Black Forest Labs FLUX API

FLUX 3 is Black Forest Labs' multimodal foundation model, a change from the company's previous FLUX.1 and FLUX.2 releases, which were image-only. It is trained jointly on images, video, and audio using an architecture BFL calls Self-Flow, and is the company's first model to generate video and audio from the same set of weights. The video capability supports text-to-video, image-to-video, video-to-video editing, video continuation, and keyframe-based transitions, producing clips up to 20 seconds long in HD (1 megapixel per frame) or Full HD (2 megapixels per frame). It generates optional native audio alongside the video, including multilingual dialogue, sound effects, and ambient noise. Black Forest Labs reports FLUX 3 Video ranks highest in Elo ratings for text-to-video and image-to-video among models it compared, including Gemini Omni Flash, Minimax H3, and Seedance 2.0. FLUX 3 Video became generally available on August 5, 2026.

FLUX.2 Klein 4B is a compact, Apache 2.0 licensed model distilled from the full FLUX.2 architecture, capable of sub-second image generation on consumer GPUs with ~13GB VRAM. It supports text-to-image, image editing, and multi-reference generation in a unified model.

FLUX.2 Klein 9B is a larger variant of the Klein family built on a 9B flow model with an 8B Qwen3 text embedder, matching or exceeding models 5x its size in quality. It offers higher output diversity than the 4B distilled variant and is ideal for fine-tuning and research.

FLUX.2 [klein] 9B is an image generation and editing model from Black Forest Labs, released on January 15, 2026. It combines a 9B flow model with an 8B Qwen3 text embedder and is step-distilled to 4 inference steps, so it returns results in under a second. It handles text-to-image generation and multi-reference image editing in a single model. It follows prompts more closely and produces higher quality than the 4B Klein variant, and Black Forest Labs says it matches or exceeds models five times its size. It suits real-time applications, rapid prototyping, and latency-sensitive workflows, and is a faster option than the larger FLUX.2 [pro] and FLUX.2 [max] models.

FLUX.1 Kontext Dev is an open-weight 12B parameter model for in-context image generation and editing, allowing prompting with both text and images to modify visual concepts. It was the first open model to deliver proprietary-level image editing performance and runs on consumer hardware.

FLUX.1 Kontext Max is the highest-quality model in the Kontext series, optimized for iteratively modifying existing images via text prompts with maximum fidelity. It offers the best editing consistency and prompt following among Kontext variants.

FLUX.1 Kontext Pro is a production-grade in-context image generation and editing model that balances quality and speed. It powers integrations in Adobe Photoshop's Generative Fill and Meta's platforms.

FLUX.2 Dev is a 32B parameter open-weight flow matching transformer for text-to-image generation and multi-reference image editing. It uses Mistral-3 24B as its vision-language backbone and features a new VAE, improved typography, and support for up to 10 reference images.

FLUX.2 Flex is a specialized FLUX.2 variant focused on typography accuracy and fine detail preservation, with a configurable 'steps' parameter for trading off between quality and speed. It exposes advanced controls like guidance and step count for precise diffusion tuning.

FLUX.2 Max is the most capable model in the FLUX.2 family, delivering the highest editing consistency, strongest prompt following, and best photorealism. It uniquely supports grounded generation with real-time web search to visualize current events and trending content.

FLUX.2 Pro is the production-grade FLUX.2 model balancing high quality and affordability, designed for professional image generation and editing workflows. It uses fixed optimal inference parameters for consistent output without manual tuning.

FLUX.1 Canny [dev] is a 12 billion parameter open-weight model from Black Forest Labs that generates images conditioned on a canny edge map extracted from an input image, alongside a text prompt. It preserves the structure and composition of the source image while a prompt drives style, texture, and content changes, useful for turning sketches into finished art, retexturing objects, or controlled style transfer. The model was trained with guidance distillation, making it more efficient than a standard classifier-free-guidance model of the same size, and its weights are downloadable from Hugging Face. Black Forest Labs positions it just below the production Canny [pro] variant in output quality.

FLUX.1 Canny [pro] is Black Forest Labs' production-grade structural conditioning model, generating images that follow the edges of an input image's canny map while a text prompt controls style and content. Black Forest Labs describes it as the best-in-class model for canny-edge-guided generation, ahead of the open-weight Canny [dev] variant in quality. It's served through the BFL API rather than distributed as open weights, aimed at teams that want production reliability without managing model weights themselves. Typical uses include architectural visualization, product retexturing, and converting rough sketches into polished renders while keeping the original composition intact.

FLUX.1 Depth [dev] is a 12 billion parameter open-weight model from Black Forest Labs that conditions image generation on a depth map extracted from an input image, combined with a text prompt. It keeps the spatial layout and perspective of the source image while a prompt changes surface appearance, objects, or style, suited to retexturing and scene-restyling tasks where geometry needs to stay fixed. Like Canny [dev], it was trained with guidance distillation for more efficient inference than a standard model of its size, and its weights are available on Hugging Face. Black Forest Labs reports its Depth models outperform proprietary alternatives such as Midjourney's ReTexture feature.

