ShotSmith

Project Description https://github.com/Zenieverse/ShotSmith-AI ShotSmith AI is a professional cinematography tool built on Bria FIBO’s JSON-native image generation, designed to replace prompt-based guesswork with deterministic visual control. Instead of relying on fragile text prompts, ShotSmith expresses creative intent through structured JSON that explicitly defines camera, lighting, composition, and color parameters. Users can input a script, storyboard, or natural-language scene description. ShotSmith’s agentic system translates this intent into schema-validated FIBO JSON, specifying shot type, lens, field of view, camera angle, lighting configuration, and HDR color palette. Because the same JSON always produces the same result, outputs are fully reproducible and suitable for real production workflows. The application exposes professional controls through an interactive UI, allowing users to adjust camera lenses, FOV, lighting ratios, and color settings with predictable results. Multi-shot timelines support continuity locking across scenes, enabling consistent cinematography over an entire sequence. ShotSmith also supports HDR and 16-bit color workflows using ACES, with export formats designed for compositing and post-production. By combining FIBO’s structured generation with an agentic architecture and a production-focused user experience, ShotSmith AI demonstrates how visual AI can function as reliable creative infrastructure rather than experimental tooling.

Inspiration Generative image models are powerful, but in real production they break down fast. Prompt-based workflows are non-deterministic, hard to debug, and impossible to version. Cinematographers don’t think in adjectives — they think in lenses, angles, lighting ratios, and color spaces. FIBO inspired us because it flips the paradigm: visual intent becomes structured, controllable, and reliable. ShotSmith was born from the idea that AI image generation should behave more like a render engine than a slot machine. What it does ShotSmith AI turns creative intent into deterministic cinematography using FIBO’s JSON-native generation. Users can input a script, scene, or shot description, and ShotSmith automatically generates structured JSON that explicitly defines: Camera type, lens, FOV, and angle Lighting setup and intensity Composition rules HDR / 16-bit color palettes The same JSON always produces the same framing and lighting behavior. ShotSmith also supports multi-shot timelines, continuity locking, and HDR-ready outputs for professional pipelines. How we built it ShotSmith AI is built as an agentic, JSON-first system: A Narrative Agent extracts emotional and pacing cues A Cinematography Agent selects shot types, lenses, and FOV A Lighting Agent creates physically plausible lighting setups A Color Agent defines HDR palettes using ACES workflows A FIBO Render Agent validates schemas and triggers generation All agents communicate strictly via JSON, with FIBO acting as a deterministic visual renderer. The UI exposes professional controls mapped directly to FIBO parameters, making changes predictable and repeatable. Challenges we ran into Translating creative language into precise camera and lighting parameters without losing artistic intent Designing agent outputs that are expressive yet fully schema-validated Maintaining visual continuity across multiple shots Making the UI feel like a professional tool rather than a chat interface Each challenge reinforced why structured control is essential for real workflows. Accomplishments that we're proud of A fully deterministic image generation pipeline using FIBO Multi-agent orchestration that communicates only via JSON Camera, lighting, and color controls that behave predictably every time HDR-ready, 16-bit outputs suitable for professional post-production A tool that feels credible for studios, agencies, and production teams What we learned Prompt engineering does not scale to production JSON is not a limitation — it’s creative freedom with guarantees Determinism is the missing ingredient for enterprise visual AI FIBO enables workflows that were previously impossible with generative models Most importantly, we learned that artists don’t want “better prompts” — they want control they can trust. What's next for ShotSmith ComfyUI node integration for pipeline-friendly workflows Shot-to-shot continuity constraints (character, lighting, lens locks) Timeline-based animation and motion planning Deeper Nuke / Unreal / DCC interoperability Team collaboration and versioned visual assets ShotSmith is just the beginning of treating generative AI as cinematography infrastructure, not experimentation.

Built With

• all