No camera, no mic, no editing software, no stock footage. You type one sentence — “make a short about the multiply-by-11 trick” — and Claude Code plans the video, writes the motion as code, checks the frames itself, records the voice-over with word-by-word synced captions, adds the SFX, and hands you a 1080×1920 MP4 ready to post.
The first 50 copies at $49 instead of $79 — then the regular price returns.
These are final clips from the very system you will receive — not one of them was edited by hand. Every video has a script.md and beats.json you can open and rebuild, or ask for a version on your own topic.
A silent product, different from the shorts: a set of Instagram slides at 4:5 (1080×1350), each slide an independent Composition over a shared frame, exported as a set of PNGs with one command. This is the complete “Claude Cowork” carousel — 9 slides in the channel’s brand, and the first slide also comes out as an animated MP4.
make a 9-slide Instagram carousel about Claude Coworkscript.md + slides.json, and builds a Composition for every slidecarousel-export.mjs Carousel1 → a PNG per slideEvery short goes through the same steps — nothing is rendered and nothing is billed before you see the plan and approve it.
Claude fact-checks first, then writes script.md: a beat table HOOK → SETUP → QUIZ → REVEAL → TWIST → LOOP, with what is on screen and the voice-over line for every beat.
It writes a Remotion Composition (TSX) using ready-made kits: chess, math, algorithms, maps, spreadsheets, terminal, browser… every movement tied to a word in the voice-over.
Before any full render, it renders phone-sized PNGs and reads them itself: clipped text, overlapping layers, text hidden under the platform UI — and fixes them first.
ElevenLabs, one line at a time (redo a single line without paying for the rest), and it writes the real timing of every word into beats.json.
It re-renders so the highlight lands on exactly the spoken word — Arabic RTL or English.
A sound-cue sheet sfx-plan.json from a library that is generated once and reused in every video. A draft mix — your ear is the final gate.
Description, hashtags, title and first comment for every platform — ready to paste. Publishing is on you, by design.
You want to post daily but hate being on cameraEverything here is faceless — the motion, the captions and the voice are the stars.
One short costs you a full day of editingHere it takes one conversation. Rendering takes minutes, and a change is one sentence.
You tried “AI video” tools and got generic, boring resultsHere every movement is coded for your topic — not stock templates.
Synced captions are always late or wrongTiming comes from the word that was actually spoken, not from an estimate.
Nothing you found works properly in Arabic12 ready Arabic shorts, RTL layout, a Modern Standard Arabic voice, and technical terms kept in English as they are.
You are afraid to spend on APIs without knowing where the money wentEvery paid tool has a --dry-run, and Claude states the estimated cost before any paid call.
One production line, and five ways to get the pixels:
Each one in its own folder: the plan, the beats contract, the SFX sheet, and the exact Composition that renders it. Open it, tweak it, or ask for a version on your own topic.
14 text-only skills that decide what to make and why — they stop at the plan, and production takes it from there:
Claude asks you which identity to use before every video. Edit one file, and every future short follows it.
A web UI on your own machine with zero dependencies: every project with its video, script and beat timeline, a form for every tool with a live preview of the command, and the render scripts as buttons.
--dry-runVoice, SFX, music, images, clips, collage cutouts, web-page captures, safe-zone checks, Google Drive upload…
Not a translation. A guide I wrote from my own experience on Windows 11: what to install and in what order, where the keys go, how to request your first short, what happens behind the scenes, the craft rules that keep 40 seconds gripping, what everything costs, and the errors that will stop you — with their fixes. + Arabic guides for daily use and for the dashboard.
A skill is a written playbook — a Markdown file at .claude/skills/<name>/SKILL.md — that tells Claude Code how to do one job well. You type the slash command, or describe what you want in your own words and Claude picks the right skill. Every card here: what it does, how it works, a command you can paste as is, and something real in the project it produced. All 26 skills come ready and explained in the guide.
/make-shortThe default pipeline for any short: from a topic to a complete video with motion and voice.
How it works: Topic → script.md + beats.json → TSX beats over one fixed board → frame-by-frame checks → ElevenLabs voice with word-by-word synced captions → a draft SFX mix. ~40 seconds, looping back to frame 0.
make a short about the Monty Hall problem/make-voxVox-style paper collage with a documentary rhythm: maps, archival cutouts, a camera traveling between scenes.
How it works: Every scene is broken into layers (Gemini images + cutout.py, an HTML→PNG page, or SVG directly) and composed on a CollageBoard camera in Remotion.
make a vox-style short about how coffee conquered the world/make-ai-shortThe generative pipeline: video clips from an AI model with a locked, consistent character.
