F Lite 7B

🚀 F Lite 7B Model Card

F Lite 7B is a 7B parameter diffusion model jointly developed by Freepik and Fal. It is created through knowledge distillation from F Lite, offering comparable image quality to its larger counterpart while being faster and more memory - efficient. Moreover, it maintains the same copyright - safe and SFW generation features as F Lite.

🚀 Quick Start

Try F Lite 7B instantly through our interactive demo on Hugging Face. F Lite 7B is compatible with both the diffusers library and ComfyUI. For detailed information, refer to the F Lite GitHub repository.

✨ Features

• Knowledge Distillation: Built from F Lite via knowledge distillation, it provides similar image quality.
• Efficiency: Faster and more memory - efficient compared to its larger sibling.
• Copyright - Safe and SFW: Preserves the copyright - safe and SFW generation characteristics of F Lite.

🔧 Technical Details

Read the technical report to gain in - depth knowledge about the model architecture, training recipe, and evaluation.

📚 Documentation

Limitations and Bias

• Malformations: Similar to its teacher model, it can generate malformed images.
• Text Rendering: Text rendering capabilities are limited.
• Biases: Despite careful curation, societal and aesthetic biases may still appear.

Recommendations

• Prompt Length: Longer, descriptive prompts can enhance image quality. Very short prompts may result in sub - optimal images.
• Resolution: Generate images above the megapixel mark as low resolutions can reduce image fidelity.

📄 License

The F Lite 7B weights are released under the permissive CreativeML Open RAIL - M license. Dependencies inherit their respective licenses (Apache 2.0 for T5 XXL and Flux Schnell VAE).

📚 Acknowledgements

This model depends on high - quality components such as T5 XXL for prompt encoding and the Flux Schnell VAE for efficient latent - space compression.

📚 Citation

@article{ryu2025flite,
  title={F Lite Technical Report},
  author={Ryu, Simo and Pengqi, Lu and Mart\'in Juan, Javier and de Prado Alonso, Iv\'an},
  year={2025}
}

⚠️ Important Note

The model has limitations such as the potential to generate malformations, limited text rendering capabilities, and possible societal and aesthetic biases.

💡 Usage Tip

Use longer, descriptive prompts and generate images above the megapixel mark for better results.