Retro Anime SDXL - Style
Things To know:
You should use the trigger word retr0me
Recommended strength is 0.1 - 1.0 (I used 1.0 for the example images)
This LoRA works best in portrait/square images
Works incredibly well with other LoRA's, so please experiment and share what you make!
SDXL Usage Tips :
The LoRA weight I have been using is 1.0 - clip skip 1.0 for testing for the example images.
When generating pictures for the examples I used Comfy UI and the settings I used are:
LoRA: strength_model = 1.0 / strength_clip = 1.0
Sampler: DPM++ 2M
Scheduler: Karras
Steps: 24 - 32
cfg: 3.0Examples of prompts used:
Positive: retr0me, masterpiece, best quality, 8k, ultra-high resolution, highly detailed, anime, 1girl, young woman, solo, close up, standing, headphones, city background, blurred lights, reflection, youthful, futuristic
Negative: worst quality, low quality, bad quality, blurry, text, low resolution, pixelated, watermark, signature, black and white, monochrome, cartoon, border, framed, signs, 2 headphones, headphones around neck
Positive: retr0me, masterpiece, best quality, 8k, ultra-high resolution, highly detailed, anime, 1girl, woman, solo, medium shot, gardener, red hair, plant setting, calm expression, soft lighting, face, light skin, natural colors, serene, slightly angled, gentle gaze, indoor scene, plants
Negative: worst quality, low quality, bad quality, blurry, text, low resolution, pixelated, watermark, signature, black and white, monochrome, cartoon, border, framed
Description
Version V1 is beautiful, but does have faint hints of a modern anime. Future plans include creating a refined and distilled version trained on purely retro anime.