V2.0 (EXPERIMENTAL, V1.0 IS STABLE):
The images were generated with different checkpoints. Upscalers & controlnet were used to enhance the details. Download the images if you want a sample of the workflow used (Comfyui). Don't use embeddings designed for SDXL1.0, they will ruin the images.
------------------------------------------------------
This is my first attempt at creating a style LORA. This LORA is intended to replicate a ‘smoother’ anime aesthetic, minimizing unnecessary complexities in the prompts and overall has smoother outlines compared to the already existing anime screencap style LORAs. READ THE DESCRIPTION!
TRIGGER WORD:
sm_an_stl, source_anime, anime screencap, anime, animification
Personal preferences for the settings:
Steps: 40
CFG: 5-8
sampler: Euler A
Lora strength: 0.7-1
Clip skip: 2 (ComfyUI users: -2)
Suggested prompts:
Positive prompt:
score_9,score_8_up,score_7_up, sm_an_stl, source_anime, anime screencap, anime, animification, (your prompt details), (perspective)Negative prompt:
rough lines,mosaic ceonsoring, censored, text, paint, crayon, graphite, abstract, glitch, 3D, realistic, semi realistic, photorealisticFeel free to experiment with the prompts!
Tips:
-> FOR NSFW: Always include "mosaic censoring" and "censored" in the negative prompt (if undesired)!
-> When using the LORA in combination with a character LORA, 0.7 strength seems to be the sweet spot.
-> Optional positive tags: shiny, shiny skin
-> Optional negative tags: motion blur, motion lines, blur, blurry background, faceless male
-> In some cases (e.g character appearance influence is too strong) it might be better to lower the LORA strength (for ComfyUI users, lower both strength model and strength clip).
-> "AutismMix" and "EDG on pony" seem to have the most stable results.
HUGE thanks to Hilucas for helping me out and facilitating the training.
I will be reading all the feedbacks and viewing the images uploaded to the gallery for future improvements.
Description
Pros:
-Better fingers (hopefully)
-Improvement on style
-Reduced strength (better compatibility with character loras)
-Better compatibility with other pony checkpoints
Cons:
-It is more faithful to the prompt, which means that it requires longer prompts to generate the desired result
-Skin tends to be lighter/paler
-Colors aren't as bright/vivid
FAQ
Comments (4)
I gonna post is as my review of this model.
With V2 model, I get "from below" images. Though I enhance "from front" prompt, it does not work.
Can you dm me the full prompts and settings please?
Really good lora, i just have a question? Why would you need to use score_3, score_2, score_1 when they aren't even used on the base pony diffusion model?
Thank you and you are right, they are not used but I use them out of bad habit. They might even negatively impact the images
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.