
A YouTube tutorial with 280K views had an open-source one-click version 6 days later: cinematic-skill, 9 days, 700+ stars
What it is A skill for coding agents like Claude Code and Codex. One prompt has it write the script, add narration, and find public-domain vintage photos, then uses code to draw an evidence-wall-style documentary short in one continuous shot, frame by frame, without an AI video model.
Why it blew up: Zane Hoyer's "Claude builds a documentary with code" tutorial hit 11-03 280K views with no open-source version of it anywhere. The author shipped his own version 6 days later, launched it on his "free resources" account, and then an Instagram repost account and GitHubDaily drove a bigger second wave.
CHow you could follow it Don't build another documentary skill. Copy the playbook: find an AI coding tutorial that got 100K views in a week and whose creator never open-sourced it, ship a one-line-install version within a week, publish the finished video before the repo, and hand repost accounts a ready-made selling point.
- Data (updated daily) · Repo created 2026-11-09 · 810 stars · 7 days +585
GitHub · stars Data 2026-11-11 → 2026-11-20 (10 days, 10 observations). Points before 10-28 come from deeply / archive backfill, not every day has one
An X account that posts only "free AI resources" packaged the approach of the trending YouTube video "Make a Netflix-style documentary with Claude writing the code" into a skill you install with one line. It hit 739 stars and 112 forks in 9 days. The second week grew faster than the first, and the author didn't drive this wave. Instagram repost accounts running "comment DOCU and I'll send you the link" did. Below: what it is, where the stars came from, how crowded this niche is now, and which parts you can borrow.
Four minutes before posting
2026 Nov 9 15:35 (UTC), Moh4696/cinematic-skill pushed a commit on GitHub with the message: "Drop the tutorial credit from the README, SKILL.md and PROMPT.md." The deleted line read: "The single-camera evidence-wall method follows Zane Hoyer's tutorial on code-built documentaries," followed by a YouTube link.
Four minutes later, 15:39, the X account @exploraX_ (26.9k followers) posted: "I created /cinematic-skill that helps Claude and Codex make cinematic films. 100% free and MIT licensed", with npx skills add Moh4696/cinematic-skill on the line below. As of 11-18 the post had 32k views, 439 likes, and 676 bookmarks. The next day the repo had gained 92 stars.
The link that later got deleted is where the whole thing started.
What it is
A skill for Claude Code, Codex, Cursor and Gemini CLI. You tell your agent "make a 90-second documentary about the Wright brothers" and it writes the script and a narration track, pulls public-domain old photos from Wikimedia Commons while logging the licenses, then draws every frame with Python and Skia: one continuous camera move across a corkboard of evidence, with photos, newspaper clippings, maps and red string sliding between each other. Stamps, highlighter strokes and rolling numbers land on the exact moment the narrator says the matching word. It pauses for human sign-off at three points: the storyboard, the visual style, and a 30-second sample render.
The most prominent line in the README: "Nothing is generated video. Every frame is drawn by code." No video model means no per-render cost, every render is identical, and swapping the voiceover re-times the whole film automatically. The repo ships a 92-second sample film, and the author says it renders at 16p on a 1080G-RAM laptop in about 6 minutes.
This pitch line was later copied word for word by repost accounts dozens of times.
The demand was lit by someone else
In Nov 3, YouTube editor Zane Hoyer (136k subscribers) published "Claude Edited This Netflix-Level Documentary (Copy Me)", which has reached 11-18 views and 283k plays. It's the same workflow: spend 20 to 60 minutes gathering reference images on Pinterest, tell Claude to "do everything in code, no AI video generator", feed it real historical photos, and get 30 to 40 seconds of one-take camera moves. The line from his video: "the secret isn't one magic prompt."
The next day another creator, Amine AI, posted "Claude Made This Netflix-Style Documentary From Code (FREE SKILL)", with 46k plays. But the "free" skill was locked behind joining his Skool community.
So by early Nov the situation was this: hundreds of thousands of people had seen "Claude can make documentaries from code", wanted something they could install and use, and nobody had shipped an open-source version with no signup. exploraX 11-08 first posted a finished Al Capone documentary made with this method ("Just a single prompt + ref images", 50k views, more bookmarks than likes 838 574), then the next day quoted that post to launch the skill. From Zane's video to the skill going live took 6 days.
