A „this works” checklist to run before you spend a cent on a provider
02
Your first film with zero API keys
16 pages
Free offline narration: when it is enough and when the machine shows
Remotion versus HyperFrames, which composition engine for which brief
Word-level subtitles and burning them in without quality loss
Demo renders with no configuration, to verify the whole chain
The honest ceiling: what the free path does well and what you won't show a client
03
How OpenMontage thinks: the agentic architecture
20 pages
No orchestrator in code: your coding assistant is the orchestrator
Pipeline manifests, stage director skills and the tool registry
The seven production stages and what exactly happens in each
The provider selection layer, its seven scoring dimensions, and reading the decision log
How to write a brief so the agent picks the right path instead of guessing
Delivery promise enforcement, blocking renders that look like slideshows
Run replay and the decision log, for when a client asks how a shot happened
08
Language layer: narration, subtitles and localization
16 pages
Choosing a voice, free for drafts and premium for client work
Voice direction inside the script: pacing, pauses, sentence stress
Word-level subtitles and the font traps that eat diacritics
The localization pipeline: one film into several dubbed language versions
Fact-checking research output before it reaches the script
09
Commercial work: licence, law and rates
16 pages
AGPL-3.0 broken down: what covers the code and what does not touch your film
When the licence actually starts to apply to you, and how to live with it in a company
Labelling AI-generated content in advertising, EU AI Act requirements
Accounting for provider spend as a business expense
Pricing agentic production and what to put in the client contract
6 projects
You finish the course with six completed productions
01
An explainer for zero
Sixty seconds with research, free narration, visuals and subtitles. The whole chain without a single API key.
final.mp4 + source brief
02
Documentary montage from archives
A corpus from open archives, semantic retrieval, a cut made from real footage. No narration, with music.
75-90 s montage
03
Cinematic trailer
Concept, script, scene plan, generated shots and composition. Budget calculated before the start.
30 s trailer, ~$3
04
Clip factory from a webinar
One long recording in, a dozen short clips out, ranked and captioned.
12 vertical clips
05
Product film from stills
Approved photos pinned as first and last frame, so the model invents motion without losing product identity.
30-60 s spot
06
Localization for three markets
A finished film through the localization pipeline: transcription, translation, dubbing, per-language subtitles.
3 language versions
Who it is for
Five groups this course pays back fastest for
AI video creators who hit the generator ceiling
You can pull a good clip out of Sora or Kling, but you still assemble every film by hand. OpenMontage sits one layer up: it drives those models and makes sure the clips become a film.
Freelancers and small studios
Clients pay for finished material, not for your hours in a timeline. A pipeline that handles research, script, assets and edit moves you from operator to director, and the margin follows.
Marketers with a content library
Webinars, podcasts and training recordings are sitting there collecting dust. Clip factory and the repurposing pipeline turn them into social clips; localization turns them into new markets.
Developers and vibe coders
You already have a coding assistant and the terminal is your home. For you this is the shortest path to video production without learning an NLE.
Educational creators and faceless channels
Repeatable format, many episodes, a budget near zero. The archive-based free path genuinely closes that business case, and paid providers enter only where they pay off.
Who it is not for: people who do not want to open a terminal, anyone who needs one clip right now (Sora 2 or Kling will be faster), and companies that cannot put AGPL-licensed software on their machines. That last case gets broken down in module nine, because it is often less severe than it sounds.
You might be thinking
Five reasons people bounce off OpenMontage
„It is open source, I will just take it from GitHub”
Please do, the repository is free and I am not selling access to it. What I am selling is the week you will not lose. An install with no walkthrough, eleven pipelines with no decision table, sixty provider integrations with no cost comparison and an AGPL licence with no explanation: that is precisely the wall most people hit in their first weekend. This course is about getting through it once.
„I cannot code, so this is not for me”
You will not write a single line of code, because you drive production in plain sentences. But I will not pretend this is CapCut: you clone a repository, run a few terminal commands and paste keys into a text file. Module one walks through all of it, including the errors that actually come up. If the terminal is a hard wall for you, honestly: start with the AI video course, where everything happens in a browser.
„I already have Sora and Kling, why add this”
Because it does not compete with them, it sits above them. Sora and Kling hand you a clip. OpenMontage runs the research, writes the script, plans the scenes, orders shots from those very models, adds narration, music and subtitles, assembles the film and checks it before showing you. Your subscriptions stay exactly where they are; what changes is who clicks between them.
