FLUX.2 [max] vs [dev]: when paying more actually changes the result
A practical guide to the FLUX.2 model tiers, where [max], [pro], [dev], [flex], and [klein] change the result, and when the cheaper model is enough.
The price gap is not the whole story
FLUX.2 is not one model with nicer labels on top. Black Forest Labs has turned it into a ladder: FLUX.2 klein for speed, FLUX.2 dev for open experimentation, FLUX.2 pro for production, FLUX.2 flex for control, and FLUX.2 max for the most demanding outputs.
That ladder matters because image generation failures are uneven. A bad sketch costs almost nothing. A bad product hero image can waste a whole review cycle. A character that changes face halfway through a campaign is worse than a slow render. Paying more only makes sense when the stronger tier reduces a failure that would cost you more than the generation itself.
Black Forest Labs launched the first FLUX.2 wave on November 25, 2025, with FLUX.2 pro, FLUX.2 flex, and FLUX.2 dev. FLUX.2 max followed on December 16, 2025. FLUX.2 klein arrived on January 15, 2026. That release order is useful context. The family started with the production and open tiers, then added a top tier for harder final work, then added a compact tier for fast interactive use.
What changed with FLUX.2
The useful jump from FLUX.1 to FLUX.2 is not just prettier samples. FLUX.2 combines generation and image editing in the same family, supports multiple reference images, handles larger images, improves text rendering, and gives better control over color, composition, and scene structure.
Black Forest Labs says FLUX.2 can edit images up to 4 megapixels and work with as many as 10 reference images in some contexts. In practice, that means you can give the model a product photo, a style reference, a background direction, and a layout instruction, then ask for one coherent result. That is a different class of work from a simple text prompt.
The hard part is choosing how much model you need. People waste money in both directions. They spend top-tier credits while still figuring out the idea, or they stay on a cheap tier after the output is already close and the remaining problems are exactly the ones the stronger tiers solve.
Current pricing, in plain English
Black Forest Labs prices FLUX.2 with a simple credit system where one credit equals one cent. Its current public pricing lists FLUX.2 dev as free for non-commercial local development. FLUX.2 klein starts at $0.014 per image for the 4B version and $0.015 per image for the 9B version. FLUX.2 pro starts at $0.03 for text-to-image and $0.045 for editing. FLUX.2 flex is listed at $0.06. FLUX.2 max starts at $0.07.
Those numbers are starting points because the FLUX.2 API pricing scales with image size and, for editing, the image work involved. Still, they show the shape of the decision. FLUX.2 klein and FLUX.2 dev are for volume and exploration. FLUX.2 pro is the everyday paid production tier. FLUX.2 flex costs more because it gives you more control. FLUX.2 max costs the most because it is meant to improve prompt following, consistency, and grounded generation.
Replicate tells a similar story from the hosted-platform side. Its FLUX.2 launch post described FLUX.2 pro as generating in about 6 seconds, or about 9 seconds with an input image, and priced it at $0.015 plus $0.015 per input and output megapixel. The same post put FLUX.2 flex at around 22 seconds for generation, around 40 seconds with an input image, and $0.06 per input and output megapixel. Replicate's model pages also frame FLUX.2 dev as a hosted option for generation and editing, while BFL's own docs treat the weights as free for non-commercial local development.
That distinction matters. Free local weights do not mean every hosted workflow is free. They mean developers can experiment locally under the license terms, while managed platforms may charge for compute, hosting, queues, and commercial access.

AI 图像生成
输入文字描述,AI 智能生成精美图片