RELEASE ON TENSORART FOR ONLINE GENERATION

This model was not trained using images of celebrities. Please do not generate any images of celebrities using this model. Don't use this model to gen any NSFW content.

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

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

VERISIMILLUSTRIOUS V2 EARLY ACCESS ONLY ON TENSORART

💠VERSIONS

🔷V1:

Initial release
This Version allows Realism style to be trigger with Prompt and keeps the flexibility of Illustrious to create quality Anime and Artist styles.
The Model has an updated characters database thanks to WAI + Raehoshillust models.

🔷V2:

Available for testing in TENSORART, share results here too.
Version 2 improves realism, but default style becomes a bit more semi-realistic.
Characters tags have an influence if you get anime or realism style default style, same with quality tags.
Updated characters data thanks to Raehoshiillust V8.0 amazing work.

__________________________________________________________________________________________________

💠ABOUT THIS MODEL

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.

After 3 months of trial and error, I present Verisimillustrious V1, 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, my laptop limits what I'm able to do unfortunately. 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 and for free Generation at TensorArt

✔Mergers allowed

✔Share art outside CivitAI

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

⚠Availability outside of CivitAI is my prerogative, will make it known in the description

❌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.

V2 TIPS:

Using BREAK before an Artist/Style Tag can diminish their effect
For Realism ethnicities using some Nationalities + gender can change facial structure Ex: Russian girl
Lower Character STRENGHT aka (character:0.9), helps if the hair or face is not realistic enough, you can also bump the Realism tags, it depends on the situation.
Adding extra tags to enforce Realistic style or Anime Styles in both Positive and Negative prompts is useful. Here is an example for Realistic Style::

Positive: hyperrealistic, photorealism, realistic, photo real, Negative: 3D, 2.5D, 2D, unreal, unity, blender, cartoon, anime, cgi, cg, illustration, comic, toon, oil painting \(medium\), flat color, outline, shiny skin, smooth skin, semi-anthro, impasto,

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

__________________________________________________________________________________________________

Verisimillustrious - v1