Sign In

TamReal-Ish

63
472
9
Verified:
SafeTensor
Type
Checkpoint Merge
Stats
313
0
Reviews
Published
Jan 10, 2024
Base Model
SD 1.5
Usage Tips
Clip Skip: 1
Hash
AutoV2
3787D1326A

Now with a pruned version!

v2 has more RealCartoon Anime, mostly in the OUT areas, but some IN as well. Happy to provide details if anyone is interested.

Just what we need, another 2.5D model, but after some kind words I decided to go ahead and upload this. It's a MBW merge of three realistic models, Pirsus Epic Realism merged with Clarity, then that into Divas, with Real Cartoon Anime over that. It uses the base-alpha from Divas. I'll link everything below, and I might throw up the realistic merge later. It hits a nice sweet spot between realism and vibrancy without looking too overcooked the way some do.

Most of the sample images are made with the usual suspects, hires. fix 4x-UltraSharp and ADetailer. Play around with different negative embeddings, I noticed they alter the final image quite a bit. I also use pretty high step counts, compared to most, but I like to 'walk' through my prompts with [different|celebrities|and|descriptions] to get certain looks. From some very limited testing this model should handle booru style tags, but I don't typically use them, I'm a little wordier.

Definitely check out the Divas page, Alex does some really cool work and there are quite a few subjects baked into Divas. I tend to prefer embeddings to Loras when generating celebrities/characters, so I'm not entirely sure how well this model plays with person Loras, but things like clothing etc. seem fine. Anime loras seem to need slightly higher weights, but they work. Play around with the descriptions in your prompts, using 'photoreal, ultra realistic, fine film grain' etc. vs 'anime, illustration, 3d art' definitely tweaks the final image.

Pirsus Epic Realism here

Clarity v3 here

Divas here

RealCartoon Anime v7 here