Today, we are proud to release FLUX.1 Kontext - a suite of generative flow matching models to support image generation and editing. Unlike existing text-based image generation models, the FLUX.1 Kontext family supportscontext-sensitiveImage generation, which can extract and modify visual concepts to generate coherent new images by co-composing cues with text and images.
Consistent, context-aware text and image generation and editing.
Your images. Your language. Your world.
FLUX.1 Kontext is a significant extension to the traditional text-generated image model that combines on-the-fly text-based image editing with text-generated image capabilities. As a multimodal flow model, it leads the way in maintaining character consistency, understanding context, and local editing capabilities, while providing powerful text-generated image capabilities.
Enhanced text-to-image capabilities
Whether it's for creative ideation, sketching, conceptual design, or just for fun, text-to-image is still a core element of today's image generation, and the FLUX.1 Kontext model delivers leading edge cue-following, photo-realistic rendering, and competitive font generation performance, with inference speeds of up to 8x faster than today's dominant models, such as GPT-Image.
Play it, create it, get your hands dirty ......
The FLUX.1 Kontext model does more than just generate images from text. Unlike traditional streaming models that can only be generated based on text, FLUX.1 Kontext also understands and creates based on existing images. Simple text commands are all that's needed to modify the input image, resulting in a flexible and instant image editing experience - without the need for fine-tuning or a complex editing process.The core capabilities of the FLUX.1 Kontext suite include:
- Character Consistency: maintains the stability of unique elements of an image (e.g., characters or objects) across different scenes.
- Localized editing: make precise changes to specific elements of an image without affecting other parts.
- Style Reference: Generate a new scene that retains the unique style of the reference image, guided by textual cues.
- Interaction speed: Very low latency in image generation and editing.
...... and supports iteration: step-by-step modifications
Flux.1 Kontext supports step-by-step addition of commands through multiple rounds of editing, continuously optimizing creations with very low latency while maintaining image quality and character consistency.
FLUX.1 Kontext [pro] model
As part of the FLUX.1 Kontext suite, we are introducing two contextual image models into the BFL API.
- FLUX.1 Kontext [pro] -- a pioneer in fast iterative image editing
The model combines localized editing, context-based generative modification, and classic text-to-image functionality to deliver the signature quality of FLUX.1. FLUX.1 Kontext [pro] supports both text and reference image inputs, allowing precise editing of specific areas of the image, as well as complex transformations of entire image scenes. The inference speed is an order of magnitude faster than existing state-of-the-art models, and it is the first model to support multiple rounds of iterative modification by the user, while maintaining consistency of character, identity, style, and distinctive traits across different scenes and perspectives.
- FLUX.1 Kontext [max] -- Extreme Performance at Extreme Speed
This experimental new model delivers vastly improved performance in terms of cue adherence, font generation and editorial consistency, while maintaining super-fast speeds with no compromises.
FLUX.1 Kontext [max] and FLUX.1 Kontext [pro] are now live and can be experienced on the following platforms:KreaAI,Freepik,Lightricks,OpenArt,LeonardoAI, and our infrastructure partners:FAL,Replicate,Runware,DataCrunch,TogetherAI together with ComfyOrg. We are particularly grateful to OpenArt together with KreaAI Support in preference data collection.
FLUX.1 Kontext [dev] Private Beta is open for application!
We strongly believe that open research and weight sharing are the cornerstones of security technology innovation. We have developed an open weights version - FLUX.1 Kontext [dev], a lightweight 12B diffusion transformer suitable for customized development and compatible with the previous FLUX.1 [dev] inference code. FLUX.1 Kontext [dev] is now open as a private beta for scientific use and security testing. If you are interested, please contact kontext-dev@blackforestlabs.ai. Upon official release, the model will be available through the FAL,Replicate,Runware,DataCrunch,TogetherAI together with HuggingFace Distribution.
Performance Evaluation
To validate the performance of the FLUX.1 Kontext model, we performed a comprehensive evaluation of theTechnical ReportThe release is in the middle of the day. Here's a brief summary: We built the KontextBench benchmarks covering real-world text-to-image and image-to-image tasks, which will be released later.
We show the results of the evaluation in the six contextual image generation tasks.FLUX.1 Kontext [pro] consistently ranks among the top in all tasks, especially scoring the highest in text editing and character retention
We evaluated the image-generating image models, including FLUX.1 Kontext, on six KontextBench tasks. FLUX.1 Kontext [pro] performed consistently across all tasks, with the highest scores for text editing and character retention (see above), and consistently outperformed other state-of-the-art models in terms of inference speed (see below).
The FLUX.1 Kontext model consistently has lower latency than other state-of-the-art models in the text-generated image (left) and image-editing (right) tasks
We evaluated the text generation image model on multiple quality dimensions, and FLUX.1 Kontext performed competitively on the Aesthetics, Cue Following, Font Generation and Realism benchmarks.
Left: Enter image; Center: Edit: "Tilt her head toward the camera"; Right:"Make her smile."
Left: Enter image; Center: Edit: "Change 'YOU HAD ME AT BEER' to 'YOU HAD ME AT CONTEXT'"; Right:"Change the scene to a nightclub."
Failure Cases:
The current version of FLUX.1 Kontext still has some limitations. Too many rounds of iterative editing may introduce image artifacts that degrade the image quality; the model occasionally fails to follow the cues accurately and ignores some specific requirements; the limited world knowledge affects its ability to generate contextually accurate content; and, in addition, the distillation process may trigger output artifacts that affect the quality of the generation.
Schematic of the failure case of FLUX.1 Kontext: after six rounds of iterative editing, the image quality deteriorates and significant artifacts appear.
FLUX API Demo: BFL Playground Debuts!
Since the release of our product, we've continued to receive feedback from people wanting to streamline the model testing process. Today, we are officially launching the FLUX Playground - a convenient interface for testing our state-of-the-art FLUX models without technical integration.
Playground allows developers and teams to validate use cases, demonstrate capabilities to stakeholders, and experience advanced image generation in real-time. Whether evaluating technical feasibility or demonstrating results to decision makers, Playground provides an instant validation portal before full access to the API.
At BFL, our mission is to build state-of-the-art media generation models and infrastructures.Playground is an introductory platform to the BFL API, designed to accelerate the process from evaluation to production deployment. Access it now:https://playground.bfl.ai/
English 
简体中文 
日本語 
Deutsch 
Português do Brasil