CrashBench

Generate auto-damage data to test your vision AI

The Auto-damage preset paints controlled damage onto a real vehicle photo via masked inpaint (FLUX Fill pro). You get coherent, localized, labelled claims — the raw material to stress a damage detector.

How it works

flux-fill-pro3 credits
flux-fill-proauto-damage

How it works

Plain text-to-image models hallucinate a plausible dent but never a precise one, in the right place, physically consistent with the bodywork. Masked inpaint fixes that: you generate on a real photo and keep the rest intact.

Each base image can then go through the deterministic stage (WhatsApp compression, angle, downscale) to mimic how a claimant actually sends a photo. One paid generation becomes a full test set.

Key points

  • Masked inpaint: damage painted on the real photo, not invented from scratch.
  • Labelled output: model, seed, ops and human rating stored with the image.
  • Ideal for claim detectors, auto expertise and field robustness.