FLUX.3 Image Klein lands in Diffusers

Black Forest Labs' FLUX.3 Image Klein is available in Hugging Face Diffusers for text-to-image, multi-reference image-to-image, and bounding-box layout control. Distilled and base checkpoints ship with LoRA training scripts.

Black Forest Labs' newest open image model, FLUX.3 Image Klein, is supported in Hugging Face's Diffusers library from launch day, so developers can start generating and editing images with a few lines of Python.

One model, three jobs

FLUX.3 Image Klein handles text-to-image generation and multi-reference image-to-image editing in a single model. Its new feature is layout control: you can write bounding boxes into the prompt to specify where each element of an image should appear, or which parts of a reference image should be moved. For anyone who has fought with prompts to get a logo in the top left or a product in the center, that is a meaningful step toward predictable composition.

Speed and hardware

According to the Hugging Face post, the distilled checkpoint produces a 1024×1024 image in about two seconds on an H100 GPU. For smaller GPUs, quantized NVFP4 and FP8 weights are available and load through the standard single-file method. A base, non-distilled checkpoint is also provided for people who want maximum quality or plan to fine-tune.

The release includes LoRA training scripts for both text-to-image and image editing, which makes it practical to teach the model a brand style, a product, or a character with a modest dataset. The pipeline ships in Diffusers' modular format, so you can swap out components or insert your own steps in the denoising loop.

Why it matters for builders

Open image models with strong layout control are especially useful for product teams that need consistent visuals at scale: marketing banners, thumbnails, e-commerce shots, or in-app illustrations generated per user. Hosted tools like Midjourney remain easier for one-off creative work, but an open model you can run and fine-tune yourself gives you control over cost, style and data.

The practical checklist before adopting it: confirm the license fits your commercial use, test the quantized weights on your target GPU, and try a small LoRA on your own assets to see how quickly it locks in your style.

FLUX.3 Image Klein is listed in our AI Images category alongside other image generators.

Source: Hugging Face. This article is PickAI’s own summary and analysis based on the original reporting.