November 1, 2024

Introducing FLUX.1 Tools

Today, BFL is excited to announce the release of FLUX.1 Tools, a suite of models that enhance the control and steerability of our base text-to-image model, FLUX.1. This update introduces four powerful features: FLUX.1 Fill for inpainting and outpainting, FLUX.1 Depth for structural guidance using depth maps, FLUX.1 Canny for edge-based structural guidance, and FLUX.1 Redux for mixing and recreating input images with text prompts. These tools are available as open-access models in the FLUX.1 [dev] series and through the BFL API as part of FLUX.1 [pro].

What’s New in FLUX.1 Tools

1. FLUX.1 Fill: Inpainting and Outpainting

• Inpainting: FLUX.1 Fill excels at inpainting, which involves removing or replacing parts of an image based on a binary mask and a text description. It outperforms existing tools like Ideogram 2.0 and open-source variants such as AlimamaCreative’s FLUX-Controlnet-Inpainting.
• Outpainting: The model also supports outpainting, allowing users to extend images beyond their original borders seamlessly.

2. FLUX.1 Depth: Structural Guidance with Depth Maps

• Depth Map Integration: FLUX.1 Depth uses a depth map extracted from an input image and a text prompt to guide the generation process. This feature is particularly useful for maintaining structural integrity in complex scenes.
• Use Cases: Ideal for tasks like adding or removing objects while preserving the scene's 3D structure.

3. FLUX.1 Canny: Structural Guidance with Edge Detection

• Edge-Based Guidance: FLUX.1 Canny leverages canny edges extracted from an input image and a text prompt to guide the generation process. This ensures that the generated content aligns with the structural lines of the original image.
• Applications: Useful for tasks requiring precise edge alignment, such as architectural or technical drawings.

4. FLUX.1 Redux: Mixing and Recreating Images

• Adapter Functionality: FLUX.1 Redux is an adapter that allows mixing and recreating input images with text prompts. It’s particularly useful for creating composite images or generating variations of existing content.
• Flexibility: Offers a high degree of flexibility in combining different elements to create new, coherent images.

Performance Benchmarks

We conducted a benchmark comparing FLUX.1 Fill [pro] and [dev] with other leading inpainting models. The results are publicly available here. Key findings include:

• FLUX.1 Fill [pro] outperforms all other competing methods, making it the state-of-the-art inpainting model to date.
• FLUX.1 Fill [dev] outperforms proprietary solutions while being more efficient at inference.

Accessing FLUX.1 Tools

BFL API

• FLUX.1 [pro] Variants: Available through the BFL API, offering best-in-class capabilities for production use.
• FLUX.1 [dev] Variants: Open-access models with inference code and weights available for research and development.

Partnerships

Our released models are also available via our partners:

• fal.ai
• Replicate
• Together.ai
• Freepik
• [krea.ai