Replicating lighting language and texture identity was inconsistent.
A fast-growing creative brand needed a way to generate unlimited on-brand images without manually crafting style prompts for every request. Their design team had built a signature visual identity across 15+ reference images but replicating that same cinematography, lighting language, and texture identity was inconsistent and time-consuming.
They wanted a solution that could learn the brand's style, store it, and automatically apply it to any new user-generated prompt, all while maintaining aesthetic consistency, camera logic, and the brand's "old-money, luxury, retro" vibe.
A fully automated style-extraction and generation engine.
Our team built a fully automated style-extraction and image-generation engine powered by n8n, Gemini AI, Supabase, and multi-model image generation APIs, delivering 100% consistent brand-style images on demand.
Phase 1: Perfect Style Extraction
The system ingests 15+ reference images and performs deep cinematic style extraction using Gemini's multimodal capabilities, focusing on:
- High-precision camera logic breakdown: Lens compression, framing rhythm, angle behavior, bloom, contrast physics.
- Lighting identity analysis: Warm sunset hues, side-lit highlights, retro glow falloff.
- Texture & material fingerprinting: Wood varnish details, fabric tension, micro-scratches, oil-sheen physics.
- Imperfection pattern capture: Analog dust, chromatic fringe, bloom leaks, atmospheric haze.
Each extracted style is stored in Supabase as structured text linked to the image filename, creating a permanent repository of the brand's aesthetic DNA.
Phase 2: Object Recognition for Context Matching
When a user submits a prompt, the system automatically scans the entire stored dataset, finds the closest matching reference image, and extracts its semantic content. It returns the best-aligned style description ensuring every new output matches the correct style segment.
Phase 3: Final Image Generation Engine
The system merges the user prompt, matching style, aesthetic motifs, imperfection profile, camera DNA, and lighting rules into a single Alpha-Prompt.
This Alpha-Prompt is sent simultaneously to multiple models (Gemini 2.5 Flash, Imagen 4, Nano Banana Pro, Ideogram V3, Grok Imagine). Each model returns a variation with the same brand identity preserved, and the best output(s) are uploaded to Google Drive automatically.
