Flux 1 Image generator – Black Forest Labs Open Source model

Black Forest Labs, founded by a team of distinguished AI researchers and engineers behind the groundbreaking Stable Diffusion project, has rapidly captured attention in the AI world. Their recent surge in popularity stems from a viral AI video that showcased the stunning realism achievable by combining Flux 1’s images with Runway ML’s animation capabilities.

Riding this wave of success, Black Forest Labs secured a high-profile partnership, powering the image generation features of Elon Musk’s newly launched Grok 2 chatbot on X. However, this collaboration has ignited controversy due to the lack of safeguards in Grok 2, leading to concerns about the potential for generating and spreading misleading or harmful content (now available in visual form too).

Nevertheless, Black Forest Labs is backed by a successful $31 million seed funding round led by Andreessen Horowitz, with notable participation from angel investors and follow-up investments from General Catalyst and MätchVC, the lab is poised to make a significant impact. With an advisory board boasting industry veterans like Michael Ovitz and AI pioneers like Prof. Matthias Bethge, which makes it a startup firmly positioned to drive innovation and accessibility in the field of generative AI.  

Review and Comparison:

• Unmatched Image Quality and Diversity: Flux 1 models generate images with exceptional visual quality, detail, and adherence to prompts. They consistently outperform popular models like Gemini, DALL·E 3 (HD), and SD3-Ultra in visual quality, prompt responsiveness, output diversity, aspect ratio variability, and typography.  
• Open-Source Accessibility: Flux 1 offers an open-source model (Flux.1 [schnell]) under an Apache 2.0 license, empowering the AI community to customize and build upon its capabilities. This commitment to accessibility and transparency aligns with Black Forest Labs’ core belief in fostering innovation and collaboration.
• Efficiency at Scale: Flux 1 models are based on a hybrid architecture of multimodal and parallel diffusion transformer blocks, scaled up to 12 billion parameters. They leverage flow matching for improved training and generation efficiency, incorporating rotary positional embeddings and parallel attention layers for enhanced performance.
• Speed: The Flux.1 [schnell] model is specifically designed for speed, outperforming even powerful non-distilled models like Midjourney v6.0 and DALL-E 3 (HD) in the few-step model category.
• Less guardrails: Flux.1 allows generating images that Dall E 3 and Gemini refused to create such as the one below. Moreover, you can generate NSFW with it, but there are no examples to be added in this article.

Flux.1 Model Family:

Flux 1 offers three variants, each catering to different needs:

1. Flux.1 [pro]: The flagship model, delivering state-of-the-art performance with exceptional image quality, detail, and diversity, perfect for professional applications. Access it via their API, Replicate, or fal.ai. They also offer dedicated enterprise solutions.
2. Flux.1 [dev]: An open-weight, guidance-distilled model for non-commercial applications. It offers similar quality to the pro model but is more efficient. Available on HuggingFace, Replicate, and Fal.ai.
3. Flux.1 [schnell]: The fastest model, designed for local development and personal use. It is openly available under an Apache2.0 license, with weights on Hugging Face and inference code on GitHub and HuggingFace’s Diffusers.

Access Guide for Beginners:

1. GoEnhance AI, Replicate, or Fal.ai: For beginners, these platforms offer user-friendly interfaces to experiment with Flux.1 models.
2. Hugging Face: Access the open-source “dev” and “schnell” models and community resources on Hugging Face.
3. Local Installation: If you’re comfortable with technical setups, you can install and run the “schnell” model locally using the provided code on GitHub or HuggingFace’s Diffusers.

Training and Licensing:

Flux 1 models are trained on a massive dataset of images and text. The specific training details are not fully public, but the models utilize innovative techniques like flow matching, rotary positional embeddings, and parallel attention.  

• Flux.1 [pro] and [dev]: Licensing details for commercial and non-commercial use can be found on their website or by contacting them directly.
• Flux.1 [schnell]: Available under the permissive Apache 2.0 open-source license.

Head-to-head Image generation comparison

A hyper realistic image taken with a DSLR camera of a busy street in central London

Cyber punk futuristic bar where aliens are drinking – neon lights, run down 

Conclusion

Flux 1 represents a major step forward in open-source AI image generation, fueled by the expertise and vision of Black Forest Labs. With its commitment to quality, efficiency, and accessibility, Flux 1 is poised to empower creators and researchers alike. As Black Forest Labs continues to push the boundaries of generative AI, we can anticipate even more impressive innovations in the future.