black-forest-labs/flux-schnell

FLUX.1 [schnell] is a cutting-edge 12 billion parameter rectified flow transformer designed for rapid and high-quality text-to-image generation. This model excels at creating images from textual descriptions, offering a powerful and efficient solution for various creative and practical applications. For a comprehensive overview of FLUX.1 [schnell] and its capabilities, please refer to our detailed blog post.

Core Features and Capabilities

FLUX.1 [schnell] distinguishes itself through a combination of advanced features that contribute to its exceptional performance:

1. State-of-the-Art Image Quality: Experience cutting-edge output quality with remarkable prompt adherence. FLUX.1 [schnell] achieves image generation performance that rivals closed-source alternatives, ensuring high fidelity and detailed results.

2. Ultra-Fast Image Generation: Leveraging latent adversarial diffusion distillation, FLUX.1 [schnell] achieves rapid image synthesis, generating high-quality visuals in a remarkably short span of just 1 to 4 steps. This accelerated generation process significantly enhances workflow efficiency.

3. Open and Accessible Licensing: Released under the permissive apache-2.0 license, FLUX.1 [schnell] empowers users with the freedom to explore its potential for personal projects, scientific research, and commercial applications alike. This open license fosters innovation and broad accessibility.

Getting Started with FLUX.1 schnell

To facilitate seamless integration and experimentation, we provide a comprehensive reference implementation of FLUX.1 [schnell], accompanied by readily accessible sampling code. These resources are available in our dedicated github repository. We encourage developers, researchers, and creative professionals to utilize this repository as a foundational starting point for building and innovating with FLUX.1 [schnell].

Local Inference with ComfyUI

For users seeking local inference capabilities within a node-based visual workflow environment, FLUX.1 [schnell] is fully compatible with Comfy UI. This integration allows for flexible and customizable image generation pipelines, directly on your local machine.

Important Considerations and Limitations

While FLUX.1 [schnell] offers powerful generative capabilities, it's important to be aware of the following limitations:

• Non-Factual Content Generation: It is crucial to understand that this model is not designed to provide or guarantee factual information. Output should not be relied upon for accuracy in real-world contexts.
• Potential for Societal Biases: As a statistical model trained on large datasets, this checkpoint may inadvertently reflect and amplify existing societal biases present in the training data. Users should be mindful of this potential.
• Prompt Adherence Variability: The model may not always perfectly generate output that precisely matches the provided prompts. Variations in interpretation and generation can occur.
• Prompt Style Sensitivity: The effectiveness of prompt following is significantly influenced by the prompting style employed. Experimentation and refinement of prompts may be necessary to achieve desired outcomes.

Responsible Use and Prohibited Applications

To ensure the ethical and responsible application of FLUX.1 [schnell], the following use cases are explicitly prohibited:

The model and any derivatives thereof must not be used:

• In any manner that violates any applicable national, federal, state, local, or international laws or regulations.
• For the exploitation, abuse, or endangerment of minors in any form, including but not limited to the solicitation, creation, acquisition, or distribution of child exploitative content.
• To generate or disseminate verifiably false or misleading information and/or content with the intent to cause harm to others.
• To generate or disseminate personal identifiable information with the potential to cause harm to individuals.
• To engage in harassment, abuse, threats, stalking, or bullying of individuals or groups.
• To create non-consensual nudity or illegal pornographic content.
• For fully automated decision-making processes that negatively impact an individual’s legal rights or create or modify binding obligations.
• To generate or facilitate large-scale disinformation campaigns or malicious content dissemination.

Optimized Inference Performance

For users requiring accelerated inference speeds, we offer a go_fast flag within our API. Activating this flag engages an optimized version of flux-schnell specifically tuned for rapid inference. Currently, this optimized variant incorporates fp8 quantization and an enhanced attention kernel. We are committed to continuous improvement and will provide updates to the model and corresponding documentation as further performance enhancements are developed and deployed.