Case study

Giga Production Studio

An AI production crew in a single app. Researcher, writers' room, voice, director, art department, editor. AI does the work, you keep the taste.

RoleConceived, designed, built end to end
Year2026
StatusFeature-complete, v1.0
StackPython · FastAPI · React · Electron
The Timeline: a real multi-track editor inside the studio, with shots, voiceover, and auto-placed sound effects. (Interactive UI recreation.)
The problem

Fast and soulless, or great and unaffordable.

The market had split into two bad options. One-click AI video is fast, but generic, soulless, and uneditable, the "AI slop" audiences are already tuning out. The traditional studio is beautiful and on-brand, but slow, expensive, and it needs a whole team: writer, researcher, voice artist, animator, editor.

No tool gave you a real team's output and a real director's control, at the speed of AI. That gap is the whole reason this exists.

What I built

The whole studio, end to end.

I conceived, designed, and directed the build: the multi-agent architecture, the model-agnostic LLM gateway, the production pipeline and its approval gates, the in-app non-linear editor, the self-improving "Brain" memory, and the entire React studio UI.

I am also the first user. I run the channel the studio was built to produce, so every feature was pressure-tested against real work, not a demo.

How it works

One guided pipeline, with you in the chair.

01 Idea and inspiration 02 Research the topic 03 Script writing 04 Voice-over generation 05 Scene direction 06 Timeline editing 07 Channel profile

✓ you approve at every handoff. AI does the labor; you keep every creative decision.

Inside the studio

A walk through the whole pipeline.

Each screen below is the actual studio UI, embedded live. Try them.

The Scout · Inspiration

Ideas in your angle, ranked.

Turns a topic, link, or stray thought into ranked video ideas in your voice, checked against what actually performs.

The Researcher · Research

Reads the sources in full.

Reads real articles, transcripts, and threads end to end and builds a fact-grounded brief with SEO signals. Facts, not hallucinations.

The Writers' Room · Writing

Three writers, one final script.

Three AI writers draft in parallel, each on a different model; a senior writer synthesizes one final script in your tone and target length.

The Voice Artist · Voiceover

Studio narration, section by section.

Studio-quality narration with different voices per character, reviewable and re-takeable, timed to the script.

The Director · Direction

A shot for every single line.

The magic step. The director breaks the finished script down sentence by sentence and proposes the right visual for each: an image, a pose, a motion graphic, a title. You override any choice; approved sections lock and survive re-runs.

The Publicist · Profile

Packaged for publishing.

Packages the finished video for its platform: title candidates with rationale, description, tags, and thumbnail concepts.

The Brain · Memory

A memory that learns your taste.

Every edit and decision is captured as a lesson and woven into a living memory, visualized as a knowledge galaxy, and re-applied to future videos. The studio you use in month three is sharper and more you than the one you started with.

Tech stack

Under the hood.

Backend
Python 3.11 · FastAPI + WebSocket (real-time agent streaming)
Frontend
React + TypeScript + Tailwind · Vite
Launcher
Electron app: installs the studio and every dependency, self-updates from GitHub, starts and stops the backend and frontend
AI orchestration
Multi-agent team via a LiteLLM gateway, model-agnostic (Claude, GPT-4o, Gemini, local via Ollama)
Image generation
ComfyUI · channel-style LoRA · automatic background cut-outs
Voiceover
AI text-to-speech (per-character voices) · Whisper for timing
Memory (the Brain)
Markdown lesson files · captures the difference between generated output and the user's edit · re-applied on later runs
Motion graphics
Custom HTML/GSAP engine rendered to alpha video
Status and outcome

Shipped, and it gets better with use.

Feature-complete at v1.0, the full pipeline works end to end, from an idea to a finished, edited video. It is self-improving: the Brain captures lessons from every project and re-applies them. The studio itself is a web app, wrapped in an Electron launcher that installs like any normal application. The launcher sets up every dependency, pulls new versions from GitHub, and starts and stops the servers, so a Python and Node stack installs in one click for someone who should never have to think about either. Source is a private repository, available on request.

Want something like this built?

I build AI systems and automations that do real work, with a human in control.