Black Forest Laboratory recently released a new FLUX.1Tools toolkit, which aims to improve the controllability of its basic Vincent graph model FLUX.1 and enable accurate modification and recreation of real and generated images. The toolkit contains four main functions: FLUX.1Fill (image repair and expansion), FLUX.1Depth and FLUX.1Canny (using image structure information to guide image generation), FLUX.1Redux (generating image variants). Each tool provides open source and API versions, and has provided services through multiple platforms to provide users with a more convenient and efficient image editing experience. The following is a detailed introduction to each function.
FLUX.1Tools contains four main functions:
FLUX.1Fill: This is an advanced image repair and expansion tool that goes beyond existing tools such as Ideogram2.0 and the open source AlimamaCreative's FLUX-Controlnet-Inpainting. FLUX.1Fill can edit and expand real pictures and generated pictures based on the text descriptions and binary masks provided by the user, and can seamlessly integrate the modified content into the image to achieve a natural and smooth effect.
After testing, FLUX.1Fill [pro] outperforms all other competing products, making it the most advanced image repair model currently available. The open source version FLUX.1Fill [dev] also outperforms other proprietary solutions and is more efficient at inference.
FLUX.1Depth and FLUX.1Canny: Both tools use the structural information of the image to guide image generation, thus maintaining the original structure of the image texture when modifying it.
Among them, FLUX.1Depth uses the depth information of the image for modification, while FLUX.1Canny uses edge detection information for guidance, which is more suitable for precise detail adjustment.
Evaluation results show that FLUX.1Depth outperforms proprietary models such as Midjourney ReTexture, with FLUX.1Depth [pro] providing greater output diversity and FLUX.1Depth [dev] providing more consistency in depth perception tasks result. For edge detection models, FLUX.1Canny [pro] has the best performance, followed by FLUX.1Canny [dev].
FLUX.1Redux: This is an adapter for all FLUX.1 base models for generating image variations. Given an input image, FLUX.1Redux can make subtle changes to it, generate different versions, and even redesign the image based on text descriptions provided by the user.
In addition, FLUX.1Redux also supports image style adjustment through API. Users only need to provide an image and a text description. This feature has been integrated into the latest FLUX1.1[pro] Ultra model, which combines input images and text cues to create high-quality 4-megapixel output and supports flexible aspect ratios. Test results show that FLUX.1Redux achieves state-of-the-art performance in image variant generation.
FLUX.1Tools provides open source version and API version
FLUX.1Tools is provided in two versions: open source version (FLUX.1[dev]) and API version (FLUX.1[pro]). The API version of each tool is released as the FLUX.1[pro] variant, while the open source version is released as the guided distillation FLUX.1[dev] variant, with inference code and weights provided. Additionally, Black Forest Labs announced that its published models will be available through its partners fal.ai, Replicate, Together.ai, Freepik and krea.ai.
Black Forest Lab stated that the FLUX.1Tools toolkit released this time will inject new vitality into the Flux ecosystem, and they look forward to seeing community users create more exciting works using these new tools.
Model: https://huggingface.co/black-forest-labs/FLUX.1-Redux-dev
Detailed introduction: https://blackforestlabs.ai/flux-1-tools/
All in all, the FLUX.1Tools toolkit has brought new breakthroughs to the field of image editing and creation with its advanced functions and efficient performance. It is worth looking forward to its more possibilities in future applications. Developers can access the links provided for more information and resources.