FLUX.2 ([max], [pro], [flex], [dev], [klein])
The previous FLUX generation, combining a 24B vision-language model with a rectified flow transformer for both generation and editing. It spans closed API tiers and open-weight [dev] and [klein] models, which remain the main open option from Black Forest Labs.…
Best for
Open-weight image generation and editing that you can run locally or fine-tune ([dev], [klein]).
Overview
The previous FLUX generation, combining a 24B vision-language model with a rectified flow transformer for both generation and editing. It spans closed API tiers and open-weight [dev] and [klein] models, which remain the main open option from Black Forest Labs. Marked legacy because FLUX 3 Image is now the flagship, but still widely used and sold.
Strengths and limits
- Open weights: [dev] 32B on Hugging Face; [klein] under Apache 2.0
- Up to 10 reference images and editing up to 4 MP
- Klein models run on-device (e.g. ASUS ProArt laptops)
- [dev] commercial use requires a BFL license
- Ranks well below FLUX 3 Image and top closed models on Artificial Analysis
Published benchmark scores
Scores are as published by the source; settings (reasoning effort, tools, attempts) differ between labs, so compare with care. How to read benchmarks →