1.2 - [LoRA version] - Extraction from the Dreambooth trained checkpoint available as Version 1.1

------

High Quality Dreambooth (JP's repo) trained LoRA of musician Ed Sheeran. Created using recent hi-resolution images of the television newscasting star.

Trigger with "Edsheeran Person" (yes, put person also). Use "Hi-Res fix" with the settings you'd normally use for best results. Use prompt weighting and/or Controlnet to get maximum flexibility out of this likeness model.

Trained locally using JP's repo (24gb VRAM req / arguably the best for likenesses). LoRA was extracted directly from the checkpoint trained by Dreambooth available as Version 1 (best possible quality). The A1111 extension "Supermerger" was used to extract/create the LoRA file.

The checkpoint (Version 1) is a clone of what was trained and thus should produce slightly better results. That said in generating samples the LoRA held up really well - so pick the one that makes more sense for your workflow.

Some samples images were generated w/ Controlnet for initial guidance so you won't be able to replicate them with just the metadata (don't think Civitai lists whether or not CN was used).