AI Potluck
Back to Gap Map Model components / Image, video, 3D & music generation

FLUX

Black Forest Labs
restricted / Overall score: 3.4

FLUX is Black Forest Labs' family of rectified-flow transformer models for text-to-image generation and image editing. The open-weight releases span FLUX.1 [dev] and [schnell] with their Kontext, Krea, Fill, Redux, Depth and Canny variants, the 32B FLUX.2 [dev], and the smaller FLUX.2 [klein] models that run on consumer GPUs. The strongest tiers, FLUX.2 [pro], [flex] and [max], are served only through the vendor's API.

Scored on the downloadable [dev], [schnell] and [klein] weights; the API-only [pro], [flex] and [max] tiers are not part of the reading. FLUX 3 Video is a hosted model, and the downloadable FLUX 3 Action is a robotics world-action model outside this entry.

Openness

2 high confidence
2.0
weights
open(fourteen downloadable image checkpoints across FLUX.1 and FLUX.2
data
closed(the model cards describe filtering of the pre-training data and release none of it)
code
partial(the flux and flux2 repositories carry minimal inference code
license
FLUX-Non-Commercial-License-2.1(the most restrictive license among the distributed checkpoints: FLUX.2 [dev] and the [klein] 9B models, alongside FLUX [dev] Non-Commercial License v2.0 and the FLUX.1 [dev] Non-Commercial License v1.1.1

FLUX.1 [dev], FLUX.2 [dev] and the 9B [klein] models may not be used commercially without a paid license from Black Forest Labs, while [schnell] and the [klein] 4B models are Apache 2.0. Only inference code is published and the training data is not released. FLUX-Non-Commercial-License-2.1 does not permit commercial use, so the license tier is commercial_forbidden.

Adoption

4 high confidence
4.0

Adoption is measured as Hugging Face downloads summed across the fourteen FLUX.1 and FLUX.2 base checkpoints, led by FLUX.1 [dev]. Quantized and LoRA repositories are left out.

Capability

3 medium confidence
3.0

FLUX.2 [dev] places in the upper half of the text-to-image arena and among the strongest open-weight image models, a step below Qwen-Image-2.1, which leads the open models there. The vendor's better-placed FLUX.2 tiers are API-only and are not counted.

Verified 2026-09-26