black-forest-labs/flux-2 models

flux-2 by Black Forest Labs — AI Model Family

The flux-2 family from Black Forest Labs represents a cutting-edge suite of AI image generation models, officially branded as FLUX.2, designed to deliver photorealistic images with exceptional prompt adherence, sharp text rendering, and intricate detail. This family solves key challenges in AI image creation by providing ultrafast, high-fidelity text-to-image and image-to-image editing capabilities across a range of model sizes, from compact 4B-parameter variants to powerful Pro editions, enabling creators to generate professional-grade visuals efficiently on consumer hardware. Comprising over 20 specialized models, flux-2 spans categories like Text to Image, Edit (Image to Image), Base, Turbo, Flash, Max, Flex, Klein (4B and 9B), Pro, and LoRA adaptations, making it the ultimate toolkit for diverse workflows from rapid prototyping to custom fine-tuning.

flux-2 Capabilities and Use Cases

The flux-2 family excels in text-to-image generation and image editing, with variants optimized for speed, control, and quality. Core models include Flux 2 (Text to Image) for standard high-res outputs, Flux 2 | Edit (Image to Image) for precise refinements, and advanced options like Flux 2 Pro (Text to Image) and Flux 2 Pro | Edit (Image to Image) for production-grade fidelity. The Klein series—Flux 2 | Klein | 9B | Base | Edit (Image to Image), Flux 2 | Klein | 9B | Base | Text to Image (Text to Image), Flux 2 | Klein | 4B | Base | Edit (AI Models), Flux 2 | Klein | 4B | Base | Text to Image (Text to Image), Flux 2 | Klein | 4B | Edit (Image to Image), and Flux 2 | Klein | 4B | Text to Image (Text to Image)—offers sub-second generation on modest GPUs, supporting up to 3 reference images for consistent edits. Speed-focused models like Flux 2 | Turbo | Edit (Image to Image), Flux 2 | Turbo | Text to Image (Text to Image), Flux 2 | Flash | Edit (Image to Image), and Flux 2 | Flash | Text to Image (Text to Image) prioritize real-time workflows. Premium tiers such as Flux 2 | Max | Edit (Image to Image), Flux 2 | Max | Text to Image (Text to Image), and Flux 2 | Flex (Text to Image), Flux 2 | Flex | Edit (Image to Image) balance quality and efficiency, while Flux 2 | Text to Image Lora (Text to Image), Flux 2 | Lora Edit (Image to Image) enable custom styles via LoRA fine-tuning.

Use cases span creative and technical applications. For designers, Flux 2 | Klein | 4B | Base | Text to Image prototypes concepts rapidly: prompt ""transform this line art into a photorealistic cyberpunk cityscape at dusk, neon lights reflecting on wet streets"" to yield coherent 1024x1024 composites. Developers use Flux 2 Pro | Edit for e-commerce, replacing objects in product photos with prompts like "add floating geometric shapes in pastel colors around the central vase." Models support up to 4MP resolutions, adjustable aspect ratios, 25-50 inference steps, and CFG scales around 4.0-5.0, with native editing for semantic changes, object replacement, and multi-reference composition.

Pipelines combine models seamlessly: start with Flux 2 | Turbo | Text to Image for quick drafts, refine via Flux 2 | Klein | 9B | Base | Edit, and fine-tune styles using Flux 2 | Text to Image Lora. This unified approach runs on 8-12GB VRAM GPUs, with quantized options like FP8/NVFP4 for even lower requirements, producing photorealistic outputs with strong spatial logic and character consistency.

What Makes flux-2 Stand Out

flux-2 sets benchmarks in AI image generation through its rectified flow transformer architecture, outperforming traditional diffusion models in nuance capture for textures, lighting, and arrangements. Key differentiators include unrivaled text rendering (up to 60% first-attempt accuracy), HEX color code support for precise branding, and multi-image reference handling (up to 10 in Dev variants, 3 in Klein) for superior consistency. The Klein models achieve sub-second speeds with NVFP4/NVFP8 acceleration, rivaling larger 32B-parameter systems in Elo quality scores while fitting consumer RTX GPUs. Base models offer maximum control via extended sampling steps, unlike distilled variants focused on velocity, enabling high-fidelity fine-tuning without high-end servers.

Strengths like photorealistic diversity, professional-grade sharpness, and integrated editing make flux-2 ideal for designers prototyping concepts, developers building image tools, marketers creating brand visuals, and researchers experimenting with LoRAs. Its efficiency—0.3-1.2s inference on optimized hardware—combined with robust prompt adherence ensures reliable, controllable outputs for complex scenes, from cinematic landscapes to detailed product variants.

Access flux-2 Models via each::labs API

each::labs is the premier platform for harnessing the full flux-2 family, providing unified API access to all models including Flux 2 Pro, Klein variants, Turbo, Flash, Max, Flex, and LoRA editions through a single, scalable endpoint. Integrate effortlessly with the each::labs Playground for instant testing or the robust SDK for production pipelines, supporting long-polling predictions and custom workflows on eachlabs.ai. Sign up to explore the full flux-2 model family on each::labs.