Where the stars came from: two peaks, and the second one wasn't the author's
Daily star gains (Trendshift, by UTC day):
| Date | New stars | What happened that day |
|---|---|---|
| 11-10 | +92 | Day one after the author's X launch |
| 11-11 | +63 | |
| 11-12 | +28 | Instagram repost accounts start posting Reels |
| 11-13 | +55 | |
| 11-14 | +117 | GitHubDaily publishes a Chinese introduction |
| 11-15 | +103 | |
| 11-16 | +131 | Hindi-English repost account "YE FILM CODE HAI!" |
| 11-17 | +97 |
The author never changed a line of code after 11-09, but from 11-14 to 11-16 the repo gained more than 100 stars a day, higher than on the two launch days. The second wave came from two sources:
The first was small Instagram and Facebook accounts. Starting 11-12, Bibek yadav, StackWhiz ("One prompt. A whole documentary, built from code"), Omar | Building with AI ("Follow me and comment DOCU and I'll send you the repo link") and Confuse Coder ("Comment NOTES") posted Reels with nearly identical pitches: skip AI video generators that blur and hallucinate, and use this to start a faceless YouTube channel. These accounts were growing their own followers with comment-keyword link giveaways, and I found no sign they were paid. Per-post view counts weren't available.
Another route is Chinese. GitHubDaily (92K followers) 11-14 posted a Chinese write-up that opened with: "Wanted to make a history explainer video: script, voiceover, images, editing, each step by step, and then you have to check the copyright on every old photo one by one. An entire afternoon gone." It drew 11K views and 329 saves, and gained +117 that day. Later a few Chinese accounts rewrote it and reposted, with very little engagement.
A few other things worth noting: the fork-to-star ratio was about 15%, high for a new repo of this size, which suggests people actually installed it (though it's also possible that people arriving via the comment-for-link posts are more likely to fork; that's my inference). The repo was less than 60 days old, too new for GitHub trending, and Trendshift never placed it in the top 25, so the star growth had nothing to do with rankings. I found no discussion on HN or Reddit.
The author already did this for a living
@exploraX_ is an account that only posts "free AI resources". It has 59.5k posts, and its pinned long-form thread, "20 agentic skills", has 6.6M views. Its owner's GitHub repos in the same vein each have hundreds of stars: 300-free-resource-websites 376, build-ai-agents-free 390, open-3d-skills 422. open-3d-skills was built in late Oct using the same template: a demo GIF, a one-line install, plugins for several agents, and one complete example.
His followers already come to save free repo links, and bookmarks often outnumber likes. A copycat can't get this: you don't have those 26.9k people waiting to bookmark, so the same post would bring in 92 stars on day one.
Who it's competing with
The "agent writes code to make video" lane was already crowded by November, and even the narrower "code-drawn documentary" niche wasn't theirs alone.
| Project | Launched | Stars (11-18) | Compared with alternatives |
|---|---|---|---|
| sarpstar/cinecode | 11-03 | 2,677 | Same wave, three.js plus a headless browser, for any kind of film. 11-09 gained 872 in one day, more than 3 times this repo's rate |
| kaomei/stickman-video-director | 8-03 | 6,808 | Also aimed at faceless YouTube, stick-figure art style |
| vincentsch/explainroo | 9-25 | 1,359 | Local voiceover + word-timed captions + canvas rendering. Closest route, but no old photos or evidence-board look |
| Leonxlnx/cinetic | 9-28 | 1,275 | Remotion for product launch videos |
| bangbuilds/code-doc-film, parallax-doc | October | 20 / 8 | Earlier code-documentary skills that nobody promoted and that have almost no stars |
| kairo-labs-ai/vintage-documentary and 4 others | 11-14~17 | 0 | Copycat archival-photo documentary skills with nearly the same look |
The last two rows are the most telling: the earlier and later projects do roughly the same thing, yet one has 20 stars and the other 0. The star count wasn't decided by the code. It came down to shipping it in the week the tutorial was hottest, getting it into the hands of people who had just watched the tutorial, and having an account that could hand out the link.
If it were you
C: There's a clear playbook to learn here, but building another documentary skill isn't worth it. Within two weeks cinecode (2,700 stars) had already taken most of this slot, and any identical clone after 11-14 has been getting 0 stars. What you can borrow is the method, not this topic.
