Freepik has introduced an open-source AI image generator model named F Lite. This model stands out due to its unique training on a specifically curated dataset, comprised of approximately 80 million commercially licensed and business-safe images sourced internally by Freepik.
Developed in collaboration with the artificial intelligence startup Fal.ai, F Lite comes in two distinctive versions: the Standard Model and the Texture Model. The Standard Model is engineered for general-purpose rendering, offering more predictability in outputs. Conversely, the Texture Model specializes in producing richer textures and enhanced details, though it requires more intricate prompts and might exhibit distortions, making it less ideal for vector-style image creations.
F Lite is a 10 billion parameter diffusion model, a joint creation by Freepik and Fal.ai, meticulously trained on content that is both copyright-safe and suitable for all audiences. A graphics card boasting at least 24 GB VRAM is necessary to run the model natively. Notably, F Lite excels with detailed, extended prompts, enhancing its usability in AI rendering workflows. It is compatible with ComfyUI, a renowned tool for AI rendering workflows, and supports an advanced workflow utilizing SuperPrompt to transform simple text inputs into nuanced outputs.
F Lite is accessible under the permissive CreativeML Open RAIL-M license, encouraging developers to explore and customize the model with their own datasets. Its core components, the T5 XXL text encoder and Flux Schnell VAE, are licensed under Apache 2.0, further facilitating adaptation and innovation by developers.
Yorum Yap