3D ARTSTYLE LORA (LOKR) is on ANIMA 1.0 now!
Use for 3d realistic and 3d cartoonish pics in r34 style. For more specific 3d loras check my profile.
TLDR:
It has the same dataset as my LoRA for NoobAI.
Sadly I didn't have time for adding new pics in dataset because I tested tons of training settings for Anima but I have plans to update it in future for better visual and more characters.
Activation tags:
3d, realistic
OR
3d, cartoonish
Also you can use "3d" without second tag. I prefer 3d, realistic even with wow and overwatch chars.
Optional effects:
shiny skin, wet, sweat, shiny clothes - for effects on characters
sidelighting, backlighting, underlighting - choose one for your scene
bloom, shadow, light paricles, sunlight, moonlight, dappled sunlight, dappled moonlight, dark, day, evening, night - extra lighting effects. Use properly - for example sunlight for day, moonlight for night. Bloom and shadow are good if you have lighting source.
For background:
nature, scenery, blurry background, blurry, depth of field
Characters:
My lora knows many franchises. I'll post character list little later.
Anyway you can prompt every character you want if danbooru knows it and you'll get it in 3d like I got Tohru from Dragon maid.
Settings for Anima image generation:
Read in my article about settings for image generation on Anima.
Special info about 3d lora settings:
Some scenes are better on specific samplers and schedulers.
Usually ER SDE sampler is your friend but sometimes euler a can make better scene in txt2img.
If you use "beta" scheduler don't use too many steps. Usually 25-30 are ok.
Don't beat your head against the wall—if you're having trouble with a specific scene, try a different scheduler.
If the visuals are too contrasty, try lowering the CFG (3D lora allows for a reduction to 2.5-3 CfG).
SGM Uniform is still good scheduler but sometimes you need more time for good scenery.
Description
Anima V1.0 version.
FAQ
Comments (10)
Hell yeah!
3D SEGGGGS❤️🔥
YESSSSSS! YESSSSSS! It's time to create art! (¬‿¬)
Nice! I am waiting for your masterpiece!
@Enigmata Nah, the result is incredible and beautiful!
https://civitai.red/posts/28918628
I'm quite curious about how many images and how many epochs are needed to train a 3D LoRa dataset for optimal results, and whether there are more suitable large 3D models.
It depends on what you train.
I have good 3d output on all character loras where were many 3d art in same style with same character. Good example - Jaina and Lara Croft.
But for style you need to try hard. It is not so difficult to train 3d style if you use came + rex with higher alpha but it is difficult to save characters in lora. That's why I use so many epochs - you need to understand that you can have 5000, 10000, 20000 or more steps, but if you have 2-3 arts with certain character for his appearance only epochs * pic count counts.
I can do loras with less amount of steps but some characters in styles force to feel me like Joel saving Ellie in TLoU because I want to save certain characters in style.
(☆▽☆)
😍 👍🏻
Great! <3
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.



















