Black Forest Labs: FLUX Erase outperforms GPT Image-2 (68.5%) and Finegrain (63.2%) in prompt-free object removal

Black Forest Labs (BFL) launched FLUX Erase on 21 May 2026 — a specialised inpainting tool that removes objects, shadows, watermarks, and text from images using only a binary mask, without any textual description of what to remove. The tool is immediately available through the BFL commercial API and a public demo at flux-tools.bfl.ai/erase.

What are the concrete benchmark results?

BFL published a benchmark on a set of 198 test images with the following results:

• 68.5% win-rate against OpenAI GPT Image-2 in object removal tasks
• 63.2% win-rate against Finegrain Eraser Standard, the previous category leader

Win-rate means that human annotators chose the FLUX Erase output as better in that percentage of cases. The result is significant because GPT Image-2 (OpenAI multimodal model) and Finegrain Eraser are currently considered the leading tools for professional image editing.

Why is prompt-free operation technically challenging?

Classic inpainting tools require a textual prompt — “remove the person in the foreground”, “erase the watermark”. This is inherently difficult because the user must describe what they want removed, and the model must correctly understand the description and selectively act on the right part of the image.

FLUX Erase reverses the problem — the user provides only a binary mask (a black-and-white image where the white area marks what to remove). The model must independently:

1. Understand what is inside the mask (object, shadow, text, watermark)
2. Reconstruct the background based on context from unmarked parts of the image
3. Prevent artefacts — without introducing objects that do not belong in the scene

This requires a model with strong understanding of the entire scene, not just the local area around the mask.

What are the use cases?

FLUX Erase is designed for professional creative workflows — photographers, designers, and content creators who need fast, high-quality tools for removing unwanted elements from images:

• Removing watermarks from stock photos (legal only if you have the rights to the image)
• Removing background people from tourist photographs
• Erasing text and logos from background elements in videos
• Real estate photography — removing furniture, cars from the scene
• E-commerce — removing backgrounds from product images

What does this mean for BFL positioning?

Black Forest Labs is increasingly positioning itself as a specialist in professional creative tools, in contrast to OpenAI and Google which build generic multimodal models. FLUX models have consistently been top performers in text-to-image categories, and now with Erase they are entering the image editing segment.

The strategy makes sense — image editing is a segment where corporate users (marketing agencies, photographers, video productions) are willing to pay for better tools. BFL offers both an API and a web UI, covering both channels.

It is worth watching whether BFL will continue expanding into other professional creative niches — perhaps video erasing, 3D inpainting, or audio cleanup. With solid performance and clear use cases, FLUX Erase is a good example of a specialised AI tool that outperforms generic models in a specific domain.