---
tags:
- text-to-image
- lora
- diffusers
- template:diffusion-lora
widget:
- text: sza holding microphone emoji. curly long hair. dark skin tone
  output:
    url: images/example_nk0p3d0uy.png
- text: >-
    billie eilish holding microphone emoji. blonde hair. 
    seafoam blue-green eyes. happier than ever era. light skin tone.
  output:
    url: images/062.png
- text: >-
    man sticking out tongue and wink emoji. blue wavy hairstyle. one eye
    closed.  facing viewer. medium  skin tone
  output:
    url: images/043.png
- text: >-
    emoji of green-clad Hylian warrior holding Master Sword. 
    enlarged head in cartoon style. detailed texture. 3D lighting. no cast shadows.
  output:
    url: images/083.png
- text: squirrel on surfboard emoji. cute. facing viewer.
  output:
    url: images/047.png
- text: sza holding microphone emoji. curly long hair.
  output:
    url: images/051.png
- text: ed sheeran holding microphone light skin tone emoji.
  output:
    url: images/056.png
- text: unicorn on bicycle emoji.
  output:
    url: images/057.png

base_model: black-forest-labs/FLUX.1-dev
license: other
license_name: flux-1-dev-non-commercial-license
license_link: https://huggingface.co/black-forest-labs/FLUX.1-dev/resolve/main/LICENSE.md

---
# Generate Emoji

<Gallery />

## Model description 

This model can generate emoji in a familiar style. Trained on emoji with diverse skin tones.

## Download model

Weights for this model are available in Safetensors format.

[Download](/caspersimon/Diverse-Emoji/tree/main) them in the Files & versions tab.

## Usage


The model was trained on small images (160x160) and has mostly been tested to generate images of that size. Generating larger images is possible and limited testing shows that this can yield improved results (see images below).

<div style="display: flex; gap: 20px; align-items: flex-start; margin: 20px 0;">
    <div style="text-align: center; width: 200px;">
        <img src="images/example_nk0p3d0uy.png" alt="SZA emoji" width="200">
        <p style="margin: 5px 0; font-size: 0.8em">1a: Generated 1024x1024</p>
    </div>
    <div style="text-align: center; width: 200px;">
        <img src="images/051.png" alt="SZA emoji" width="200">
        <p style="margin: 5px 0; font-size: 0.8em">1b: Generated 160x160, upscaled to 800x800</p>
    </div>
    <div style="text-align: center; width: 200px;">
        <img src="images/062.png" alt="Billie Eilish emoji" width="200">
        <p style="margin: 5px 0; font-size: 0.8em">2a: Generated 320x320, upscaled to 1600x1600</p>
    </div>
    <div style="text-align: center; width: 200px;">
        <img src="images/061.png" alt="Billie Eilish emoji" width="200">
        <p style="margin: 5px 0; font-size: 0.8em">2b: Generated 160x160 (using same prompt and seed as 2a), upscaled to 800x800</p>
    </div>
</div>

However, increasing output size can also cause unexpected outcomes.
<div style="display: flex; gap: 20px; align-items: flex-start; margin: 20px 0;">
    <div style="text-align: center; width: 200px;">
        <img src="images/063.png" alt="Billie Eilish emoji" width="200">
        <p style="margin: 5px 0; font-size: 0.8em">2b: Generated 1024x1024 (using same prompt and seed as 2), upscaled to 5120x5120</p>
    </div>
</div>

See [this github repo](https://github.com/EvanZhouDev/open-genmoji/) for exactly how to use this model. Note that this implementation currently uses a different model by default, so you should change some of the code to use the diverse emoji model.

### *Skin tone variations*

The model was trained on a set of emoji and their descriptions. For emoji with available skin tone variations, skin tone was explicitly present in the descriptions of the emoji, so generating an emoji with a specific skin tone is possible by adding the specific skin tone to the prompt.[^1] See the table below for example descriptions of an emoji with skin tone variations.

| Emoji | Description\*|
|-------|----------------------------------------------------|
| 👩‍🦱    | woman with curly hair *default skin tone* emoji      |
| 👩🏻‍🦱   | woman with curly hair *light skin tone* emoji        |
| 👩🏼‍🦱   | woman with curly hair *medium light skin tone* emoji |
| 👩🏽‍🦱   | woman with curly hair *medium skin tone* emoji       |
| 👩🏾‍🦱   | woman with curly hair *medium dark skin tone* emoji  |
| 👩🏿‍🦱   | woman with curly hair *dark skin tone* emoji         |

\* Added emphasis is not present in training set

[^1]: I have found that "default skin tone yellow emoji" works more reliably than just "default skin tone"