Creative Campaign Engine

Multi-stage image campaign orchestrator that progresses generated images through a structured workflow: prototype generation, vision model analysis, informed refinement, and variation production. Each stage feeds the next — analysis output from one stage shapes the parameters for the next generation pass, creating an iterative improvement loop rather than one-shot generation.

Campaigns define prompts, parameters, quality thresholds, and stage-specific configuration. The prototype stage generates initial candidates. Vision model analysis evaluates each output for quality, composition, style adherence, and defects. The refinement stage uses analysis feedback to adjust generation parameters and produce improved versions. The variation stage creates alternatives from successful outputs.

Asynchronous task processing with distributed queue dispatch across available GPU nodes. Full status tracking per task and per campaign — pending, processing, completed, failed, cancelled. Manual intervention supported at any stage: review intermediate results, adjust parameters, skip stages, or branch a campaign in a new direction from any completed output.

Campaign state persists to database across sessions. Resume a campaign days later, review results from any stage, refine individual outputs, or generate new variations without re-running the full pipeline. Designed for systematic creative work where the goal isn't a single image but a curated set that meets specific quality and consistency requirements.