„The project is young, everything will change in a month”
It will, and that is true: the repository moves fast, tool names shift, providers arrive. That is why the course is built around what stays: the seven production stages, pipeline selection, provider economics, quality gates and licensing. All of it carries to the next version. Updates are free, and the licence guarantees that even if the project stalled, the code stays in your hands.
„AGPL is a legal problem at my company”
Worth separating, because this is where most confusion starts. The licence covers OpenMontage's code, not the films you produce with it: the material is yours and you sell it normally. Obligations begin only when you modify the code and offer it to others as a network service. Module nine breaks this down case by case and says plainly when you need a lawyer and when you do not.
Your instructor
Who writes this course
ŁK
Łukasz Kowalski
Author of the KursVideoAI courses. I use AI video production and coding agents daily on commercial work, so this course distils the workflow I actually run: with costs calculated, with the places where the pipeline breaks, and with an honest line between what this system does brilliantly and what it will not do at all. No rewriting the GitHub docs. More about me, and my testing method is written up in how we test.
Bonuses
What comes with the 176 pages
Pipeline decision table
Client brief in, the right pipeline and a cost range out. One page you will come back to on every job.
Production prompt bank
60+ ready briefs for the agent, spread through the modules and collected at the end: explainer, archive montage, cinematic, clip factory, product film, localization.
Production cost sheet
A formula for film cost plus worked examples from real productions. You size the budget before you run, not after the invoice.
Quality gate checklists
What to check at the script and storyboard gates before you let a render start. A separate list for accepting delivery from a client's side.
Licensing and AI Act cheat sheet
AGPL mapped onto use cases, a contract clause template and a checklist for labelling AI content in advertising.
Discord, #openmontage channel
A channel for questions and updates. The repository moves fast, and I post workflow-breaking changes there as they land.
Get the course
One payment. Lifetime access. The complete OpenMontage course.
September 2026 EditionOne-time payment
See price →
one-time payment · lifetime access · no subscription
✓176-page PDF, 9 modules
✓Step-by-step install on macOS, Linux and Windows
✓Decision table for all eleven production pipelines
✓The free path: a finished film with zero API keys
✓60+ production prompts
✓Production cost sheet, size the budget before you run
The OpenMontage documentation assumes you already understand these. Short definitions follow.
Production pipeline
A complete recipe for one type of film, from research to render. OpenMontage ships eleven, and picking the right one drives both cost and outcome.
Stage director skill
A markdown instruction file that teaches the agent how to execute one production stage. These replace control code.
Orchestrator
The component that runs the whole production. In OpenMontage there is none in code: your coding assistant plays that role.
Approval gate
A point where the pipeline stops and waits for your decision. The two that matter most are the script and the storyboard.
Contact sheet
A per-scene view of takes with their prompts, cost and quality scores. You approve the visuals on it before the render.
Delivery promise
A record of what the film must actually contain. The system blocks any render that fails it, such as a slideshow where a montage was asked for.
Self-review
The mandatory post-render inspection: file analysis, frame sampling, audio levels and subtitle checks.
Decision log
A record of every provider choice, style decision and fallback, so you can reconstruct why the film looks the way it does.
Asset corpus
A searchable set of footage assembled for one production from open archives and stock libraries.
Semantic retrieval
Matching shots by what the image means rather than by tags and descriptions. It is why archive montage works at all.
Composition engine
The layer that assembles finished material into a film. OpenMontage chooses between React-based and HTML-based rendering.
AGPL-3.0
An open source licence. It covers OpenMontage's code, not the films you produce with it.
FAQ
Sixteen questions about OpenMontage and this course
What is OpenMontage?
OpenMontage is an open-source agentic video production system released under AGPL-3.0. Instead of generating single clips from a prompt, it runs the entire production: web research, script, scene plan, asset generation or retrieval, editing, narration, music, subtitles, render, and a self-review of the result. Your coding assistant is the orchestrator rather than a separate application. There is no browser version: you install it locally and drive it in plain language.
Is OpenMontage free?
The software is. You clone the repository and pay nobody for access. You only pay for what you use externally: video, image, voice or music generation at providers. You can also stay at zero, because right after install you have free offline narration, open film archives and composition engines that cost nothing. This course is paid; the tool is not.
Do I need to know how to code?
You do not need to write code. You drive production in plain language and your coding assistant does the work. You do need to be comfortable opening a terminal: installation means cloning a repository, running a few commands and pasting keys into a text file. Module one walks through that on macOS, Linux and Windows, including the errors that actually show up.