Who it's for: developers who want stars for their own open-source project, who already watch AI coding tutorials on YouTube, who move fast enough to turn a prompt into an installable, runnable skill with an example within a week. Ideally they have an X or Xiaohongshu account with a few thousand followers, or know a Chinese repost account like GitHubDaily.
Who it's not for: anyone who wants another version in the "code documentary" slot (it's too late unless you find an audience it hasn't covered); anyone with no distribution who expects GitHub organic traffic to do the work (the earlier code-doc-film proves that doesn't happen); anyone trying to make money from it (this category is about building reputation and followers).
How to do it: watch for viral AI-coding tutorials on YouTube with titles like "Copy Me" or "FREE SKILL" that pass 100k plays within a week, where the creator only shares the video and prompts, or locks the skill behind a paid community. Within a week, turn it into a version you can install with one npx skills add line, and use the same README template: an animated finished film above the fold, a one-line install, plugins for Claude Code / Codex / Gemini, and one complete runnable example. To differentiate, change the audience or platform: vertical 9:16, Chinese narration, aimed at history explainer accounts on Douyin and WeChat Channels (my inference: it only outputs landscape right now, and Chinese voiceover would need to be added yourself). Keep the tutorial credit in the README; don't delete it.
How to get users: post the finished film first, then the repo. On day one, post "a film made from one prompt, prompt in the comments" to test the response, then quote that post on day two with the link. Prepare one line that repost accounts can read out directly, like the "no AI video generator, every frame is drawn by code" pitch. The second wave of stars came from that line. For Chinese coverage, go after accounts like GitHubDaily.
Deadline: this kind of window is measured in weeks. The tutorial-to-open-source-release gap was 6 days; anything launched after 11-13 (10 days after the tutorial) was still at 0 stars by 11-18 (my guess: not just because it was late, but because nobody promoted it).
Where this analysis could be wrong: ① cinematic-skill was still gaining 100+ stars a day after late Nov, even though the author stopped posting, which suggests the product is being used on its own merits and not just riding the trend; ② In Dec, a clone reached several hundred stars through a Chinese-language or vertical version, which means audience-switching is a wider path than I assumed, and this slot deserves an upgrade to B; ③ if Zane Hoyer or another major creator calls out the deleted credit, the backlash could hurt a "repackage someone else's tutorial" play, which would show it has a clear ceiling.
Data and sources
- Star and fork counts: GitHub daily snapshots (11-11 to 11-18), GitHub API (read 11-18 with 739 stars / 112 forks), Trendshift daily stats.
- Commit history: github.com/Moh4696/cinematic-skill, 11-09 07:58 first commit, 15:35 removal of the Zane Hoyer credit.
- X: @exploraX_'s Al Capone film (11-08) and skill launch (11-09), GitHubDaily's Chinese introduction (11-14). Engagement numbers as read on 11-18.
- YouTube: Zane Hoyer, "Claude Edited This Netflix-Level Documentary (Copy Me)" (11-03), Amine AI, "Claude Made This Netflix-Style Documentary From Code (FREE SKILL)" (11-04).
- Instagram / Facebook repost posts: public Reels and posts from 11-12 to 11-16; play counts not retrieved. TikTok not checked.
- Competitor star counts: each project's GitHub page, 11-18. Where it says "inference", it's my inference.