FLUX.1 Depth [pro] is Black Forest Labs' production-grade depth-conditioned image generation model, using a depth map from an input image to guide the structure of new output while a text prompt drives its content and style. It targets the same retexturing and scene-restyling use cases as Depth [dev] but is served through the BFL API for production workloads instead of distributed as open weights. Black Forest Labs reports its Depth models outperform proprietary tools such as Midjourney's ReTexture feature at preserving structure during text-guided edits.

FLUX.1 Fill [dev] is a 12 billion parameter open-weight inpainting and outpainting model from Black Forest Labs, editing or expanding an image using a binary mask and a text prompt. It can remove objects, add new elements inside a masked region, or extend a canvas outward, while maintaining lighting, perspective, and scene consistency across the edit. Black Forest Labs reports Fill [dev] outperforms other proprietary inpainting solutions while being more efficient at inference, ranking just behind the production Fill [pro] variant in quality. Its weights are available on Hugging Face for teams that want to self-host or fine-tune an inpainting model.

FLUX.1 Fill [pro] is Black Forest Labs' production-grade inpainting and outpainting model, editing or expanding an image from a binary mask and a text prompt. Black Forest Labs describes it as a state-of-the-art inpainting model, outperforming competing methods including Ideogram 2.0 at understanding complex scenes and text while keeping edits consistent with the rest of the image. It handles object removal, adding new elements, and canvas expansion, preserving lighting and perspective around the edited region. It's served through the BFL API for production use, ahead of the open-weight Fill [dev] variant in output quality.

FLUX.1 Redux [dev] is an open-weight adapter from Black Forest Labs that takes an image as input and produces variations of it, optionally guided by a text prompt for restyling. It works by aligning image embeddings from a SigLIP encoder with the T5 text-embedding space used by the FLUX.1 base models, letting an input image condition generation the way a text prompt normally would. Redux is built to combine with Black Forest Labs' other FLUX.1 Tools, such as Fill or Canny, inside a single workflow, and can mix multiple input images together. It's the dev-licensed counterpart to Redux [pro], suited to self-directed restyling and remixing workflows rather than production deployments.

FLUX.1 Redux [schnell] is Black Forest Labs' speed-optimized image-variation adapter, built on the fast FLUX.1 Schnell base rather than the higher-quality Dev base used by Redux [dev]. Given an input image, it generates variations or restyled versions guided by an optional text prompt, trading some output quality for lower latency. Black Forest Labs positions it for rapid prototyping and creative exploration, such as real-time creative tools or fast iteration loops, where turnaround speed matters more than the highest fidelity output. It's released under the permissive Apache 2.0 license, consistent with the rest of the Schnell family.

FLUX1.1 [pro] Ultra is Black Forest Labs' high-resolution mode built on top of FLUX 1.1 Pro, generating images at up to 4 megapixels, four times the resolution of the standard model. Generation still takes about 10 seconds per image; Black Forest Labs reports this is over 2.5x faster than comparable high-resolution offerings from other providers. Ultra mode includes an optional Raw mode toggle that favors a candid, less processed photographic look over the typical polished AI aesthetic, with reported gains in subject diversity and realism for nature photography. It supports flexible aspect ratios and suits product photography, marketing images, and other cases where resolution and photorealism matter more than low latency.

FLUX 1.1 Pro is an improved flagship model released October 2024, offering better quality and efficiency than the original FLUX.1 Pro. It added Ultra mode for 4x higher resolution (up to 4MP) and Raw mode for hyper-realistic candid photography-style images. Generation time is approximately 10 seconds per sample.

FLUX.1 Dev is a 12B parameter open-weight text-to-image model released under a non-commercial license. It offers quality comparable to DALL-E 3 and Midjourney 6 in prompt fidelity and photorealism, and is the most popular open image model globally. It's designed for developers and researchers to run on consumer hardware.

FLUX.1 Dev LoRA is a fine-tuning adapter layer built on top of FLUX.1 Dev, enabling customization of image generation for specific styles, subjects, or concepts. It allows developers to train lightweight adaptations without retraining the full model.

FLUX.1 Krea Dev is an open-weight text-to-image model developed in collaboration with Krea AI, trained to achieve better photorealism and more varied aesthetics than standard FLUX.1 models. It overcomes the oversaturated 'AI look' common in other generators.

FLUX.1 Pro is Black Forest Labs' original flagship proprietary text-to-image model, offering high prompt fidelity and photorealistic output comparable to Midjourney 6. It supports fine-tuning via the FLUX Pro Finetuning API for enterprise customization.

FLUX.1 Schnell (German for 'fast') is the speed-optimized variant of FLUX.1, designed for rapid image generation with lower latency at the cost of some quality. It is the most permissively licensed model in the FLUX.1 family, released under Apache 2.0. Ideal for real-time applications and high-throughput workflows.