What are the system requirements?
Python 3.10 or newer, Node.js 18 or newer (22 for motion graphics), FFmpeg, and a coding assistant: Claude Code, Cursor, Copilot, Windsurf or Codex. It runs on macOS, Linux and Windows. API keys are optional; without them the free path works. A GPU is not required but unlocks local video generation instead of a cloud bill.
Does it work on Windows?
Yes, though it is the bumpiest of the three paths. Installing Node dependencies can fail with an ERR_INVALID_ARG_TYPE error. The workaround is a single line and module one covers it, along with the WSL route that usually saves time. The project ships no Docker image, so there is no shortcut around this.
Do I specifically need Claude Code?
No, although that is what I teach on. OpenMontage works with any coding assistant that can read files and run code, which includes Cursor, Copilot, Windsurf and Codex. The differences come down to comfort in long sessions and to usage limits, not to whether the pipeline executes. If you want to get the most out of the agent itself, I have a separate Claude Code course.
What does one film actually cost?
It depends on how much material you generate with paid models. The project publishes the cost of its own productions and the range is concrete: roughly $0.15 for a short animated scene, $1.33 for a one-minute short with narration and music, about $4 for a fifty-second vertical film, about $5 for a showcase built from seven generated worlds. The archive-based path costs nothing. The course includes a sheet that sizes this before you run a production.
Can you really make a film without spending anything?
Yes, and it is this system's strongest feature. A free offline engine narrates, footage comes from Archive.org, NASA and Wikimedia Commons, Remotion or HyperFrames composes, and FFmpeg burns in subtitles. Crucially it does not have to be a slideshow: the documentary montage pipeline builds a searchable corpus of real footage and cuts a timeline from it. You just have to ask for that explicitly, and module five shows exactly how.
How many pipelines does OpenMontage have?
The project describes more than ten and its table lists eleven: animated explainer, animation, avatar spokesperson, cinematic, clip factory, documentary montage, hybrid, localization and dub, podcast repurpose, screen demo and talking head. Each is a complete recipe for a different kind of film with its own tooling and its own economics. Picking the pipeline is the first and most consequential production decision, which is why the course gives it a dedicated decision table.
How is this different from Sora, Runway and CapCut?
Sora and Runway hand you a single clip from a prompt; CapCut gives you tools to edit by hand. OpenMontage does something categorically different: it runs the whole production and treats those models as suppliers of material. What you get is an edited film with narration, music and subtitles, not a raw clip or an empty timeline. The cost of entry differs too: those tools open in a browser, this one installs locally.
Can I sell films made with OpenMontage?
Yes. AGPL-3.0 covers the code, not the material you produce with it, so the finished film is yours and you bill for it normally. Licence obligations arise only if you modify the code itself and offer it to others as a network service. Do watch two things that are independent of the licence: rights to source material from archives, and the obligation to label AI-generated content in advertising. Both get their own sections in module nine.
Can I start from a video I already like?
Yes, and it is usually faster than inventing the perfect prompt. You paste a link to a video, Short, Reel or a local file, and the agent analyses transcript, pacing, scenes and style. You get back several differentiated concepts, a note on what it keeps and what it changes, an honest tool path and a cost estimate, before full production starts.
Do I need a graphics card?
Not for cloud models and not for the free path. A GPU only matters once you want to generate video locally without paying per generation. That is a sensible direction at volume, but it needs a separate install and patience for render times. Module six covers when that investment pays off and when it is cheaper to pay a provider.
Is the course in English?
The course PDF is written in Polish, and this page exists so English-speaking readers can evaluate whether the material covers what they need. Everything technical in it — commands, configuration, pipeline names, prompts — is in English anyway, since that is how the tool itself works. If you read Polish comfortably, or are happy running the PDF through a translator, the content transfers cleanly.
How does this differ from your AI video and Claude Code courses?
The AI video course is about browser generators and getting a good clip out of them without installing anything. The Claude Code course is about the agent itself: CLI, hooks, MCP, Agent SDK, the tool that plays director here. This course is about the production system that ties the two together and takes you from brief to an edited film. Three separate products, bought independently, with no overlap in material.
When do I get the PDF and in what format?
The course is in production and this page is an announcement. The 176-page PDF goes out by email on release day, along with an access code for the #openmontage Discord channel, where I post progress. PDF rather than video, because prompts and commands are easier to copy from text. Lifetime access, free updates, VAT invoice, and a 14-day withdrawal right under consumer protection law.