What is FLUX 3 Image? Black Forest Labs’ model, explained

FLUX 3 Image is the image generation and editing model in the FLUX 3 family from Black Forest Labs. This post sticks to what Black Forest Labs has published about it, read on October 2, 2026, and leaves out anything we cannot back.

Who makes it

Black Forest Labs is the lab behind the FLUX models, the line that started with FLUX.1 and continued through the FLUX.2 generation. FLUX 3 is the third generation. BFL announced the FLUX 3 family on July 23, 2026, and FLUX 3 Image is the family's image model: available through the BFL API at flux-3-image and in BFL's own playground. The model page's one-line pitch is "maximum control over every pixel."

Flux 3 Image, this site, is an independent workbench that accesses FLUX 3 Image through fal's text-to-image and edit-image endpoints. We are not affiliated with, endorsed by, or sponsored by Black Forest Labs; FLUX is their trademark, used here only to identify the model we provide access to.

What it does

Four things, each documented by BFL with examples.

Text to image. Describe the subject, setting, light, and framing and the model renders the whole scene from the prompt. BFL highlights strong prompt following and accurate text rendering, including multilingual text: the published samples include a Congolese airmail stamp with its denomination, a magazine cover with its title, and a collage with a designer's name set in type.

Editing by instruction. Send an image and write the change as an instruction. The documented examples are small and specific: change the color of one bird in a flock, recolor an owl's feathers, swap a plate for a burger, remove a cat. Name what should change and what should remain. The fal integration forwards your prompt without enabling optional prompt expansion.

Multi-reference composition. Up to ten images in one request. You refer to each by its position, image 1, image 2, and say what it contributes: the person from image 1 wears the jacket from image 2 and stands in the kitchen from image 3. By default the output keeps the aspect ratio of the first image.

Bounding boxes. To fix where every element goes, you append boxes to the prompt, each one [top, left, bottom, right] on a 0 to 1000 grid with a short description. On a finished image the same boxes target an edit: recolor, move, replace, or remove one element while everything else stays unchanged. We cover this in detail in how bounding boxes work.

BFL's direct API documents web grounding, including an Oktoberfest poster example. Our fal endpoints do not expose a grounding parameter, so this workbench does not offer that control or promise web-grounded output.

Resolutions and ratios

The fal endpoints accept 512sq, 768sq, 1k, 2k, and 4k. Our workbench offers 1K (about one megapixel), 2K (about four), and 4K (about sixteen). Fourteen aspect ratios are supported: 21:9, 2:1, 16:9, 3:2, 7:5, 4:3, 5:4, 1:1, 4:5, 3:4, 5:7, 2:3, 9:16, and 1:2, plus auto, which follows the first reference or falls back to square.

The current composer uploads one reference. fal accepts reference URLs or data URIs, at least 256 pixels on each side and at most 4 megapixels each. The preview credit price follows the selected resolution; see how credits work.

How it differs from the FLUX.2 generation

We will only say what BFL says, and BFL's FLUX 3 Image pages do not contain a side-by-side comparison with FLUX.2 that we can quote. What they do put at the front of FLUX 3 Image is control: bounding boxes for layout and targeted edits, up to ten references in one request, grounding against web search, and a result that returns the expanded prompt it actually rendered from. If you want a benchmark number comparing the generations, BFL has not given us one, so we will not invent it.

Timeline

• July 23, 2026: Black Forest Labs announces the FLUX 3 family.
• October 2, 2026: BFL model documentation and fal's FLUX 3 Image generation/editing endpoints are publicly available. This site remains a local preview; live generation has not been verified.

Try it

The generator on the homepage runs FLUX 3 Image with the credit price printed on the prompt box. Common questions, including sizes, editing, and refunds, are answered in the FAQ.

Sources: Black Forest Labs, bfl.ai/models/flux-3-image; docs.bfl.ai FLUX 3 Image overview and editing pages; bfl.ai/blog/flux-3; fal text-to-image schema and fal edit schema. Read October 2, 2026.