---
license: other
base_model: black-forest-labs/FLUX.1-dev
tags:
- text-to-image
- diffusers
- lora
- flux
- emoji
- avatar
- apple-style
library_name: diffusers
---

# Mike0021/flux-dev-apple-emoji-lora

This repository contains LoRA weights for generating clean, centered Apple-style emoji avatars with `emj icon` as the trigger phrase.

The LoRA was trained for research and experimentation on Apple-style emoji assets. It is not affiliated with, endorsed by, or sponsored by Apple. The raw Apple emoji training images are not included in this repository.

## Usage

```python
import torch
from diffusers import FluxPipeline

pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16)
pipe.load_lora_weights("Mike0021/flux-dev-apple-emoji-lora", weight_name="pytorch_lora_weights.safetensors")
pipe.enable_model_cpu_offload()

image = pipe(
    "emj icon of a smiling dragon, centered on a plain white background",
    height=1024,
    width=1024,
    num_inference_steps=30,
    guidance_scale=4.0,
).images[0]
```

## Training Summary

- Base model: `black-forest-labs/FLUX.1-dev`
- Trigger phrase: `emj icon`
- Dataset: 220 curated Apple emoji image/caption pairs prepared from `iamcal/emoji-data`
- Training images are RGB 512x512 composites on a plain white background
- Captions were metadata-derived per-image captions with the shared `emj` style trigger

## Samples

The comparison sheet uses avatar-style prompts from the target Hub avatar generator prompt distribution. The right column is `fofr/sdxl-emoji` rendered with the same subject prompts where feasible.

![Avatar prompt comparison](samples/avatar-comparison-sheet.png)

## Limitations

- This model is intended for research and experimentation.
- Apple emoji source assets are Apple IP; review the relevant terms before using derivatives commercially.
- The model may reproduce Apple-like visual conventions and should not be presented as an official Apple model.