SDXL Emoji LoRA

Prompt
The girl with a pearl earring flat

(CivitAI)

Model description

A small (<30 MB) rank 4 LoRA for SDXL which can help you style your images as Emojis.

It can make "Regular 3D emojis" or "Flat emojis"

Please see the LoRA weight scale <-> Trigger word chart below, before using this model

This model is very sensitive to having the correct LoRA weight scaling per Trigger word (Not to be confused with Guidance scale)

I intended for the trigger words "Emoji" and "Flat" to be mutually exclusive. So when "Emoji" is part of your prompt, do not have "Flat" in your prompt and vice versa. When you use "Flat", give a LoRA scale of 0.7-0.8. Alternatively, if you chose "Emoji" to be in your prompt, give a lower LoRA scale of 0.2-0.5.

In the Showcase you can see various prompts examples by pressing the little (i) icon.
For example:
"Gal Gadot as wonderwoman flat <lora:SDXL-Emoji-Lora-r4:0.7>"
Which has "Gal Gadot as wonderwoman" as the prompt "flat" as the trigger word and 0.7 as the LoRA scale

In the image below, the same seed was used. Also the same basic prompt. What changed is the "Trigger" word and the "LoRA weight scale"

Note that in order to get a proper result, you need to have a relatively low LoRA weight scale for the "Emoji" style (In the 0.2-0.5 range) and for the "Flat" style you would want a higher weight scale (In the 0.7-0.8 range)

Trigger words

You should use emoji, flat to trigger the image generation.

Download model

Weights for this model are available in Safetensors format.

Download them in the Files & versions tab.

Use it with the 🧨 diffusers library

from diffusers import AutoPipelineForText2Image
import torch

pipeline = AutoPipelineForText2Image.from_pretrained('stabilityai/stable-diffusion-xl-base-1.0', torch_dtype=torch.float16).to('cuda')
pipeline.load_lora_weights('Norod78/sdxl-emoji-lora', weight_name='SDXL-Emoji-Lora-r4.safetensors')
image = pipeline('Gal Gadot as wonderwoman flat ').images[0]

For more details, including weighting, merging and fusing LoRAs, check the documentation on loading LoRAs in diffusers