What is copyright-safe?
I noticed that the F-Lite model is described as being trained on “copyright-safe” images. I understand that this likely refers to images from Freepik’s internal dataset, but I would like to clarify the specific meaning of “copyright-safe” in this context.
Could someone please elaborate on:
• Does “copyright-safe” mean that all training images are either in the public domain, under permissive licenses (e.g., CC0, CC-BY), or otherwise legally cleared for use?
• Is there any official documentation or definition from Hugging Face or Freepik that outlines what qualifies as “copyright-safe”?
• Are there any legal considerations or potential risks associated with using models trained on such datasets, especially for commercial applications?
Understanding this would help ensure proper compliance in projects where copyright considerations are critical.
Thank you in advance for your assistance!
Keen to learn more about this too
What we specifically mean is that we have trained a DiT-based model using images from which we had the right to train a model on. Note that there are other external models also involved in the generation though, the VAE from Flux Schnell and the T5 XXL from Google, from which we don't know what they were trained on.
So the DiT part was trained on images that are okay to use—thank you for clarifying.