What is Cinematic Skill?
Cinematic Skill is an agent skill for Claude Code, Codex, Cursor, Gemini CLI and OpenCode that makes narrated films from code rather than generated video. You ask for a documentary or explainer on a subject; the agent researches and writes the script, times it with a scratch voice, collects images, and lays out one large board of photos, typeset documents, cards, maps and counters. A single continuous camera then travels across that board, with stamps, highlights and numbers timed to the narrator's words.
Frames are drawn with Skia in Python, so a film renders the same way every time and a new voice track re-times it without editing the scene. A Wikimedia Commons fetcher downloads archive images, records each licence and refuses anything that is not public domain, CC0, CC BY or CC BY-SA unless you override it. Sound effects are synthesised rather than sampled, and the finish adds a loudness-normalised mix, four caption styles and an SRT file. The agent stops for approval at four gates, including the beat table and a 30-second draft.
The repository is MIT licensed, has one author and was first released as 1.0.0 on 9 November 2026, followed by 1.0.1 the same day. It has 260 stars and 16 forks.
Is Cinematic Skill worth it?
Cinematic Skill makes the right bet for factual video: no generated footage at all. Every frame is drawn in Python with Skia over real archive photos, so the result is reproducible, cheap to change and cannot invent a picture of an event that never happened. The skill writes that into its rules: an AI image may illustrate, but never stand as evidence of a real person or event, and a document drawn in code must be labelled a recreation and quote only words that were actually written.
The sourcing tool is the most useful part. fetch_commons.py searches Wikimedia Commons, records the credit and licence for each file, writes credits.md, and refuses anything outside public domain, CC0, CC BY and CC BY-SA unless you pass an override. Its pattern also rejects non-commercial and no-derivatives variants. For a history channel or a teacher, that removes the most common way a documentary goes wrong after publication.
Timing is built on the narration. A free scratch voice from macOS say or espeak-ng sets the timings first; every hit is keyed to a spoken word, so when you swap in a final voice the film re-times itself. The README reports that the bundled 92-second example renders at 1080p in about 6 minutes on a 16 GB laptop, a figure we have not reproduced.
The catch: the project is four days old. It has 260 stars and 16 forks, three commits from one author, and no issues filed yet, so there is no public record of how it behaves outside the author's own films. The look is also deliberately narrow: one camera over one board, in seven looks such as a cork evidence wall or blueprint. That suits history, true crime and explainers; it will not produce live action, characters or product footage, and on Linux or Windows you must fetch fonts separately to match the Mac typefaces.
Our take: for anyone whose videos are built on real photographs and need to be right about their sources, this is a better starting point than any generator-led skill. For anything else, the evidence-board format will feel like one trick played well.
Where it earns its place
A secondary-school history teacher
Make a 90-second film on a historical event for a lesson
Ask the agent for the documentary, approve the beat table and facts, let fetch_commons.py pull public-domain photos with credits, and render with the free scratch voice. No paid services are required.
A solo YouTube creator running a true-crime channel
Produce a narrated episode on the cork-wall look with sourced images
Record the final voice in sections, let split_takes.py cut it into timed lines and flag suspect cuts, and hand over credits.md with each upload. The cost is the agent plan and any TTS service used.
A startup founder preparing a short brand story
Tell the company's origin with team photos, a map and a few big numbers
Use the studio or charcoal look with your own photos graded in natural colour, counters for the figures and a route on the map. Rendering is local, so no per-minute video fees.
Before you commit
- Setup
- npx skills add (Node 22.20+), a plugin command for Claude Code or Codex, or a shell installer; then pip install skia-python, numpy and pillow, put ffmpeg on PATH and run check_setup.py.
- Integrations
- Claude Code, Codex, Cursor, Gemini CLI, OpenCode; Wikimedia Commons; macOS say, espeak-ng or any TTS command; your own photos or AI stills.
- Data and privacy
- Rendering is local. Script prompts go to your agent's model provider, Commons searches go to Wikimedia, and text sent to a TTS service follows that service's terms.
- Cost at scale
- Not published. The code costs nothing; spend is the agent plan, a paid TTS voice if used, and render time on your own machine.
- Limits
- Needs Python 3.9+; renders above 1080p need the performance notes; only one board-and-camera format; non-Mac systems need get_fonts.py per project.
What’s good
- No generated video: every frame is drawn from code and renders identically
- Commons fetcher records licences and refuses non-free files
- Animation hits keyed to spoken words, so a new voice re-times the film
- Synthesised sound effects avoid sample licensing
- Installs for Claude Code, Codex, Cursor, Gemini CLI and OpenCode
What’s not
- Four days old, one author, three commits, no track record
- One visual format: a camera moving over a single board
- Linux and Windows need a separate font fetch to match the Mac look
- Render speed figures come only from the README
Best for
History and science creators, teachers and journalists who want narrated documentaries built from properly licensed archive images.
Cinematic Skill alternatives
Cinematic Skill: common questions
- Is Cinematic Skill worth it?
- A disciplined way to make Ken Burns-style documentaries with an agent, where code-drawn camera work and a licence-checking image fetcher matter more than the four-day-old packaging. We rate it 3.8 out of 5.
- How much does Cinematic Skill cost?
- Free under MIT. Runs on your own coding agent plan; the scratch voice uses macOS say or espeak-ng at no cost, and a final voice or score costs whatever TTS or music service you choose.
- What are the downsides of Cinematic Skill?
- Four days old, one author, three commits, no track record; one visual format: a camera moving over a single board; linux and Windows need a separate font fetch to match the Mac look; render speed figures come only from the README.
- Who is Cinematic Skill best for?
- History and science creators, teachers and journalists who want narrated documentaries built from properly licensed archive images.
- How much does Cinematic Skill cost for a team or in production?
- Not published. The code costs nothing; spend is the agent plan, a paid TTS voice if used, and render time on your own machine.
- Is Cinematic Skill safe for company data?
- Rendering is local. Script prompts go to your agent's model provider, Commons searches go to Wikimedia, and text sent to a TTS service follows that service's terms.
- What are the best Cinematic Skill alternatives?
- The closest alternatives we have reviewed are Video Director, HyperFrames and Stickman Video Director.
Answers checked Nov 13, 2026.