FLUX.2 Klein 9B for Image Editing

Flux.2 Klein 9B is a 9‑billion‑parameter image model from Black Forest Labs that unifies text‑to‑image, image‑to‑image, and powerful image editing (including multi‑reference) in a single fast architecture.

What it is

• A rectified flow transformer with a 9B image model plus an 8B Qwen3 text encoder, designed for high‑quality generation and editing at 1024×1024.

• Released in both Base and Distilled variants: Base for maximum control and quality, Distilled for 4‑step, sub‑second inference on modern GPUs.

• Positioned as the “small flagship” of the FLUX.2 line, matching or beating much larger models in quality while staying efficient enough for interactive use.

Key features

• Unified workflow: one model for text‑to‑image, image‑to‑image, and image editing (single‑reference and multi‑reference).

• 4‑step distillation in the 9B Distilled variant for very low latency while keeping strong visual fidelity.

• Multi‑reference editing: can use multiple reference images (e.g., up to three) to control subject, style, or composition.

• Strong prompt adherence and output diversity, with good handling of detailed instructions and photorealistic scenes.

• Runs on high‑end consumer GPUs; the distilled 9B is explicitly targeted at production, real‑time, and interactive applications.

Typical use cases

• Fast, high‑quality image editing:

  • Text‑guided edits (“make it nighttime”, “change the outfit to a blue jacket”, “remove the logo”).

  • Object replacement/removal and style transformation for photos, product shots, and scenes.

• Multi‑reference creative work:

  • Combine elements from several images (e.g., “place the person from image 1 in the room from image 2 with the lighting of image 3”).

• Interactive and production tools:

  • Real‑time or near‑real‑time preview in editors, design tools, and web apps where latency is critical.

  • As a default image model in pipelines that need both generation and later fine‑grained edits without swapping checkpoints.