This is the 8-Step Turbo-LoRA for Anima-Preview. Lora has a strong style trending! Still in the experimental phase, this version demonstrates turbo-training capabilities within the Anima architecture.
This model was baked by merging my Turbo LoRA and RDBT - Anima LoRA (by @Reakaakasky).
Recommended Settings:
Steps: 8–12
CFG Scale: 1
Samplers: er_sde or res_2m
Original workflows are included in the examples. Please experiment and share your feedback.
Description
Anima-Preview 8 Step Turbo-LoRA
FAQ
Comments (2)
at CFG=1, everything just ends up with this grey-green-brown-ish hue, like a The Matrix filter or something.
turning up the CFG to 3 and lowering the LoRA to 0.5-0.8 however gives some decent results when used with artist tags or other quality LoRAs.
I also found that 8-steps is very optimistic. while the LoRA does counter the haziness of low steps, you still need the extra steps or else you lose a ton of detail. 12 can be done, 15 feels like the sweetspot. that still cuts inference time in half.
Try out my Anima-Preview2 LoRA. It’s working way better for me https://civitai.com/models/2460007/anima-preview2-turbo-lora