SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI imagesFLUX.2 Klein

FLUX.2 Klein

FLUX.2 Klein is an open-source AI image generation model released by Black Forest Labs, focused on ultra-fast generation, capable of producing high-quality images in under 1 second. It offers two variants: 4B (Apache 2.0 license) and 9B, where the 9B model's output quality can rival models 5 times its size, and the 4B model can run on consumer-grade GPUs with a full commercial license. Technically, it uses a Rectified Flow Transformer architecture, 4-step inference distillation, an 8B-parameter Qwen3 text encoder, supports multi-reference image editing and C2PA watermark provenance, and emphasizes safety by design. It can seamlessly integrate with Hugging Face Diffusers, ComfyUI, the Black Forest Labs API, NVIDIA TensorRT, Automatic1111 WebUI, and custom inference pipelines. The official website offers free daily credits (50 images/day, 4B model, 1024x1024) as well as paid plan options for professional creators. It is suitable for AI researchers and creators who need fast, high-quality, commercially usable image generation.

Open website

About this tool

Purpose

FLUX.2 Klein is an open-source AI image generation model developed by Black Forest Labs, with its core positioning being to generate high-quality images in an extremely short time, reportedly within 1 second. It is aimed at AI researchers, creators, and developers who need efficient image generation capabilities, and is suitable for local deployment, cloud API integration, and various content creation scenarios.

Main Capabilities

  • Ultra-fast generation: Officially claimed to produce production-quality images in under 1 second.
  • Dual model variants: Offers two scales, 4B and 9B. The 4B model uses the Apache 2.0 license, can run on consumer-grade GPUs, and has a full commercial license; the 9B model pursues the highest quality, and is officially said to match the output of models 5 times its size.
  • Unified generation and editing: Supports multi-reference input editing, enabling image modification based on multiple reference images within a single workflow.
  • Text understanding: Uses an 8B-parameter Qwen3 text encoder to improve prompt understanding and adherence.
  • Efficient inference: After step distillation, only 4 inference steps are needed to complete generation without sacrificing image fidelity.
  • Content provenance: Built-in C2PA watermarking can provide cryptographically signed metadata.
  • Safety first: Pretraining-stage filtering and post-training fine-tuning to defend against misuse vectors.

How to Use

FLUX.2 Klein can integrate with a variety of mainstream AI frameworks and platforms for easy deployment:

  • Hugging Face Diffusers: Native support is provided in the Diffusers library, with pipelines optimized for the 4B and 9B models.
  • ComfyUI: Full integration, supporting node-based workflows and custom sampling options.
  • Black Forest Labs API: Cloud API access, suitable for production deployments that do not want to manage infrastructure.
  • NVIDIA TensorRT: Optimized inference for NVIDIA GPUs to maximize performance.
  • Automatic1111 WebUI: A community extension is available for this popular Stable Diffusion WebUI.
  • Custom pipelines: Full access to model weights and architecture to build custom inference pipelines.

The official website also provides a direct generation entry point, where users can get started with free daily creditsGenerate without local deployment.

Use cases

  • AI researchers conducting model evaluation and experiments.
  • Creators who need to quickly produce high-quality images for content creation.
  • Developers integrating image generation capabilities into their own applications or workflows.
  • Projects that require commercial licensing (the 4B model is under the Apache 2.0 license).
  • Real-time or batch production scenarios with high requirements for generation speed.

Limitations and notes

  • Regarding the specific GPU models and VRAM requirements needed for local operation, the official FAQ mentions them but does not provide specific values in the provided text, so they are not currently available.
  • The specific prices, billing methods, and subscription details of the paid plans are not currently available.
  • The free plan explicitly includes 50 images per day, access to the 4B model, 1024x1024 resolution, and community support.
  • Tiered plan options are provided, but the corresponding prices and complete terms are not currently available.
  • Specific generation quality and speed may vary depending on hardware configuration, usage method, and workflow.

Technical highlights

Based on Black Forest Labs' latest Rectified Flow Transformer technology, the 9B parameter architecture can compete with larger models in output quality. Through step distillation, the number of inference steps is reduced to 4, achieving a balance between speed and quality. Multi-reference editing and C2PA watermarking enhance editing flexibility and content traceability.

Related tools

Browse category