Render-Compare Loop Runner

Turn brute-force render-diff-fix loops for visual generation into a measurable pipeline

Problem

A developer converting images to PPT found the hard part isn't the first version but matching the original block by block, ending up driving the agent through render, diff, find the worst block, fix, repeat for a dozen rounds per image. Each round needs a real render, and controlling rounds and cost is guesswork.

Solution

A local iteration orchestrator where you define a target image and comparison metrics (block diff, SSIM, OCR text delta), and it drives the renderer or agent through render, score, locate worst block, revise. It visualizes per-round difference heatmaps and enforces round and cost caps with early-stop rules, plus tiled downsampling for huge images.

Users

Developers building image/document-to-editable-format conversion and agent apps that need repeated visual comparison of generated output

MVP

Hook one render command and one comparison metric; per-round diff heatmap and worst-block locator; caps on rounds, time and cost; snapshot comparison across rounds; CLI plus minimal local UI

Revenue

Per-project desktop license; open core with a hosted version billed per render round

Why now

Visual generation and reconstruction agent workflows are exploding, yet iteration control, cost caps and observability are still hand-rolled scripts, a classic narrow but deep gap