Model Releases
Black Forest Labs Releases FLUX 3 Image Klein 5.6B Open Weights Model
Black Forest Labs expands its FLUX 3 ecosystem with the release of FLUX 3 Image Klein, a 5.6B parameter model optimized for text-to-image and multi-reference generation.
October 15, 2026
Model ReleaseFLUX 3 Image Klein
Introduction
On October 15, 2026, Black Forest Labs (BFL) officially announced the release of FLUX 3 Image Klein, marking a significant milestone in the company's staggered rollout of the FLUX 3 family. Following the successful launch of FLUX 3 Action in September, Klein represents the second major open-weights release, designed to bridge the gap between lightweight edge deployment and the massive FLUX 3 Dev backbone expected later this year.
For developers and AI engineers, this release is a pivotal moment for local visual intelligence. Unlike many frontier models that remain locked behind proprietary APIs, FLUX 3 Image Klein provides high-fidelity visual generation capabilities directly to the community. By offering open weights, BFL is enabling a new wave of customization, fine-tuning, and specialized deployment on consumer-grade and enterprise hardware alike.
- Release Date: October 15, 2026
- Model Type: Open-weights image generation
- Parameter Count: 5.6 Billion
- Primary Focus: Text-to-image, multi-reference, and image editing
Key Features & Architecture
FLUX 3 Image Klein is built on a sophisticated 5.6B parameter architecture that balances computational efficiency with high-dimensional semantic understanding. The model is engineered to handle complex prompt adherence, a common pain point in smaller-scale diffusion models. It excels particularly in multi-reference generation, allowing users to provide up to 10 distinct reference images to guide the stylistic or structural output of a new generation.
The release includes several specialized variants to suit different workflows. Developers can choose from the standard base model, ComfyUI-optimized versions for node-based workflows, and 'pup' variants designed for ultra-fast inference. The architecture is highly compatible with the Hugging Face ecosystem and is ready for immediate integration via diffusers-based tooling, ensuring that the transition from research to production is seamless.
- 5.6B parameter dense architecture
- Multi-reference support (up to 10 images)
- Available variants: Base, ComfyUI, and 'pup' (optimized)
- Native support for diffusers-based tooling
Performance & Benchmarks
In internal blind evaluations conducted by Black Forest Labs, FLUX 3 Image Klein has emerged as the leader in its weight class. The model demonstrates superior performance in three critical domains: text-to-image prompt alignment, multi-reference consistency, and nuanced image editing. While specific MMLU or SWE-bench scores are typically reserved for LLMs, BFL utilizes visual-semantic alignment scores and human-preference ranking to validate Klein's capabilities.