Black Forest Labs has shipped FLUX Erase, a new endpoint that removes whatever you mask in an image and reconstructs the background in a single pass. Announced May 21, the model handles objects, shadows, reflections, text, and clutter together, instead of treating cleanup as a separate inpainting prompt.
Try it: Clean a product shot in one mask
Upload a product photo to the FLUX Erase endpoint or paste it into the public demo at flux-tools.bfl.ai/erase. Paint a binary mask over what you want gone (a stray hand in the frame, a power cable, an embedded watermark) and submit. The model returns the same image with the masked area filled to match surrounding lighting and texture. No prompt required. The whole loop replaces the usual two-step path of running Stable Diffusion inpainting plus a second pass for shadows or reflections.
Why it matters
Object removal tools have historically left visible halos, smearing, or shadow ghosts. FLUX Erase treats removal as one reconstruction task at the model level, which is the same architectural shift that top image models have been moving toward for edit-mode work. BFL's benchmarks show the new endpoint outperforming GPT Image-2 (68.5 percent win rate) and Finegrain Eraser (63.2 percent), and matching Google Nano Banana 2 while approaching Nano Banana Pro performance at lower price and latency.
Key details
FLUX Erase is available through the BFL API and the developer dashboard at dashboard.bfl.ai. Per-call pricing is listed on the BFL pricing page next to FLUX.1 Fill and FLUX.1 Kontext. The model accepts an input image and a binary mask, and ships with a public playground for testing without writing code. It joins FLUX Outpainting (released May 14) and FLUX.2 [klein] as part of BFL's editing-suite push for the FLUX.2 generation. The announcement was also cross-posted to X by the BFL team.
What to do next
Pick the last three frames where you had to manually clean up a stray element (a passerby, a wire, a logo) and re-run them through FLUX Erase. Compare the output to whatever inpainting workflow you currently use. If the result handles shadows and reflections without a second pass, swap it into your pipeline before the next product or social shoot.