Research notes: competitor table, acquisition evidence, full research report
- Competitors
Name Where Launched Scale How it differs from this one sarpstar/cinecode Open source 2026-11-03 2,677★ / 459 fork Same wave, three.js + headless Chrome, for any kind of film. More than 3 times the scale. It only does narration + archival-photo documentaries. Zane Hoyer's tutorial "Claude Edited This Netflix-Level Documentary" Other 2026-11-03 YouTube 283k views The source of the method, only a video and prompts. It packaged this workflow as an installable skill. Amine AI's "Claude Made This Netflix-Style Documentary From Code (FREE SKILL)" Other 2026-11-04 YouTube 46k views The skill is locked in a Skool community. This one is on GitHub, MIT-licensed, with no signup. Vincentwei1021/anything2explainer, video-talkcraft Open source 2026-08~09 3,386★ / 2,403★ Remotion for motion explainers. This one has the look of an archival-photo documentary with an evidence board. Leonxlnx/cinetic Open source 2026-09-28 1,275★ Remotion for product launch videos and motion pieces. This one leans toward history and documentary. vincentsch/explainroo Open source 2026-09-25 1,359★ Local voiceover + word-timed captions + canvas rendering. Similar route, but no photo search, license logs, or evidence-board style. kaomei/stickman-video-director Open source 2026-08-03 6,808★ Also aimed at faceless YouTube, stick-figure art style Fern-style 3D documentary skills (tmoody1973 and others) Open source From ¥2026-08 Single digits to a few dozen stars Relies on Higgsfield / Seedance to generate video. It costs money, and results aren't repeatable. This one is pure code and uses no video model. bangbuilds/code-doc-film, sanskarprajapati/parallax-doc Open source 2026-10 20★ / 8★ Earlier code-documentary skills that nobody promoted remotion-dev/skills, heygen-com/hyperframes Open source 2026-01 / 03 7,440★ / 86,082★ General-purpose code-to-video base with no documentary workflow or visual style Anthropic (official) Big-company features — — No official video skill exists. Bare Claude with a good prompt can also make these, and what the skill saves you is trial and error. - Where users come from
- Verified Upstream YouTube tutorial wave: Zane Hoyer's tutorial 11-03 283k views; the first committed README says "follows Zane Hoyer's tutorial", deleted at 11-09 15:35 UTC
- Verified Author on X: posted the finished video before the skill: 11-08 posted the Al Capone video, 50k views / 839 bookmarks; 11-09 15:39 quoted it to post the skill, 32k views / 439 likes / 676 bookmarks; next day +92
- Verified Existing audience of the author's "free resources" account: @exploraX_ with 26.9k followers, pinned long post with 6.6M views; same-template repos open-3d-skills 422★, build-ai-agents-free 390★
- Verified Instagram / Facebook repost accounts: 11-12~16 Bibek yadav, StackWhiz, Omar ("comment DOCU"), Confuse Coder ("Comment NOTES") and others posted Reels; Facebook accounts such as nadeemkhancyber reshared. This lines up with the daily peak of +11-14 on 16~100
- Verified Chinese X: GitHubDaily and copycat accounts: GitHubDaily posted at 11-14 01:00 UTC, 11k views / 329 bookmarks, +117 that day; xqliu and 陈发财 followed
- Inferred Aggregator sites and awesome lists: reveneau, reviewramble, hotrepo, skillsllm, claudepluginhub have dedicated pages; awesome-claude-video-skills lists it
Moh4696/cinematic-skill: why it could run
Data as of 2026-11-18. Sources: data cards, our github repo (repo_day / repo_profile), GitHub API (README, commit history), Trendshift, X posts (engagement via fxtwitter API read through Jina), Jina / Exa / Brave search, Instagram and YouTube pages (read via Jina). Stargazer details were not pulled; TikTok was not checked (the TikHub legacy account balance has run out).
One line
This is a skill for Claude Code / Codex / Cursor / Gemini CLI. One sentence has the agent write the script, add narration, find public-domain old photos on Wikimedia Commons, then use Python + Skia to draw frame by frame a "single long take sweeping across a cork evidence wall" documentary short, with stamps, highlighter marks and rolling numbers landing on the word being narrated. It took off mainly by riding an existing wave: 11-03 YouTube editor Zane Hoyer (136k subscribers) posted "Claude Edited This Netflix-Level Documentary (Copy Me)", covering exactly this kind of code-built single-shot evidence-wall documentary, with 280k views. 6 days later, exploraX (26.9k followers) on X, who runs a "free AI resources" account, packaged the method as a one-command, MIT-licensed free skill. It was then spread further by Indian / English Instagram repost accounts ("comment DOCU for the link") and by Chinese GitHubDaily, and grew faster in its second week than in its first.
1. Growth data
- Repo created 2026-11-09 07:58 UTC, pushed 3 commits that day, no updates after. As of 11-18:729 stars and 110 forks (data card); GitHub API read 738 stars and 112 forks that day. 0 open issues, 0 merged PRs.
- Daily star growth (Trendshift trendshift.io/repositories/347385, by UTC day): 11-09 +0, 11-10 +92,11-11 +63, 11-12 +28,11-13 +55, 11-14 +117,11-15 +103,11-16 +131, 11-17 +97.
