April 30th.Freepik Together with AI startup Fal.ai, the latest open-source AI image model, F Lite, is unique in that itThe model is trained entirely on an in-house dataset of approximately 80 million commercial licenses and secure content (SFW).
Freepik's move avoids common legal risks and ensures that the model outputscopyrightSecurity. Officially, the model utilizes 10B parametric diffusion technology to provide users with a high-quality image generation experience.
F Lite comes in two versions: Standard Model and Texture Model. The Standard Model is suitable for general-purpose image generation, with more stable and controllable output results, while the Texture Model focuses on rendering richer textures and details, but is more demanding in terms of cue words, may have distortion problems, and is not suitable for generating vector-style images.
Note: The model's README file states that F Lite requires a graphics card with at least 24GB of VRAM to run locally, and also suggests the use of detailed long cues for best results.
In addition, Freepik provides workflow support for ComfyUI, a popular AI image generation tool, including advanced options for extending simple text input with the SuperPrompt feature.
F Lite's model weights are released under the generous CreativeML Open RAIL-M license, allowing developers to freely use or even fine-tune them with their own data. Its core components, the T5 XXL text encoder and Flux Schnell VAE, are licensed under the Apache 2.0 license, further enhancing the openness and flexibility of the model.