For those don't want to read: This is a distilled model, you need to lower CFG scale to 1~1.5.
RDBT [Anima]
See the main page for more info. (usage, faq, int8 ckpt...)
See this page for update log and version info.
Sharing merges using this model is not allowed. This "restriction" won't affect anyone. It's only aimed at those who steal others' models to sell.
This is a fully finetuned model, but in LoRA. It needs the correct base model to function properly. Don't mismatch base models. If you don't know how LoRA works or what this means, use the ckpt version.
Links to the LoRA base models:
AnimaYume (ym) (hf link) (civitai link) (int8 version).
Anima pretrained (b1,p3,p2,p1) (hf link)
FYI: "Merged" base model usually is NOT recommended. It's weights might be very messy. Probably merged different pretrained versions. A nightmare for all LoRAs, not just this LoRA.
Description
FAQ
Comments (14)
Only tested p3 v.0.24f dmd2 b so far. But stacks nicely with Preview3 trained character Loras.
Special prize for the most obscure version names.
One random thing I've noticed is that wide resolutions (e.g. 1344 x 768) have a tendency to produce realistic images if no artist or style is prompted. Not really a problem, just interesting!
Has anyone done a comparison between this and the distill LoRA for Cosmos?
If I blend dmd2 and f_cfg_b using different weights (like 0.75 and 0.35), would the results be better? Also, would this setup work with 8 steps and a CFG of 3?
Can confirm dmd2_b is an improvement, keep up the good work!
use this dmd2 lora with positive weight(0.5) and cosmos dmd2 with same negative weight(-0.5). It is such a glow up!
Hm, interesting. Let's see what that does over running a negative cfg distillation instead
Hm, is that done in a normal CFG and high stepcount?
v0.29b minus ('anima turbo 0.1' x 0.33) plus (v0.6cfg_distill x 1.0) (3x the anima turbo value roughly cancels out) - does result in something Interesting
RDBT has the superior compositions and Tdrussel's officla turbo LoRa has the superior style preservation (especially when you starto to play with low or even very low shift values). Combing both traits into one and we could have the perfect turbo option for Anima.
I know this is not suppose to be like this, but RDBT and anima turbo are like natural complements.
Anima turbo has biases of high-brightness, high-saturation, simple images. RDBT has biases of low-brightness, low-saturation, complex images.
Mixing these two together produces ... the most perfect model I've ever used.