- Our repo tracking (repo_day, one snapshot per day): 11-11 126 → 11-14 277 → 11-16 492 → 11-18 729; over the last 7 days, +603,11-15 or more per day since 100.
- The curve has two peaks: the first comes after the author's post (11-10~11), the second on 11-14~16 is higher than the first. The second peak came from others' reposts, not the author.
- Forks / stars at about 15%, high for a new repo of this size. One possible reason is that people who arrived through Instagram's "comment a keyword for the link" are more likely to fork, but this was not verified.
- Under 60 days old, so it could not reach GitHub trending; Trendshift also shows it never made the top 25. Star growth had nothing to do with the charts.
2. Where users come from
| Channel | Evidence | Confirmed | |
|---|---|---|---|
| Upstream wave: YouTube tutorial | Zane Hoyer posted "Claude Edited This Netflix-Level Documentary (Copy Me)" on 11-03 (https://www.youtube.com/watch?v=e2Fykz51Nxg), reaching 11-18k views by 283. It shows using Claude with a reference board and a voiceover script, writing code to make a "continuous camera move" documentary. From 11-04 to 11-07, sites such as Frank's World, trydemotion and JAPAN AI NEWS transcribed this video. The same week Amine AI posted "Claude Made This Netflix-Style Documentary From Code (FREE SKILL)" (46k views, 11-04). Its skill is gated behind a Skool community and is not open source. Demand had already been ignited, yet no open-source ready-made skill existed. | Yes | |
| The author explicitly followed this tutorial | The first commit's README and SKILL.md said "The single-camera evidence-wall method follows Zane Hoyer's tutorial on code-built documentaries" with the video link above. The 11-09 15:35 UTC commit removed that credit, and 4 minutes later (15:39) the author posted a promo tweet. The "Inputs" entry in PROMPT.md was also removed. | Yes | |
| Author posts the work on X before the skill | On 11-08 08:45 UTC the author first posted the Al Capone documentary: "created by Opus 5.5, Just a single prompt + ref images, I'll drop the sample prompt in the comments", 574 likes, 839 bookmarks, 50k views (status/2119350502961889592). 11-09 15:39 quoted it to post the skill: "I created /cinematic-skill … 100% free and MIT licensed … install: npx skills add Moh4696/cinematic-skill", 439 likes, 676 bookmarks, 32k views (status/2119817050008748064). The next day (11-10) +92 | Yes | |
| The author's account is itself a "free AI resources" account | @exploraX_ has 26.9k followers and has posted 59.5k times; the pinned long post "20 agentic skills" has 6.6M views. On GitHub, repos of the same kind each have a few hundred stars: 300-free-resource-websites 376, build-ai-agents-free 390, 100-free-open-source-github-repos 241, freecut 288, open-3d-skills 422 (10-31, same README + demo webp + npx skills template). The followers were already looking for "free repo links". | Yes | |
| Instagram / Facebook repost accounts | Since 11-12 a batch of small accounts has reposted it as Reels: Bibek yadav (11-12), StackWhiz (11-12, "One prompt. A whole documentary, built from code"), Omar \ | Building with AI (11-13, "Follow me and comment DOCU and I'll send you the repo link"), Confuse Coder (11-16, Hindi-English "YE FILM CODE HAI!", "Comment NOTES"), plus posts like "The agent that writes your unit tests made this documentary" and "This free GitHub tool could help you build a faceless YouTube channel"; on Facebook, nadeemkhancyber and michaelgarveyjr also reshared it. The pitch is nearly identical: "skip AI video generators that blur and hallucinate", "build a faceless YouTube channel". The timing falls right on the second peak of 11-14~16. | Yes (the posts exist; per-post view counts were not retrieved) |
| Chinese X: GitHubDaily and copycat accounts | GitHubDaily (92k followers) posted a Chinese introduction at 11-14 01:00 UTC, 11,170 views, 191 likes, 329 bookmarks (status/2121407768527827398); +117 that day. Afterwards xqliu (11-15) and 陈发财 (11-16, rewriting GitHubDaily's original) followed, but engagement was small (430 / 1,190 views). 11-10 there is also one more Chinese post (picked up by Trendshift; author not identified). | Yes | |
| Aggregators, ranking sites and awesome lists | reveneau (11-14, "A Claude Code skill draws every frame of a documentary from code"), reviewramble (11-13, score 3.8/5), hotrepo, risingtechrank, githubsignals, skillsllm, skillfoxx, claudepluginhub, learn-skills.dev all have dedicated pages; zhuyansen/awesome-claude-video-skills and its companion agentskillshub.top list it (the list shows 718 stars). There is also an English "6 FREE GitHub repos" list post (bradlantic, only 70 views). | Yes (listed; how many stars it brought is unknown) | |
| HN / Reddit | Our hackernews repo has no related entries; Brave search finds no Reddit posts | Nothing found | |
| Paid promotion | No sign of ads or paid influencers. The Instagram repost accounts grow their own followers through "comment a keyword" prompts; nothing suggests they were paid. | Nothing found |
3. Competitors
The "have an agent write code to make videos" lane is already crowded; narrowed to "code-drawn history / true-story documentaries", several others launched in the same period.
