FilmBam
One brief. A coordinated production.
Original frontend design. Sample data.
The problem.
Making an AI film requires more than generating clips. References, continuity, sound, cost and delivery need to work together.
What I built.
An AI studio that plans shots, manages generation, adds sound and checks the finished film.
- My role
- Product concept, production workflow and AI-agent orchestration
- For
- Creators and small production teams
How it connects.
A production workflow with budget approval, parallel picture and sound work, and selective retakes.
Read the connections as text
- Brief → Production plan
- Production plan → Budget approval
- Budget approval → References: Approved scope
- References → Generate shots
- Budget approval → Voice + sound
- Generate shots → Assemble + finish
- Voice + sound → Assemble + finish
- Assemble + finish → Quality checks
- Quality checks → Generate shots: Selective retake
- Quality checks → Delivery: Accepted output
Built in stages.
The V2 web interface, queue, private sessions and media-delivery code are built.
Public V2 deployment, payments and a complete unattended new-order run remain to be validated.
Inside the product.
Open a module to see its capabilities,
decisions and boundaries.
Brief & planningTranslate an idea into the format, references and shots it needs.
Brief input
Film or storyboard from a short written brief
Output formats
Landscape, portrait and square outputs
Shot planning
Structured shot plans with selected duration and quality
Character references
Character references reused across shots to guide continuity
Generation routes
Generation routes selected for the shot and production format
Generation & continuityCoordinate the production instead of treating each shot as an isolated prompt.
Reference-guided generation
Reference-guided image and video generation
Supported continuity routes
Native multi-shot generation and frame chaining where the selected route supports them
Fallbacks
Alternative model routes when a generation request fails
Selective retakes
Selective shot regeneration instead of repeating the full production
Stage checks
Stage checks before moving to the next production step
Sound & finishingBring generated material into a coherent, deliverable sequence.
Audio generation
Voice, music and sound-effect generation
Audio fallbacks
Audio-provider fallbacks when a route is unavailable
Assembly & export
Shot assembly and final export through FFmpeg
Loudness & upscale
Loudness normalisation and supported video upscaling
Quality checks
Technical checks for black frames, freezes, duration and loudness; visual review rubric
Cost controlsKnow the allowance before production and the actual spend afterwards.
Budget approval
Cost estimation and budget approval before paid generation
Production caps
Production caps and per-order cost records
Fallback costs
Recorded fallback and retake expenditure
Reservation ledger
V2 reservation ledger, daily allowances and monthly spending limits
Model research
A documented model-research workflow that proposes changes for review
Studio & deliveryThe built web product connects a simple interface to the production engine.
Production interface
FilmBam web interface with brief, format, duration, quality and order history
Private sessions
Private user sessions and an owner view
Queue recovery
V2 production queue, runner claims and recovery for abandoned jobs
Media access
Per-user media delivery with expiring access links
Owner cost view
Owner ledger for reserved and actual spend
Tools & systems.
Python / FFmpeg / TypeScript / Hono / Cloudflare Workers / D1 / R2 / fal.ai / ElevenLabs / GitHub Actions
Where the project stands.
- The product is a working prototype; a public V2 launch is not established.
- The seven films in Rafael’s directing portfolio are independent work, not Video Bam outputs.