FLUX 2 Klein 9B Base LoRA

FLUX 2 Klein 9B Base LoRA is a nine billion parameter image generation model from Black Forest Labs that generates images from text descriptions and supports multi-reference editing with custom LoRA adapters. This is the undistilled foundation model designed for maximum flexibility and control.

What makes this different

Klein means “small” in German, but this model punches above its weight. It matches or exceeds models five times its size

This is the base variant, not the distilled version. That means you get the complete training signal without step or guidance distillation. Higher output diversity, more flexibility for customization, and better results when you add your own LoRA adapters for specialized styles or subjects.

What you can do with it

Text-to-image generation

Write a description and the model generates an image. The nine billion parameter architecture delivers high-quality, photorealistic results with strong prompt understanding.

Image editing with reference images

Upload an existing image and describe the changes you want. The model understands both the visual context from your reference image and your text instructions to make precise edits.

Multi-reference generation

Combine up to five reference images to maintain character consistency, product appearance, or visual style across generations. This is useful for creating variations while keeping specific elements consistent across multiple outputs.

Custom LoRA support

Apply your own trained LoRA adapters to add specialized styles, characters, or visual aesthetics. The undistilled base model preserves the complete training signal, making it ideal for fine-tuning and LoRA training workflows.

When to use this model

Use Klein 9B Base LoRA when you need maximum control and customization. This model works well for:

• Research projects exploring image generation capabilities
• Fine-tuning experiments and LoRA training workflows
• Custom pipelines where control matters more than speed
• Creative exploration with higher output diversity
• Applying custom LoRA adapters for specialized visual styles

If you need faster results for production applications, Black Forest Labs’ distilled Klein 9B model generates images in under half a second with four steps. But if you’re training your own LoRAs or need the full model capacity for customization, this base version gives you what you need.

Model details

This model uses a rectified flow transformer architecture with nine billion parameters. Black Forest Labs trained it without step or guidance distillation, preserving the complete training signal for maximum flexibility.

Unlike typical image models that separate generation and editing into different checkpoints, Klein 9B Base handles both in a unified architecture.

Safety and responsible use

Black Forest Labs filtered pre-training data for not safe for work content and known harmful material. The model went through multiple rounds of safety fine-tuning, with internal and external evaluations focused on preventing abuse.

This model is released under the FLUX Non-Commercial License. You can use it for research and non-commercial projects, but commercial use requires a separate license from Black Forest Labs.

Try it out

You can try this model on the Replicate Playground