| Competitors | Where | Launched | Scale | How it differs |
|---|---|---|---|---|
| sarpstar/cinecode | Open source | 2026-11-03 | 2,677★ / 459 fork(11-18) | A Turkish author; three.js + headless Chrome rendering, covers any kind of film (shorts, YouTube long-form, ads, music videos), README in four languages. 11-09 gained +872 in a single day, more than 3 times the size of cinematic-skill. cinematic-skill only does "narration + archival photo" documentaries, a narrower scope. |
| Zane Hoyer's tutorial and prompts | YouTube | 2026-11-03 | 283k views | The source of the method; it only provided the video and prompts, not a packaged skill. cinematic-skill turned it into a one-command, reusable tool. |
| Amine AI's "FREE SKILL" | YouTube + Skool | 2026-11-04 | 46k views | The same "code-built documentary" idea, but the skill is gated behind a Skool community; cinematic-skill is on GitHub under MIT, with no sign-up needed. |
| Vincentwei1021/anything2explainer, video-talkcraft, video-shotcraft | Open source | 2026-07~09 | 3,386★ / 2,403★ / 20,981★ | Remotion does black-background motion explainers, voiceover explainers and product promo videos, so it competes on motion design. Cinematic-skill goes for a documentary look: archive photos, an evidence wall and long takes. |
| Leonxlnx/cinetic | Open source | 2026-09-28 | 1,275★ | A Remotion-based product launch / motion video tool, similar in name and positioning, but its focus is history and documentary. |
| vincentsch/explainroo | Open source | 2026-09-25 | 1,359★ | Local Kokoro voiceover + Whisper word-level captions + canvas rendering for explainer videos. The technical route is very close, but it has no archive photo search, licensing records or evidence-wall visual style. |
| kaomei/stickman-video-director | Open source | 2026-08-03 | 6,808★ | Stick-figure, faceless YouTube videos. Same faceless-channel audience, completely different art style. |
| A Fern-style 3D documentary skill (e.g. tmoody1973/claude-skill-fern-documentary) | Open source / tutorial | From ¥2026-08 | Single digits to a few dozen stars | Relies on Higgsfield / Seedance to generate video. That costs money and the output changes every run. Cinematic-skill is pure code and uses no video model. |
| bangbuilds/code-doc-film, sanskarprajapati/parallax-doc | Open source | 2026-10 | 20★ / 8★ | Both are code-rendered documentary shorts (three.js terrain / 2.5D parallax). They came earlier but nobody pushed them, and they have almost no stars. |
| remotion-dev/skills, heygen-com/hyperframes | Open source (official) | 2026-01 / 03 | 7,440★ / 86,082★ | A general-purpose 'make video with code' base with no documentary workflow or visual style |
| Built into Claude / Anthropic | Big-company features | — | — | No official video skill exists. Zane's tutorial shows that plain Claude plus good prompts can already do it. The value of a skill is saving you the trial and error. |
Verdict: it isn't the first to do a "code documentary skill." code-doc-film and parallax-doc came earlier and nobody pushed them, and cinecode the same week was bigger. Its spot in this niche is: a free, open-source, install-and-run version of the single-shot evidence-wall tutorial (by Zane Hoyer) that was the hottest one that week.
