Verisimillustrious

TLDR: Surgical Merge of several models to create a model that can do realistic style when prompted while keeping all the 2D styles of Illustrious + new character data.

ALL IMAGES DON'T USE ANY LORAS, ADETAILER OR HIRESFIX, ONLY TEXT PROMPTS

Share your generations or opinions, any tips to improve the model are welcome.

If you would like to donate for me to get better equipment to improve this model or make others in the future, my DMs are open.

💠ABOUT THIS MODEL

After 3 months of trial an error, I present Verisimillustrious V1, Verisimilitude + Illustrious, Verisimilitude means "the appearance of being true or real" hence the name. My goal for this model was to know if there was a way to merge models and not have realism/realistic style be the default. I wanted to keep Illustrious styles as default and just allow Realism style to be prompted. I decided to do mergers because it's faster for me and my laptop limits what I'm able to do. To do this merge I had to modify SuperMerger code a bit to make it easier for me to merge specific transformer block layers and its elements per block, hence the Surgical Merge.

Every time I do a merge, I try a new method, last merge was with DARE, but I didn't feel I had enough control. So, I learned about elements and all the layers inside each block, so I started testing and making A LOT of mistakes for the last 3 months. I might plan to improve this model more, since I think I have a good understanding now of what to do, but I need a faster way to test the changes, so I will see if I can vibe code a ComfyUI node or something.

Either way, I wish I could provide a recipe... but I'm using a modified SuperMerger, so even if I provide it, it might not work. But here is basically what I merged:

MERGE A: illumiyumeXL_v31EPred + (waiNSFWIllustrious_v150 - illumiyumeXL_v31EPred)
MERGE B: MERGE A + (raehoshiIllustXL_v70 - MERGE A)

MERGE C: MERGE B + (realcosplay1536px_v21 - Illustrious-XL-v2.0)

MERGE D: MERGE C + (realcosplay1536px_v21 - Illustrious-XL-v2.0)

Please check the models used in this merge, I wouldn't have been able to create this merge without the help and time they provided by the model's creators. Also my database isn't as updated as the newest versions of some of these models and that was on purpose, so you can download their latest models and try them out.

__________________________________________________________________________________________________

💠PROS:

✔Realism/Realistic style triggers with PROMPT

✔Realism/Realistic style is NOT the default style

✔Updated Characters Data

✔Minimal Body Anatomy modifications to keep artist styles accurate

💠CONS:

❌Realism could be better, but I'm figuring out how to improve it more

❌Some Data could have been lost, not sure which, I couldn't test it completely

❌New Character Data might not be perfect, but with prompting it can

❌There might still be some style leak, but its minimal. (Check TIPS)

__________________________________________________________________________________________________

💠PERMISSIONS AND RESTRICTIONS:

✔Use only on CivitAI

✔Mergers allowed

✔Share art outside CivitAI

✔Credit me or name the model when used in any capacity

❌Use this model or Merges of this model outside of CivitAI

❌Sell Art

❌Sell Model

❌Claim it's your Art

__________________________________________________________________________________________________

HOW TO USE IT:

Follow the typical NoobAi/Illustrious prompt structure.

#gender, character, series, character description, artist tag or style, quality tags, camera tags, theme or background tags, actions, clothing

Check the images for EXAMPLES for prompts, I will try to keep them simple.

TIPS:

• Avoiding quality tags (Masterpiece, great quality, etc) can produce better art styles.

• When using 2D styles, writting realism/3D tags in negative can help and vice versa.

• CFG 6 is a good start, I noticed higher CFG could add more color to the generations.

Settings I used:

Sampling method: Euler a mostly, other Samplers can work
Sampling steps: 20 - 30
CFG Scale: 6 - 10

__________________________________________________________________________________________________

REGARDING AI:
My stance is that AI doesn't create/produce real art, but it allows to materialize concepts/ideas, but I don't consider it art. So, IF I include prompts in the examples, I will most likely not include artist tags in the prompts. It's my way to respect all the awesome artists, so please understand.

__________________________________________________________________________________________________

License

• v1 → based on Illustrious v0.1 → Fair AI Public License 1.0-SD

Licenses of models used in the merger apply to this model too