FLUX 3 Image

FLUX 3 is Black Forest Labs’ image model. You give it a prompt and it generates an image. Give it one or more reference images and it edits them instead.

What it does

• Text to image. Describe what you want and get an image back.
• Image editing. Pass up to 10 reference images and the prompt describes the edit or composition to apply. When you supply reference images, the first one sets the aspect ratio if you leave aspect_ratio on auto.

Inputs

• prompt — what to generate, or the edit to make.
• images — up to 10 reference images (jpeg, png, gif, or webp). Supplying any image switches the model into edit mode.
• aspect_ratio — auto (default) picks a ratio from the prompt and the first reference image, or choose a fixed ratio from 21:9 down to 9:21.
• resolution — output size: 768sq, 1k (default), 1.5k, 2k, or 4k.
• grounding — when on (default), the model researches the prompt with web and image search before generating. Turn it off for faster, self-contained generation.
• safety_tolerance — moderation strictness from 0 (strictest) to 4 (most permissive).
• output_format / output_quality — webp, jpg, or png, and the quality for lossy formats.

Reference images

Reference images must be at least 256x256 px and at most 16 MP. Use them to keep a subject, product, or style consistent while the model builds a new scene around it.

Prompting tips

• Be specific about style, lighting, and composition — FLUX 3 holds a look well when you name it.
• For edits, describe the change clearly (“replace the background with a snowy mountain”) rather than re-describing the whole scene.
• Leave aspect_ratio on auto when editing so the output matches your first reference image.