4. Why it took off
- Caught a demand already ignited by YouTube. "Using Claude to write code and build Netflix-style documentaries, running faceless YouTube channels" was a hot topic in the November creator scene: Zane Hoyer's episode had 11-03k views, Amine AI's had 11-04 280 46k, and there were plenty of transcript sites. People who watched the tutorials wanted something "you can install right away", but Amine's version required joining Skool. cinematic-skill launched 6 days later, built on Zane's method (the first commit included it, then it was removed).
- The author runs an X account that posts free resources, so they already have a playbook. They post the finished video first (Al Capone, 50k views, 839 bookmarks, "prompt in the replies"), then the next day post a skill-referencing follow-up (32k views, 676 bookmarks). Bookmarks outnumber likes, which means this audience is saving the posts to use later. The same template 10-31 was also used for open-3d-skills and picked up 400 stars.
- The second wave came from repost accounts, not the author. From 11-12 to 11-16, Indian and English Instagram Reels accounts reposted it with a 'comment DOCU / NOTES for the link' hook, selling it as 'no AI video generator needed' and 'make faceless YouTube videos'. Around the same time GitHubDaily published a Chinese write-up (11k views, 329 bookmarks). 11-14 to 16 averaged 100+ stars a day, higher than the author's first two days. The author 11-09 hasn't pushed code updates since, so growth came entirely from outside promotion.
- The product has an easy-to-explain hook. 'No frame is generated video, it's all drawn by code': the same render every time, cheap edits, automatic re-timing when you swap the voice, and licensing recorded for every photo. Almost every repost uses this line. The finished videos show the look at a glance: cork evidence wall, red string, stamps, rolling numbers.
- Installing is frictionless. One
npx skills addcommand, plus plugins for Claude Code / Codex / Gemini CLI and a shell install script. It ships a 90-second Wright Brothers sample video (public-domain photos). On a 16G laptop it renders 1080p in 6 minutes. It runs on Python + Skia, with no Node or headless browser. For repost accounts, 'one prompt, one video' is easy to pitch.
5. Differentiation
- Compared with general code-video bases (Remotion, HyperFrames), it is a full documentary pipeline: scripting, timeline checks, voiceover segmentation, Wikipedia photo search with licensing records, sound-effect synthesis and subtitles. It pauses at three points for your approval: the shot list, the visual style, and a 30-second draft.
- Compared with Fern-style and Higgsfield / Seedance-type generative-video skills: it uses no video model, costs nothing, and the results are repeatable.
- Compared with motion explainer tools (anything2explainer, explainroo, cinetic): its look is archive photos plus an evidence wall, aimed at the history and true-crime faceless-channel crowd.
- Compared with Zane's and Amine's tutorials: free, open source, one-command install.
The differentiation is what makes it repostable: repost accounts want exactly the line 'no AI video, a documentary built by code'. But the look and method weren't original to it; they came from Zane Hoyer. What got it running is the position of 'open-source packaged version of a hot tutorial', plus the author's distribution habits, not a technical moat. Cinecode released a broader version the same week and got more than 3 times its stars, which shows more than one project came out of this wave.
6. How visible it is
It's spreading in niche circles. On X only the author (26.9k followers) and GitHubDaily (92k followers) posted about it, with views in the tens of thousands at most. The rest is Instagram / Facebook repost accounts, Chinese copy-paste accounts, and a pile of auto-generated aggregator pages (reveneau, reviewramble, hotrepo, etc.). No media coverage, no major account wrote about it in depth, and it hasn't made trending. No revenue involved.
7. Copycats
- Same-name reuploads: Taha-hub63/cinematic-skill (11-16), jyotiinfotech2021/cinematic-skill (11-13), Sandeep-Presents/cinematic-skill (11-17), m171kumar/cinematic-skills (11-15). Each has 0 stars. They look like repost-account or user copies (contents not checked one by one).
- Similar new repos: kairo-labs-ai/vintage-documentary (11-17, 'script + voiceover + real archive images → vintage evidence-board', almost the same look), AIViewsLab/documentary-skill (11-14), Swarnavo09/2d-documentary-skill (11-17), raphael321/DocumentaryStyle (11-17, a Remotion archive-documentary version). Each still has 0 stars.
- Bigger projects from the same wave: sarpstar/cinecode (11-03,2.7K★) and zakariaf/cinecode (11-08,31★) aren't copying it. They emerged from the same trend.
