Black Forest Labs: FLUX.2 Flex

FLUX.2 Flex is an image generation model from Black Forest Labs, the studio behind the FLUX family of image models. It sits in the image category on api14, and the provider released it on 25 November 2025. If you already know the FLUX line, this is the variant tuned for detailed rendering and reference driven editing rather than a general purpose text model.

The model takes text and image inputs and returns images. Its context window is 67,344 tokens, with a maximum output of 60,609. The provider describes FLUX.2 Flex as strong at rendering complex text, typography and fine details, and it supports multi-reference editing inside the same unified architecture. In practice that means you can send a prompt, or a prompt plus up to eight input references, and get a generated or edited image back. Aspect ratio is selectable from 1:1, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16, 21:9 or auto, and output can be written as png or jpeg.

Those capabilities point at a fairly specific set of jobs. Text heavy images are the obvious one: posters, packaging mockups, banners and social assets where lettering has to come out legible and correctly spelled. Fine detail rendering suits product shots, UI mockups and illustrations that need clean edges. Multi-reference editing is useful when you want to keep a subject, style or layout consistent across variations, for example applying a brand look to several images or restyling a photo while holding the original composition. Because image input is accepted, image to image workflows and reference guided edits are both reasonable fits.

You call it through api14's OpenAI compatible API using the model id black-forest-labs/flux.2-flex. The request supports the seed parameter, and the image parameters cover aspect_ratio, output_format, n and input_references. Aspect ratio accepts the enum values listed above, output format accepts png or jpeg, n is fixed at 1 so each request returns a single image, and input_references accepts between 0 and 8 images. Seed is exposed as a boolean, which lets you control reproducibility across runs.

One constraint worth noting before you commit: n is capped at 1, so batch generation means multiple requests rather than one call. Input references top out at 8, which is generous for most editing work but a real ceiling if your pipeline expects more. And since the model outputs images only, any task that needs text back, such as captioning or reasoning over an image, belongs with a different model.