Stevie
She's the sister of Elliot and one of the two protagonists in the 2025 movie In Your Dreams.
I hope you like it and I hope to see some of your pictures made with it!
TAGS
Use the prompts in the example images to get a good baseline.
Here are some useful tags:
stvilf, oversized shirt, leggings
Try: turtleneck
EXTRA
Hi-Res fix recommended.
Try my ILLUSTRIOUS GEHENNA -- a checkpoint for the Illustrious age.
Description
Comments (3)
I think the output would be a lot more flexible in art style if you tagged "3D" in the training data. I've done enough models with polygonal training data to see this work and know that, without it, the model is forever stuck as a weird semi-real. Only negative to this is needing to use "3D" in generation prompt to get it 3D, and I think the loss of one token is worth it.
(Love Gehenna though. Best checkpoint for 3D Illustrious gens I've seen.)
I have had mixed results doing that in the past but maybe I should try again. Thanks for telling me this, I'll try using 3d again for when it fits 😊
Would you recommend it also for more stylized 3d, such as Spider Man Spiderverse movies?
@interfusor In my limited experience the answer/best guess is "yes, but also maybe add another appropriate Danbooru style/medium tag on top of it". It's how I've gotten results with "cel shading" and "low poly"
Edit: This also works to isolate other styles in general (I've gotten it to work with photos of clothing, pixel art, and few other oddities). Your Network Dim/Alpha in training may impact the success of isolating styles by using tag (higher dim+alpha=more likely to have contamination anyways. Only had problems with 32/32 in Anima and dropping those to 8/8 fixed it)