Steady Diffusion is a latent text-to-image diffusion mannequin educated on 512×512 pictures from a subset of the LAION-5B database.
The mannequin makes use of a frozen CLIP ViT-L/14 textual content encoder to situation its output on textual content prompts.
It incorporates an 860M UNet and a 123M textual content encoder, making it a comparatively light-weight mannequin that may run on a GPU with not less than 10GB VRAM.
The coaching was made potential by a beneficiant compute donation from Stability AI and assist from LAION. For extra particulars, consult with the part under and the mannequin card.
Steady Diffusion was created by a collaboration between Stability AI and Runway and since its launch, Steady Diffusion has garnered important recognition, with over 200,000 builders worldwide downloading and licensing the device.
Further particulars on the venture can be found right here.
