EveryLoRA

• Works with 99% of LoRAs I've tested, Pony-based and non-Pony.

• Supports booru artist tags and characters normally obfuscated in Pony.

• Properties of both artist-style and Pony models.

• Merge of CashMoney v4 (Pony) and NAI3LikeDiffusion (SDXL anime).

• Uses a special compatibility text-encoder block which can be used to merge Pony-based models with non-Pony models (instructions below)

• Will take tags quite literally, (so for example "medium breasts" without a clothing tag will lead to nudity, Pony score_X tags will work but will give you a scoreboard for example - do don't use them, use of strengthening "((tag))" and will usually result in noise). This is because I've amplified the tags in NAI3Like which allows artist styles to carry through, otherwise they'd wash out in the merge.

This took a lot of experimentation, so I'm releasing the model under Early Access for 500 buzz. As a bonus if the goal of 200,000 buzz is reached, I'll write a short post on how I've made the compatibility TE.

Instructions for merging Pony and non-Pony models using the EveryLoRA compatibility TE, using SuperMerger:

For example we will merge AutismConfetti with AnimagineXL. I've tested this, it works fine: advokat/AnimagineAutismConfettiMix at main (huggingface.co)

The results are a bit boring though, so you may want to try other models. Some other merges: https://civitai.com/models/755414

Step 1

Model A: AnimagineXL

Model B: EveryLoRA

Use Weight sum + MBW: 1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0

This transfers the EveryLoRA TE to Animagine.

= INTERMEDIATE_MODEL

Step 2

Model A: INTERMEDIATE_MODEL

Model B: AutismMixConfettiMix

Use Weight sum + MBW: 0,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5

This merges animagine and autism as 0.5 weight while keeping the EveryLoRA TE.

= Pony + Non-Pony merge

If you get muted/grey results you need to bake in the SDXL VAE in the options.

描述:

训练词语: