FLUX.2 Flex
NewReleased Nov 25, 2025
₹2.88
per image, from
Billed at provider rates converted at ₹96 per US dollar, with 0% markup.
Price per image
Output options
- Resolutions
- 1024x1024, 1024x768, 768x1024, 1536x1024, 1024x1536, 2048x2048
- Editing
- Yes
- Reference images
- up to 10
About FLUX.2 Flex
A FLUX.2 tier that trades a little speed for quality control, letting the number of generation steps vary. Spending more steps can carry it past the pro tier's quality at the same list price.
Flex is the FLUX.2 tier that hands the sampler back to the caller. Black Forest Labs exposes the number of inference steps, up to 50, and the guidance scale between 1.5 and 10, and describes the tier as the one that excels at rendering text and fine details. Everything else in the family is shared: the same Mistral-3 24B vision-language model driving a rectified flow transformer, the same multi-reference composition, the same 4MP editing at any aspect ratio, the same hex-exact colour control and structured JSON prompting.
The lab publishes the trade directly rather than describing it. Its own comparison runs the same design at 6, 20 and 50 steps and shows typography accuracy improving with each increase, and a second pair does the same for image detail. That is the reason the tier exists: on a poster, a packaging mock-up or an interface frame where the words have to be right, more steps buys correctness, and on a background or a texture where they do not, fewer steps buys latency back. Where pro picks one point on that curve for you, flex lets a pipeline pick per job.
Guidance is the other dial, and it works against prompt adherence rather than for it in a simple way: higher values push the model harder toward the prompt at the cost of naturalism, lower values loosen it. Combined with step count and the family's structured prompting, that is enough control to tune a house look once and reuse it, which is what Black Forest Labs means when it calls flex the choice for designs requiring exact specifications.
What flex gives up is predictability. A tier whose latency and cost depend on a parameter is harder to budget than one that does not, and the lab positions pro rather than flex as the general production default, and max above both for editing consistency and grounded generation. Flex earns its place on typography-led and detail-critical work; on everything else the extra dials are one more thing to get wrong.
What Black Forest Labs announced at launch
- Steps and guidance exposed
- The only FLUX.2 tier that lets a caller set inference steps up to 50 and guidance between 1.5 and 10, rather than accepting the point on the curve the lab picked.
- Typography is the specialty
- Black Forest Labs describes flex as the tier that excels at rendering text and fine detail, and shows the same design at 6, 20 and 50 steps to make the trade visible.
- Detail bought by the step
- The lab publishes a second comparison for image detail rather than text: more steps recovers fine texture, fewer steps returns latency, and the caller decides per job.
- The rest of the family, unchanged
- Multi-reference composition, editing at up to 4MP, exact hex colour and structured JSON prompting all work here exactly as they do on the other FLUX.2 tiers.
- What it is not for
- Latency and cost move with the step count, which makes flex harder to budget than a fixed tier. Black Forest Labs positions pro as the general production default and max above both for editing consistency and web-grounded generation.
Frequently Asked Questions
Frequently asked questions about FLUX.2 Flex.
When was FLUX.2 Flex released?
Black Forest Labs released FLUX.2 Flex on Nov 25, 2025.
Who built FLUX.2 Flex?
FLUX.2 Flex is developed by Black Forest Labs. 99Models AI connects directly to it at the provider's published rate.
How much does an image from FLUX.2 Flex cost?
Images start from ₹2.88 each, billed at the provider rate with 0% markup. You pay only for the images you generate.
What sizes and aspect ratios does FLUX.2 Flex support?
Available options are 1024x1024, 1024x768, 768x1024, 1536x1024, 1024x1536, 2048x2048, which you can select before generating an image.
Can FLUX.2 Flex edit an existing image?
Yes. This specifies whether the model supports modifying an uploaded image in addition to creating images from text prompts.
More models from Black Forest Labs
Catalog updated Sep 9, 2026