Case study · AI product

An AI studio that writes, voices and renders a branded video for nothing per video

My own app, built alone. One brief goes in; a finished post comes out in English, Arabic or Russian, with three options to choose from. A 26-second preview renders in about 50 seconds on my laptop, and no step costs a cent: every paid API has a free or local fallback.

  • Role: sole developer
  • Stack: Claude Code, Python, FastAPI, React + TypeScript, Remotion, Docker
  • Proof: one real delivered video, 24 seconds, 9:16, plus a 4:5 cut, no human edits
  • Code: github.com/elkushi/ai-video-image-generator
The app's live view while it makes a post: a cartoon factory floor with a copywriter, a designer, a validator, the renderer and a QA director, a timeline with real durations and a live feed of what each agent says.
The live view while the app makes an image post. Each station is a Claude agent or a Python tool, and the timeline shows real durations.
What this proves: I can build an AI system that does a marketing team's work end to end and checks its own output before a person ever sees it.
0 €paid APIs: every AI step has a free online option and a local fallback
~50 sto render a 26-second preview on a normal laptop
3 optionsper brief, each with its own design and its own copy
3 languagesEnglish, Arabic and Russian, with Arabic captions right to left

The idea

I wanted short videos that stay on brand, cost nothing per video, and still get made when an online AI service is down or out of quota.

How it works

Claude makes the decisions. Python scripts do the work.

  1. A strategist agent picks the audience, angle, hook and call to action.
  2. A scriptwriter agent writes the storyboard: voiceover, on-screen text and scene types.
  3. The storyboard is checked against a JSON schema before anything slow runs.
  4. Python tools make the voice, fetch or generate the visuals and pick music.
  5. Remotion, a React video framework, renders every aspect ratio.
  6. QA checks length, audio levels and black frames, and transcribes the voice with Whisper to compare it with the script.
  7. A reviewer agent reads the QA report and returns SHIP or a fix list. At most two fix rounds, then the run stops.

The same pipeline makes image posts and carousels. Videos come out in English, Arabic or Russian. Arabic captions run right to left.

Diagram: brief, agents, schema check, voice and visuals, Remotion render, QA. QA can send a fix list back, at most twice.
One video, start to finish. Blue boxes are checks that can stop the run.

Three options, you pick one

For an image post, the app makes three different angles, each with its own design and its own copy for LinkedIn and Instagram. You pick one, edit the text if you want, and it delivers the slides, a carousel PDF and the post copy.

Three post options for a demo brand, side by side: a single before-and-after image, a four-slide carousel and a single statement card, each with its own LinkedIn copy and hashtags.
Three options for one brief: a single image, a four-slide carousel and a statement card. Made for a fictional demo brand.
The Create page: pick a brand, describe the video in plain words, then choose the goal, platforms, language, length and style.
Start: pick a brand and describe what you want in plain words.
A delivered carousel in the gallery: the first slide, thumbnails of all four slides, the LinkedIn post copy with a Copy button and a Carousel PDF download.
Finish: slides, a carousel PDF and ready-to-post copy.
Real output · 24 seconds · 9:16

A finished video, made by the app

One brief for a fictional notebook brand. The agents wrote the script, the voice is a free neural voice, Remotion rendered it, and the QA step passed it. No human edits.

The same run also delivered a 4:5 version for the LinkedIn and Instagram feeds.

Constraints I set

  • No paid APIs. Every AI step has a free online option and a local fallback: Piper voices, local Stable Diffusion, a built-in mascot instead of a talking head.
  • Runs on a normal laptop. On mine, a 26-second preview renders in about 50 seconds. A local AI image takes about 9 minutes on the CPU, so online options come first.
  • Nothing ships unchecked. The schema check and the QA step can stop a run.

The web app

A FastAPI backend runs Claude Code in headless mode, and a React + TypeScript front end streams each agent and tool live as it works. It runs in Docker, with tests and GitHub Actions CI.

Security

There is no hosted version. You download it from GitHub and run it on your own machine, so it is locked down for that:

  • It holds none of my keys or accounts. Each copy uses its owner's Claude Code login. The full git history is scanned for secrets, and the scan is clean.
  • It listens on 127.0.0.1 only, so other machines on the same network can't open it.
  • The agent can only use an allowlist of tools: the project's own scripts, ffmpeg and a few file commands. Other integrations are blocked.
  • A hook checks every shell command and file path before it runs. In a live prompt-injection test, Claude tried 30 routes to get a planted fake key out of .env and got nothing.
  • Every uploaded file must sit inside the uploads folder. Tests try /etc/passwd, ../../etc/passwd and .env.
  • The container runs as its own user, not root. Keys live in .env, outside git.

← All work