This page contains RDBT base model, in fp16 and fp8. Also a fp8 "anima preview", just for convenience.
RDBT
RDBT is finetuned Anima with 15~20k images. Captions are from Google Gemini. Comprehensive and accurate.
Also CFG distilled to further improve quality and stability. 2x faster (CFG 1).
It was trained as a LoRA natively. Main page here: https://civarchive.com/models/2364703. Comments and images are also recommended to post at there.
This page contains the full fp16 base model, which = bf16 Anima base model + original fp32 LoRA. In theory, it has tiny bit higher accuracy. Also eliminates LoRA overhead (some systems might generate slower with LoRA).
You don't have to download this base model if you already have the "base model of the LoRA". You can just download the LoRA.
If you don't know which one is the correct "base model", you can download and use this full base model.
preview [fp8]:
fp8 anima preview
About fp8 models:
Quantized fp8 models in ComfyUI format.
Model file contains calibrated metadata for hardware fp8 linear. If you GPU supports it (aka. rtx 4000 and later), ComfyUI will use hardware fp8 automatically, which might (blame Nvidia) be a little bit faster. More about ComfyUI fp8, see ComfyUI docs TensorCoreFP8Layout.
Just ignore ComfyUI log warnings about tons of keys not loaded. Its a small bug in ComfyUI, it checked wrong keys. Those keys are metadata and they are loaded.