Trained on the diffusers library. This lora is all about creating safe and respectful images, and will occasionally touch on artistic nude expression.
In dreams and beyond, illustrations and paintings about cute things.
Description
Updated to the latest captioning model*. The captions were then verified.
The photos and artwork in the datasets are human-made, their image ratios have been kept for better variety.
*Built a custom pipeline to depart from the less performant Joytag, LLama, Phi Vision and Florence models.
The current dreambooth version conveniently doesn't target the first ten layers, only in the Large checkpoints. Thus this release is finally tuned for the Medium version.