Inspiration
Industrial design studios spend significant time and budget on physical prototyping and professional photography for product visualization. The gap between AI image generation capabilities and professional production requirements motivated development of a system that produces render-quality outputs with compositor-level control while integrating into existing professional workflows through Nuke and ComfyUI.
What it does
The system generates photorealistic product designs from text descriptions and structured parameters, processes outputs through HDR 16-bit pipelines, enables iterative refinement via ComfyUI nodes, executes parallel variant generation through multiple AI agents and exports Nuke-ready scripts with embedded JSON parameters for professional compositing workflows.
Features,
Intuitive dashboard for designers to tweak visual parameters and preview in real-time, with JSON automatically updated.
Highly demanded for automotive, consumer electronics and hardware design pipelines.
Reduces design iteration cycles and produces production-quality visuals without expensive photoshoots.
Demonstrates scalable AI workflows.
Highlights enterprise-grade visual AI capabilities that integrate JSON-native automation, agentic pipelines and professional post-production workflows.
Pushes FIBO’s capabilities in ways that define the future of generative AI pipelines.
multiple agents generate assets in parallel, saving time for large studios.
Generate multiple iterations → Nuke/ComfyUI visual comparison panel → AI agent selects the best composition.
Drag-and-drop storyboard → AI agent translates → preview panel with HDR and 16-bit fidelity → final Nuke-ready nodes.
Professional tool integration: allow compositors to tweak parameters in Nuke while preserving JSON-native control.
Each aspect of the shot (lighting, angle, composition, palette) is independently controllable, giving fine-grained artistic freedom.
How we built it
The backend implements Flask REST API orchestrating asynchronous calls to FIBO models through FAL.AI, Bria.ai and Replicate APIs. Agent orchestration uses asyncio with semaphore-based rate limiting for parallel execution. Image processing employs Pillow and OpenCV for HDR conversion and 16-bit color depth transformation. Frontend uses vanilla JavaScript with real-time JSON preview. Parameter validation relies on Pydantic schemas. The system stores generated assets with JSON metadata in local file system directories: generated_designs, refined_designs, nuke_scripts, comparisons, temp, exports.
Challenges we ran into
Synchronizing JSON parameter structures across FIBO API providers with different schema requirements required abstraction layer development. Managing asynchronous parallel generation while maintaining consistency and preventing rate limit violations needed semaphore-based execution control. Converting 8-bit API outputs to production 16-bit HDR formats without quality loss required custom image processing pipelines. Embedding JSON parameters into Nuke scripts while maintaining compositor adjustability demanded specific node structure design.
Accomplishments that we're proud of
Successfully integrated four different FIBO API providers with unified parameter interface. Implemented parallel agent system generating six variants in execution time equivalent to single generation. Created automated Nuke script generation maintaining JSON-native control through compositor-adjustable nodes. Built HDR 16-bit processing pipeline producing print-quality outputs from standard API responses. Developed AI scoring system for composition quality based on photographic principles.
What we learned
Multi-provider API integration requires flexible abstraction layers accommodating varying response formats and parameter schemas. Parallel execution at scale needs careful rate limiting and error handling to prevent cascade failures. Professional pipeline integration demands understanding of existing tool workflows rather than forcing new paradigms. HDR processing requires linear color space operations and proper bit depth handling throughout the pipeline. Agent-based architecture enables horizontal scaling but requires robust inter-agent communication and result aggregation.
What's next for AI-Driven Industrial Product Design Studio
Implement direct ComfyUI server integration for local workflow execution. Add WebSocket support for real-time generation progress streaming. Integrate additional generative models beyond FIBO for multi-model comparison. Develop batch processing system for large-scale asset generation. Add cloud storage integration for team collaboration. Implement version control for design iterations. Create plugin system for extending to additional 3D and compositing tools beyond Nuke. Add training data collection for fine-tuning models on specific product categories.
Log in or sign up for Devpost to join the conversation.