Everyone generating AI images burns credits and patience on the reroll death-spiral: the composition is perfect but the hands are claws, the sign text is gibberish, or the logo melted. So they reroll blindly with the same prompt and a new seed, wasting ten generations to salvage one. The real fix lives in the positive and negative prompt — not the seed — but crafting the right corrective terms for each specific defect and each specific model is fiddly, undocumented tribal knowledge scattered across forum threads.

Stop rerolling blindly. Describe the defect in plain English, paste your prompt, and get a rewritten positive prompt plus a lean, model-tuned negative prompt engineered to kill that exact flaw — so you fix it on the next generation instead of the tenth.

• nextjs-15
• typescript
• tailwind
• anthropic-or-openai-shim

P2 tier: the app requires a buyer-supplied LLM API key. The LLM call is fully wired, but when ANTHROPIC_API_KEY (or the configured provider key) is unset the main route surfaces a 'Set your API key in .env to run diagnoses' message at boot — no silent fallback to demo output. A small seed/examples.json ships for reference and screenshots only.

• Pro upgrade — Unlimited saved fix recipes, batch reroll-fixing across multiple images at once, and priority model presets
• Agency — Client-use rights, multi-seat access, and model-specific preset packs plus done-for-you failure libraries for hands, text, logos, and anatomy
• Done-for-you — Monthly drop of new model presets, negative-prompt recipes for the latest generators, and updated common-failure fixes as models evolve