Black Forest Labs

FLUX.2

the strongest open-weight family

3 min readImage Generation

Key facts

5klein to max
Tiers
$0.015per standard image
Pro price
Open weightdev + klein
Licence
Germanyex-Stable Diffusion
Origin

The strongest open-weight family. Five tiers as of July 2026: [klein], [dev], [flex], [pro] and [max].

What it is

FLUX.2 is the image model family from Black Forest Labs, and as of July 2026 it is the strongest open-weight offering in the field. The lineup spans five tiers: [klein], [dev], [flex], [pro] and [max], arranged from the smallest and most efficient up to the most capable. The spread is the point. The family is designed to cover the whole range of needs, from a hobbyist running a model on a single graphics card to an agency pushing thousands of brand-consistent images through an API, all from one coherent set of releases.

Why open weights count

The open-weight structure is what sets FLUX.2 apart. The [dev] and [klein] weights are published and free to run on your own GPU, which means a studio can generate images entirely on its own hardware, with no per-image fee and no data leaving the building. That is a decisive consideration for anyone handling confidential briefs or working to a fixed budget. The heavier tiers are offered as a paid API instead, and [pro] is the cheapest premium option per standard image, starting near $0.015 through some providers, which keeps the economics competitive even at the top of the range.

Open weights carry a second benefit beyond cost. Because the [dev] and [klein] models can be downloaded, a studio can fine-tune them on its own images, teaching the model a house style or a specific product range that a closed API will never learn. The middle [flex] tier sits between the free weights and the premium options for teams that want more capability than a local model provides without committing all the way to [max]. Few closed competitors allow that degree of adjustment, and for technically minded teams the ability to inspect, adapt and host the model themselves is a large part of the appeal.

What it does well

The family has earned particular praise for two capabilities that production teams care about. The first is hex colour precision, the ability to hold an exact brand colour specified by its hexadecimal code rather than drifting to an approximate shade. The second is multi-reference consistency, keeping a character, product or style stable across a set of images drawn from several reference pictures. Both are the difference between a novelty and a tool a brand can rely on, which is why FLUX.2 has settled into agency and marketing workflows where consistency is non-negotiable.

At the top of the range, [max] leads the family on editing consistency and prompt following, the two measures that decide how much manual correction a job needs afterwards. For a team that has to deliver finished assets rather than pleasing first drafts, that reliability is where the value sits, and it lets the free tiers below handle drafting and experimentation at no cost before the paid tier finishes the work.

The lab behind it

The company behind it explains a good deal about why the family exists. Black Forest Labs was founded by the researchers who created the original Stable Diffusion, the model that first opened image generation to the public. When they left to start their own venture, the centre of gravity of the open image ecosystem moved with them to Germany, and FLUX.2 is the continuation of that lineage. Keeping the strongest weights open is a philosophical stance as much as a commercial one.

What to watch

Within the wider set of image models, FLUX.2 holds a position no closed competitor occupies: professional-grade output that a team can run privately and inspect for itself. The closed leaders may edge ahead on individual benchmarks, but they cannot be downloaded, audited or fine-tuned, and for a large share of studios that freedom outweighs a few points of arena rating. What to watch is whether the open tiers keep pace with the closed frontier as models grow more demanding, and whether the German lab can sustain a business that gives its best research away.