Pilaster — AI Image Generation Platform
Pilaster is an AI image generation platform with memory. It owns three layers: a character registry (LoRAs, reference sheets, metadata for consistent characters), a generation abstraction (backend-agnostic interface for ComfyUI, Replicate, Runway, DALL-E, or any future engine), and experiment memory (tracks every generation with outcomes and quality scores, learns what works, warns before repeating failures).
The Problem
AI image generation is a cycle of trial and error with no institutional memory. Creators lose track of which prompts worked, which settings produced good results, and which approaches failed. Every session starts from scratch. Switching between backends (ComfyUI, DALL-E, Replicate) means losing all context.
The Solution
A platform that remembers everything. Every generation is tracked with intent, parameters, and quality score. Characters stay consistent via a registry of LoRAs and reference images. Structured prompt recipes decompose intent into reusable dimensions (subject, style, composition, lighting) that work identically across all backends.
The Outcome
Image generation goes from random experimentation to informed decision-making. The system learns from every attempt, reuses successful patterns, and warns before repeating known failures — across any backend.
Key Features
- Character registry with LoRAs and reference sheets for consistency
- Backend-agnostic generation — ComfyUI, DALL-E, Replicate, Runway
- Structured prompt recipes that map to ComfyUI nodes
- Experiment memory — tracks outcomes, learns from failures
- Version control with intent notes and parameter diffs
- MCP server for agent integration (8 tools)
Technology Stack
Interested in this project?
I'd love to discuss the technical details, challenges overcome, or similar projects I could build for you.
View Live AppLet's discuss this project