flux-2-klein

To generate images with text prompts using RunComfy and Flux 2 Klein, you invoke the RunComfy CLI against the blackforestlabs/flux-2-klein text-to-image endpoints. You provide a JSON input schema containing your prompt, steps, width, height, and optional reference images.

Flux 2 Klein 4B routes for speed and low-latency concepting, while 9B routes for higher-fidelity final polished output. Both variants use the same prompt grammar and endpoint shape, allowing you to switch between rapid iteration and final polish.

Yes, you can use up to 4 reference images in Flux 2 Klein to maintain multi-reference brand styling. This keeps style consistent across creative iterations and brand assets during your text-to-image generation workflow.

For text-to-image generation in Flux 2 Klein, use 4 to 8 steps for rapid iteration and approximately 25 steps for a final polished pass. Adjusting the step count lets you balance processing speed and output fidelity.

The best way to structure prompts for Flux 2 Klein image generation is using a subject-first declarative structure and specificity-first phrasing. This model-aligned prompting guidance produces sharper results and reduces trial-and-error.

Yes, Flux 2 Klein image generation allows you to adjust image dimensions by specifying width and height in the JSON input schema. You must adjust these dimensions within the model’s documented constraints.