How it works: One character sheet in character.json; every shot is image-to-video from a frame that contains the character (Seedance 1.5 Pro on fal); the loop shot uses frame 0 as its end frame. It states the estimated cost before every paid call.
make an AI video short with blue-man: he watches the same bus pass three times/avatar-reelYour own talking clip → a “two-storey” reel with evidence panels above you.
How it works: Drop avatar.mp4 + script.txt in a folder. You get: TSX panels on top, your avatar in a rounded card below, word-by-word synced captions in between, real screen recordings for the “proof” beat, a comments CTA, and a cover.
/avatar-reel/video-ad-editorA talk-to-camera video → a finished vertical ad, with no editing software.
How it works: Cuts the silences, adds word-timed Arabic synced captions, draws coded motion graphics that visualize what is being said (numbers, comparisons, charts), an end card, −14 LUFS loudness normalization, and an .srt file.
make an ad out of this video/prompt-showcaseA silent reel in the channel’s brand that shows an AI-made result + the prompt that produced it.
How it works: The result inside a frame, the reference image beside it, and the prompt scrolling automatically, with a CTA that invites comments. Music is added after the render.
make a prompt reel for this Higgsfield video, prompt in prompt.txt/vidtsx-2d-generatorThe core TSX rules every pipeline follows — so Remotion never falls apart.
How it works: Frame-based motion only, increasing ranges, Easing.bezier, no useState. Use it directly for a single animated clip or a title card.
generate a 5-second animated title card that says "Season 2"Every Composition under remotion/src/shots/ follows it /suggest-sfxThe sound-effects stage for any short.
How it works: Reads the beats and word timings, suggests cues from the shared media/library/sfx library, generates what is missing via ElevenLabs, writes sfx-plan.json, and renders a draft mix for you to hear and approve.
suggest sfx for short-5shorts/short-1-chess/sfx-plan.json — 22 cues on a chessboard /diyaaMy channel’s fixed profile: the Diyaa Albouzan | Tech identity.
How it works: Synced captions only with no voice-over, the bottom 30% of the frame reserved for a talking head you add later, and a Reels cover with every video. It overrides any skill that conflicts with it. Copy it and rename it to write your own profile.
/diyaa make a short about Claude skills/insider-styleThe AI-tool promo reel style, distilled from breaking down 10 successful reels.
How it works: Word-by-word kinetic text, floating windows, slide stacks, real footage, a comment-gated CTA. Run it with any pipeline.
make an insider-style reel about the new MCP feature/post-copyAll the publishing copy for every platform, ready to paste.
How it works: Description + hashtags + titles + hidden tags + first comment + alt text for Shorts, Reels, TikTok and Facebook — built on fresh algorithm research, in Jordanian dialect, saved to output/captions-and-tags.txt.
write the caption and hashtags for short-26 for every platformshorts/short-26-token-graph-paper/output/captions-and-tags.txt /algo-researchHow each platform ranks short-form content this month.
How it works: Searches the web and writes a dated brief with its sources in research/. /post-copy reads it automatically if the latest brief is older than ~30 days.
research the algorithmresearch/algo-brief-YYYY-MM-DD.md /caption-and-hashtagsThe lighter version of /post-copy for a single video.
How it works: Description + hashtags + first comment — text only.
write a caption and hashtags for the coffee shortText only /content-factoryThe orchestrator: runs the whole growth layer in order.
How it works: Studies a winning format → a hook and script in your voice → hands off to a production skill → publishing copy.
run the content factory for the AI-tools niche/content-researchWhat is actually working in a niche or for a specific creator, and why.
How it works: Pulls the posts, finds the outliers that won, and explains why they did. Needs a scraping tool.
what's working for the reference account right now/going-viralThe strategy layer before any video.
How it works: Pick the goal → the emotion that drives it → the hook → the funnel.
why aren't my reels getting saves?/viral-hook-writer10 ranked hooks for the first 1–3 seconds.
How it works: From a bare topic to ten opening lines ranked by strength; pick one and keep going.
10 hooks for a short on compound interest/hook-miningRewrites hooks you already have performance data for.
How it works: Give it a CSV of results; it keeps the psychological structure of the winners and swaps the power words.
mine hooks from hooks.csv/reel-analyzerA frame-by-frame breakdown of any reel you send it.
