StudioMunich
AI catalog, campaign and film production with consent built in
For: Fashion and luxury brands, e-commerce teams, design studios and agencies that commission catalog and campaign production, plus modelling agencies that license talent through VAULT.
StudioMunich replaces much of the physical studio shoot with a Director-controlled AI workflow. A brief goes in, the brand's visual look is locked once, and the platform produces catalog images, colorway variations, channel crops and storyboards that are reviewed by people before anything is delivered.
It is built for fashion studios and brands. Creative Directors and catalog producers brief jobs, cast models, trigger generation and approve variations in the production app. Brand-side buyers receive a review link they can open without an account and sign off from there.
The platform is organised into four wings: VISUAL for image and campaign work, VAULT for licensed human and AI talent, VID for film and motion, and VIBE for engineering and products. Work is priced in credits, and every journey starts with 200 free credits.
The problem it addresses is twofold: the cost and time of conventional shoots, and the legal risk of un-cleared AI likenesses. In StudioMunich, consent is a gate in the pipeline: without a verified licence token, the system will not generate or deliver an image of that talent.
200
free credits to start every journey
50 cr
per catalog unit (Catalog Accelerator)
~38s
median full-pipeline latency per 4K asset
3
maximum automatic correction cycles before human review
What it does
Catalog Accelerator: brief in, approved catalog imagery out, with a provenance record you can recompute
Dress a garment on a model from a flat-lay, then publish to Shopify
Semantic colorway edits that recolour the garment while holding model, pose and scene
Seamless pattern tiles generated from a prompt and brand hex colours, previewed in repeat
Storyboard to film: one sentence becomes a script, character cards and a 12-panel storyboard
MIRE presets that lock a brand's 7-layer visual DNA (environment, light, colour, props, pose, composition, mood)
VAULT talent roster with consent-gated licensing and a royalty event recorded on every licensed use
Hash-verified CO2e certificate per job, with a public verification route
How we architected it
- Next.js 15
- React
- TypeScript
- Supabase
- Vercel
- fal.ai (FLUX, birefnet, Seedream, LTX)
- LangGraph (Python)
- Fly.io
- Railway
- FFmpeg
- Anthropic Claude
- OpenAI
- Deepgram
- Pinecone
- Stripe
- Resend
- Three.js
- ONNX Runtime Web
- Model Context Protocol (MCP)
- 01
Next.js production app and brand site on Vercel, with Supabase (Postgres, Storage, Auth) in the EU; tenancy runs studios to clients to production jobs to scene units, scoped by row-level security
- 02
Brief intake uses Claude vision to parse PDF and image briefs into structured scene units (article codes, model types, sample sizes)
- 03
Generation pipeline on fal.ai: FLUX models for images, birefnet background removal, PuLID/IP-Adapter identity conditioning, a virtual try-on engine for re-dressing, and LTX for video
- 04
A LangGraph orchestration service (Python, Fly.io Frankfurt) runs brief parser, DXF ingestor, FLUX renderer, critic and ledger binder, with a correction loop capped at three cycles before handing to a human
- 05
Human-in-the-loop gates: automatic face-consistency scoring escalates low scores, and art-director approval is required before delivery
- 06
Rights rail: casting verifies a talent licence token before generation; approval fires a royalty event, with failed fires queued and retried by a Vercel cron every minute
- 07
Deliverables are SHA-256 hashed at approval and carry embedded XMP provenance (consent token, talent ID, scope, royalty event ID)
- 08
A self-hosted FFmpeg render worker on Railway assembles montage films; an MCP server exposes e-commerce campaign and certificate-verification tools to developers
Want something built like this?
Every build runs the same five gates: spec, architect, build, review, ship. It starts with a 48-hour audit.