jrrjrr's picture
Upload README.md
db77343
|
raw
history blame
2.73 kB
metadata
license: creativeml-openrail-m
tags:
  - coreml
  - stable-diffusion
  - text-to-image
  - not-for-all-eyes

LoRA-Merged Models For CoreML

Stable Diffusion v1.5 Type Models With Embedded LoRAs For Use With CoreML-based Apps

Models with an embedded LoRA are a workaround for achieving the effects provided by a LoRA with a CoreML based Stable Diffusion pipeline. An alternative to the direct use of LoRAs is necessary because Apple's CoreML Stable Diffusion implementation does not provide native support for LoRAs.

The LoRA effect in these models works best when the "trigger-words" that were used in the LoRA's training are included in the text prompt. Conversely, the effect of the embedded Lora can be minimized by including the trigger-words in the negative prompt, but in general, there will be residual effects from the LoRA present in any image generated by these models.

A merge is most practical when the model's basic characteristics match the LoRA's intended effect. A LoRA trained on a particular anime character should be mated with an anime-focused base model, a NSFW niche LoRA should be embedded in an NSFW model, etc.

The listings below indicate the LoRA and the base model used in each merge. The folder for each merged model typically contains builds at several image sizes, sample image file(s), and a text file with the suggested trigger-words from the embedded LoRA. (Some LoRAs don't use trigger-words.)

If there is a merge you'd like me to run for you, leave a note in the Community area here. I'm open to anything that is SD-1.5 type. SD-2.1 type base models are really not suppored by Mochi Diffusion, and require pipeline modifications on my end in order to convert them, so I'm staying away from them. Let me know the LoRA, the base model, and what size(s) you'd most like.

Take care in opening the sample image and trigger-word files here as they may contain words and images that you find offensive. Caveat emptor.

List Of LoRA-Merged Models