FLUX.1 Krea Dev
Open-weights text-to-image model Black Forest Labs built with Krea, trained for realistic photographic images without the oversaturated textures common in AI output.
Overview
FLUX.1 Krea [dev] came out of a collaboration between Black Forest Labs and Krea and is the open-weights version of Krea 1. Like FLUX.1 [dev], it has 12 billion parameters, uses a rectified flow transformer design and was trained with guidance distillation, and the model card says it can replace FLUX.1 [dev] in any system that already supports it. Training aimed at more realistic and varied images without oversaturated textures, and the card names aesthetic photography as the model's focus.
In human preference assessments, Black Forest Labs found it on par with the closed FLUX1.1 [pro], and the company says it outperforms earlier open text-to-image models. It is also meant as a base for fine-tuning.
What you can build
The photographic focus suits product and lifestyle shots for catalogs and ads, editorial portraits and interiors, and concept frames for film, game or music video pitches. Merchandise mockups are another fit: one sample on this page prints the word "wiro" legibly across a green hoodie.
Inputs
The default prompt is a full paragraph asking for a close-up of Iron Man wearing a green hoodie with "wiro" on the front, in a brightly lit high-tech workshop, captured with a cinematic lens. The prompt is the only field you must fill in.
Steps start at 25 and guidance scale at 4.5, with guidance adjustable from 1 to 20. The canvas starts at 1024 by 1024, and either side can be as large as 2048 pixels. Ask for up to 8 images per prompt, and fix the seed when a result has to be repeatable.
Outputs
Among the eight samples on this page are the Iron Man helmet above the "wiro" hoodie, a matte black Porsche 911 on a reflective concrete floor, a figure in a trench coat walking down a wet neon-lit alley, a roaring lion lit from behind, and a black and white shot of a ballerina dancing in the rain on a rooftop.
Recommended settings
The default size and guidance match the card's own example, 1024 by 1024 at guidance 4.5. The card gives no step count. Prompting style strongly affects how well the model follows a prompt, so the example prompts on this page make a good template: most name the subject, the setting and the light.
Limitations
The card warns that outputs do not always match the prompt and that results depend heavily on how the prompt is written. Krea [dev] may also amplify societal biases, and it is not designed to provide factual information.
Safety & compliance
The open weights carry the FLUX.1 [dev] Non-Commercial License, and the card states that outputs may still be used for commercial work. Before release, Black Forest Labs says it filtered NSFW and unlawful material out of the pre-training data, removed known child sexual abuse material from the post-training data with the Internet Watch Foundation, and had outside evaluators test checkpoints adversarially. Black Forest Labs and Krea also ran several rounds of targeted fine-tuning to limit misuse. Prohibited uses on the card range from unlawful activity and harm to minors to harassment, non-consensual intimate imagery and large-scale disinformation.
Example prompts
Great starting points for FLUX.1-Krea-dev.
API quick start
Run FLUX.1-Krea-dev with a single API call.
{
"prompt": "A dramatic and realistic close-up shot of…",
"steps": 25,
"scale": 4.5,
"samples": 1
}curl -X POST "https://api.wiro.ai/v1/Run/black-forest-labs/flux-1-krea-dev" \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_WIRO_API_KEY" \
--data-binary @- <<'JSON'
{
"prompt": "A dramatic and realistic close-up shot of…",
"steps": 25,
"scale": 4.5,
"samples": 1
}
JSONOther models you might like
Explore similar image generation models.



