metadata
license: creativeml-openrail-m
tags:
- stable-diffusion
- text-to-image
MEGA MERGE DIFF (MMD) VERSION 1-18. 18 MERGED MODELS IN ONE
ANNOUNCEMENT:
- DUE TO THE FACT THAT I CANNOT SEEM TO CATCH A BREAK AND GET SOME TIME TO ASSUAGE MY OWN INSECURITIES ABOUT THE QUALITY OF THIS MODEL, I AM JUST GOING TO RELEASE IT. FIRST MODEL RELEASE: MMD V1-18 MODEL MERGE ALPHA:
- LIST OF MERGED MODELS: https://discord.com/channels/900672465276116994/1035895704377368687
- DISCORD INVITE: https://discord.gg/GVJWxB3Q
- DOWNLOAD LINK: https://huggingface.co/ShinCore/MMDv1-18/tree/main
NOTES ABOUT USAGE:
- MODEL CAN BE A BIT HARSH AND RIGID, BUT PUT OUT SOME AMAZING GENS. 2ND RELEASE WILL BE LESS EXTREME.
RECOMMENDED SETTINGS:
- IMG HEIGHT/WIDTH MUST BE SET TO A MULTIPLE OF 128
- SET HIRES FIX TO ON. SET FIRST PASS HEIGHT/WIDTH TO HALF OF IMG HEIGHT/WIDTH.
- PLAY WITH DENOISING STR, I SET MINE TO .69, YMMV.
- MODEL IS SENSITIVE TO CFG. I USUALLY USE 12.5, BUT OTHERS HAVE REPORTED BETTER OUTPUTS AT LOWER/HIGHER VALUES. TRY THEM OUT.
- SET RESIZE SEED OPTION TO 512*512. (DDIM RESPONDS BETTER TO THIS SETTING. OTHER SAMPLERS MAY NOT. TURN ON AND OFF AND COMPARE)
PROMPTING:
- MERGED MODELS USE A COMBO OF STANDARD SD BLIP/CLIP AND DANBOORU TAGS. I USE BOTH IN MY PROMPTS.
2ND RELEASE, MMD V1-18 MODEL MERGE (TONED DOWN) ALPHA: THIS IS THE SAME AS THE FIRST RELEASE,BUT I MERGED BACK IN 25% OF SD 1.5. MORE FORGIVING, LESS EXTREME EDITION. EVERYTHING ELSE IS THE SAME.