Flux
The open state of the art for photorealism and text in images.
Olya's profile
- What sets it apart
- Flux is the family from Black Forest Labs, the German lab founded by the people who created Stable Diffusion. What really sets it apart is that it runs on two tracks at once: open models you can download and host yourself (the FLUX.2 klein weights, plus the older dev and schnell) and closed models via API for top quality. In July 2026 it raised the bar with FLUX 3, which tries to unite image, video, audio and even robotic actions in a single model.
- Strengths
- On photorealism and text inside images it is one of the few open models that can go toe to toe with the closed heavyweights: legible lettering and few hallucinations on the characters. The FLUX.2 line works in context — feed it several reference images and it keeps them coherent — with targeted editing like outpainting and erasing, plus a klein variant that generates in under a second. The open model gives you control and fine-tuning without depending on a cloud.
- When to use it
- It pays off when you need believable photorealism or correct text inside an image — posters, product mockups, graphics with words — and when you want to host the model in-house for privacy or high volume. The open klein version is handy for rapid prototyping or for building it into your own product without paying per image.
- When to avoid it
- If you just want to type a sentence and get an image with no fuss, ready-made consumer tools are simpler: the pay-as-you-go API and self-hosting both have a learning curve. FLUX 3, with video and audio, is still invite-only early access at the end of July 2026 — the image tier and open weights are yet to land and there is no public price — so for production today stay on FLUX.2. And the head-to-head video numbers are BFL's own, with no independent verification.