8. What others can learn
- Watch viral YouTube creator tutorials and build the open-source one-click version. Tutorial creators only share the video and prompt, or lock the skill behind Skool. Whoever packages it into one line,
npx skills add, within a week captures the people who watched the tutorial. - Post the finished film first, then the tool. Day one, use the film ('made from one prompt, prompt in the replies') to test reaction. Day two, quote that post to share the repo and funnel the same followers straight to GitHub.
- Give repost accounts a ready-made pitch. 'No AI video generator, every frame is drawn by code' is easy for Instagram 'comment a keyword for the link' accounts and Chinese accounts like GitHubDaily to use directly. That's how the second wave happened.
- Reuse one README template: demo webp, one-line install, multi-agent plugins, one complete example. The author used it for both open-3d-skills and cinematic-skill.
- Risk note: copying someone else's tutorial and then removing their credit can hurt your reputation if it gets noticed. If you're borrowing from it, keeping the source attribution is safer.
9. Uncertain
- Play counts for individual Instagram / Facebook reposts weren't collected. Whether there were TikTok reposts wasn't checked (the old TikHub account balance has run out).
- It's not possible to split out how much of the 11-10 day's +92 came from the author's X launch versus that Chinese post by 11-10.
- Stargazer details weren't pulled, so star-farming can't be fully ruled out. Still, both growth peaks line up with specific posts, and the fork ratio is high, which suggests real traffic.
- The reason the author removed Zane Hoyer's credit is unknown. I didn't find whether Zane responded.
- The author hasn't updated since 11-09, so whether growth continues after this wave depends on Zane's next video or new reposts.
Sources
- Data card: /home/ljr/fanout/why/cards/gh_Moh4696__cinematic-skill.json; our github database repo_day / repo_profile (11-18), hackernews database
- GitHub: github.com/Moh4696/cinematic-skill (README, and the commit 2026-11-09 07:58 / 08:01 / 15:35,15:35 that removed Zane Hoyer's credit); author profile github.com/Moh4696; github.com/Moh4696/open-3d-skills
- Trendshift:trendshift.io/repositories/347385
- X: x.com/exploraX_/status/2119350502961889592 (Al Capone film); x.com/exploraX_/status/2119817050008748064 (skill launch); x.com/exploraX_; x.com/GitHub_Daily/status/2121407768527827398; x.com/xqliu/status/2121832566634467523; x.com/Chenfacai0/status/2122474798555005142; x.com/bradlantic/status/2120339810564845924
- YouTube: www.youtube.com/watch?v=e2Fykz51Nxg (Zane Hoyer, 11-03); www.youtube.com/watch?v=MD2WHCG2oxQ (Amine AI, 11-04, skill behind Skool); www.youtube.com/shorts/WJUrgIYuSY8 (ConfuseCoder23 repost); moderncreator.app/2026-11-03-zane-hoyer-claude-edited-this-n…
- Instagram: www.instagram.com/reel/DfrpJlLBjBA/ (Omar, 11-13, 'comment DOCU'); www.instagram.com/reel/Dfp8jBVhxuk/ (Bibek yadav, 11-12); www.instagram.com/reel/DfqjBw8hIdy/ (StackWhiz, 11-12); www.instagram.com/reel/DfzPGeLsxfl/ (Confuse Coder, 11-16); www.instagram.com/reel/DfuoPMDiYF6/
- Facebook:www.facebook.com/nadeemkhancyber/videos/stop-using-ai-video-… ;www.facebook.com/michaelgarveyjr/posts/this-free-github-tool…
- Aggregator pages: reveneau.com/ainews/cinematic-skill-claude-skia-ffmpeg-no-vi…; www.reviewramble.com/github-repos/cinematic-skill; hotrepo.ai/Moh4696/cinematic-skill; agentskillshub.top/best/claude-video-skills/
- Competitors: github.com/sarpstar/cinecode ; github.com/Leonxlnx/cinetic ; github.com/vincentsch/explainroo ; github.com/Vincentwei1021/anything2explainer ; github.com/kaomei/stickman-video-director ; github.com/bangbuilds/code-doc-film ; github.com/sanskarprajapati/parallax-doc ; github.com/kairo-labs-ai/vintage-documentary