losslessmix
losslessmix copied to clipboard
Mixing models of stable diffusion without weights loss
Please use supermerger project: https://github.com/hako-mikan/sd-webui-supermerger
It has improved version of cosine similarity with awesome UI
Algorithm that allows you to mix models without loss of quality.
TLDR; Mixing models by cosine similarity
Usage
python3 weightedsim.py openjourney-v2.ckpt EimisAnimeDiffusion_1-0v.ckpt
Example
openjourney-v2
eimisanimediffusion_1-0v
weighted sum merge
weighted sim merge
Prompt
25 year old woman, white top, blue shorts, adorable face, piercing eyes, resting mouth, intricate necklace, short hair, looking away, standing next to beach, sunset, palm tree
Negative prompt: deformed, bad anatomy, disfigured, mutation, extra limb, ugly, poorly drawn hands, missing limb, floating limbs, disconnected limbs, malformed hands, blurry, ((((mutated hands and fingers)))), distorted hands, amputation, missing hands, double hands, watermark, censored, black and white, sepia, zombie
Steps: 28, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 0, Size: 576x832, Model hash: 628090e8bb, Model: merged-0.5, ENSD: 31337
The list of models that used this script to create them:
PS
Please stop asking me how to run this and how to work with it. I wrote it without knowing Python. This is my second and hopefully last Python script.
My Doge Wallet
DEw2DR8C7BnF8GgcrfTzUjSnGkuMeJhg83