This is a Anima based, semi-realistic / 2d / 3d / anime-style merge.
THIS IS A TURBO - model, so it works with low CFG 1-2. It contains the excellent n_Arno's turbo lora already embedded.
It's supposed to be "style neutral" in a way, so that style loras can be used and switch to 2d with anime coloring, anime screencap, etc.
Haven't really tested it with high CFG as personally I use Turbo to save time in generations.
This has no "grand direction", it's more like what I'm messing up with at the time.
The attached file is only the model, so you will also need the Anima text encoder and VAE.
2.9B Version -> Mainly converted the previous 1.0 version into 2.9B Anima model with instructions from @n_Arno, from and at the same time made INT8 convrot version.
Convrot conversion made with:
convert-to-quant -i Suomaa_Anima-Turbo_1b-29B.safetensors -o Suomaa_Anima-Turbo_1b-29B-INT8.safetensors --int8 --convrot --convrot-group-size 256 --scaling_mode row --comfy_quant --save-quant-metadata --simple --anima --exclude-layers "(first_stage_model|cond_stage_model)"Overall, just draws things bit differently. No additional training was done on the model itself, just the 2.9B conversion with a script.
From slightly testing it, some images seem better on the 1.0, some on the 2.9B version which changes the composition on some images, and some on INT8 version of it which changes the details. So in the end it's more of a preference.
Notes about the 2.9B dataset, from original Anima-2.9B page:
The dataset does not have quality tag like masterpiece, very awa, high quality, etc.
Dataset was filtered from source instead.
You can and should use absurdres, highres, year_tag.
Knowledge cutoff in July 2026, training data included both new and old samples prior to September 2025
Mixed captioning, including both tags and natural languages, using a mix of Gemini 3.1 Flash-Lite, Gemini 3.5 Flash-Lite, and Claude Sonnet 5
NO scoring
For Anima 2.9B quality improvements, you may also want to try RDBT 2.9B:
You need to load both "Distillation add-on" LoRA and the "finetuned v2" LoRA to get best quality.