How it works: A link or a file → a written teardown: the rhythm, the shots, the hook, and why it worked.
break down this reel: <link>/reel-scripterAn idea or a hook → a full 30–45-second script.
How it works: Writes the voice-over + on-screen text → a script.md ready for the production pipeline.
turn hook #3 into a 40s script/reel-builderA breakdown or an idea → a beat-by-beat storyboard.
How it works: On-screen text, scene, beat → beats.json, the single hand-off point into the production pipeline.
build the reel from that breakdown/content-repurposerAn article / podcast / long video → a batch of shorts.
How it works: Extracts multiple ideas and scripts from a single source.
turn this podcast episode into 5 reels/content-calendarA batch of ideas → a two-week publishing plan.
How it works: Arranges the ideas with hooks and formats for each day.
plan my next two weeks from IDEAS.md/comment-responderA guide for turning a comment into a direct message.
How it works: Public reply → send the resource → log the lead. Nothing here posts or messages anyone — you do it by hand.
/agent-reachPulls content from 13 platforms for research (X, Reddit, YouTube, LinkedIn…).
How it works: Needs its own tools; agent-reach doctor shows what is working.
agent-reach doctor/ai-brainLong-term memory across sessions in Obsidian.
How it works: “wrap up” saves a session note, and the next session reads it. Needs a Filesystem MCP pointed at the vault.
wrap up| Tool | What it does | Example |
|---|---|---|
gen_voice.py | ElevenLabs voice-over per line; writes the real word timings into beats.json | python tools/gen_voice.py --beats shorts/short-5-prob/beats.json --mux remotion/out/Short5Monty.mp4 |
gen_sfx.py | Expands the media/library/sfx effects library from palette.json and updates the catalog | python tools/gen_sfx.py --only whoosh-soft,pop-reveal --dry-run |
gen_music.py | Instrumental music beds, loudness-normalized | python tools/gen_music.py --only ambient-pad --length 330000 |
mix_sfx.py | A draft SFX mix from sfx-plan.json, ducked under the voice | python tools/mix_sfx.py shorts/short-2-math/sfx-plan.json --print |
mix_music.py | A music bed under a base video, ducked hard under the speech | python tools/mix_music.py --all --base out.mp4 |
gen_image.py | Gemini still images (pro / fast / lite) with a prompt sidecar file | python tools/gen_image.py --prompt "…" --aspect 9:16 --out x.png |
gen_clip.py | A video clip from fal with any model; extras via --set | python tools/gen_clip.py --model fal-ai/bytedance/seedance/v1.5/pro/image-to-video --prompt "…" |
bakeoff_clip.py | The same prompt across several models, a labeled grid with costs | python tools/bakeoff_clip.py --prompt "…" --models seedance,kling,veo-lite |
cutout.py | An image on a white background → a transparent collage cutout | python tools/cutout.py in.png out.png --method rembg |
capture_web.py | Playwright captures of a real web page / real VS Code | python tools/capture_web.py --spec spec.json |
check_safe_zone.py | Verifies the reserved avatar band is empty, frame by frame | python tools/check_safe_zone.py --comp Short21EvalAware |
gen_chords.py | Generates the piano audio for short-9 | python tools/gen_chords.py --dry-run |
upload_drive.py | Mirrors output/ + cover/ to Google Drive via rclone | python tools/upload_drive.py --project short-2-math |
Let me be clear from the start: the price is for the system and the guide. The tools it runs on have their own subscriptions, which you pay directly to their owners.
The only key you need to start: ElevenLabs. It gives you the complete default pipeline with voice, synced captions, SFX and music. Lines are cached by text hash, so you never pay twice when you re-render.
I work with AI tools every day and teach them in Arabic to thousands of followers. I built this system on top of Hassan’s open-source project (MIT) and kept developing it: the Arabic version of every short, the growth layer, the visual identities, the avatar format, the carousel, and the dashboard — and it is the same system I use for my own channel. The origin and attribution are fully documented inside the guide.
Tried the system and produced your first short? Send me your review (and the video, if you like) — it is published here under your first name.
Send your reviewbeats.json when something is off, and the guide shows you exactly what it looks like.If you follow the guide and cannot produce your first short, message me within 7 days: I will either help you get it running, or refund you. No questions asked.
“The craft lives in files you read and edit, not in a model’s mood. Change the rule, and every future short follows it.”Type one sentence. Get a finished, looping 1080×1920 short.
The complete system, 27 ready shorts, 12 in Arabic, and the guide that walks you from installation to your first video — download it now and produce your first short today.
