What AI says about Mitsua Diffusion

Mitsua Diffusion One is a latent text-to-image diffusion model trained from scratch on public-domain/CC0 imagery using a fixed OpenCLIP text encoder (ViT-H/14), as a successor to Mitsua Diffusion CC0. It is released as an open-access base model intended for Mitsua’s AI VTuber Elan Mitsua, with further training available on an opt-in basis via a Discord community (Japanese). The model is currently noted as low quality and limited in diversity; training data are ethically sourced from CC0/open-access institutions, and it is licensed under the Mitsua Open RAIL-M license with usage restrictions, including restrictions on illegal outputs and redistribution of weights for commercial use under the same terms.