klibaner/fluent-emoji-style-sdxl

Fluent-Inspired 3D Emoji LoRA for SDXL

Prompt

msemj style, a friendly blue robot holding a laptop, 3d emoji icon, smooth rounded shapes, isolated on white background

Prompt

msemj style, a golden trophy with stars, 3d emoji icon, isolated on white background

Prompt

msemj style, a small online store with a striped awning, 3d emoji icon, isolated on white background

Prompt

msemj style, a red rocket launching from a cloud, 3d emoji icon, isolated on white background

Model description

Fluent-Inspired 3D Emoji LoRA is an SDXL LoRA model designed for generating colorful, rounded 3D emoji-style illustrations and icons.

The model is suitable for:

• UI and UX illustrations
• application icons
• onboarding screens
• presentation graphics
• social media content
• game items and achievements
• educational materials
• sticker and reaction concepts

This is an independent community project. It is not affiliated with, sponsored by, or endorsed by Microsoft.

Trigger word

Use the following trigger phrase in your prompt:

msemj style

For best results, place it near the beginning of the prompt.

Recommended prompt format

msemj style, [object description], 3d emoji icon, smooth rounded shapes, soft studio lighting, isolated on white background

Example prompts

Technology

msemj style, a friendly blue robot holding a laptop, 3d emoji icon, smooth rounded shapes, soft studio lighting, isolated on white background

Business

msemj style, a small online store with a striped awning, shopping bags near the entrance, 3d emoji icon, isolated on white background

Achievement

msemj style, a golden trophy surrounded by small stars, celebratory 3d emoji icon, isolated on white background

Startup

msemj style, a red rocket launching from a soft white cloud, dynamic 3d emoji icon, isolated on white background

Prompt recommendations

The model generally works best with:

• one clearly described central object;
• simple or white backgrounds;
• descriptions of materials, poses and accessories;
• square image dimensions;
• short and specific prompts.

For a consistent icon set, reuse the same prompt structure, lighting description and generation settings.

Training information

• Base model: Stable Diffusion XL 1.0
• Model type: Standard LoRA
• Trigger phrase: msemj style
• Training resolution: 512 × 512
• Network dimension: 64
• Network alpha: 32
• Training epochs: 3
• Optimizer: AdamW8bit
• Learning rate: 0.0001
• Precision: BF16
• Model format: Safetensors

Limitations

• The model is optimized primarily for single centered objects.
• Complex scenes with many objects may be less consistent.
• Generated text and letters may be inaccurate.
• White or simple backgrounds usually produce the most reliable results.
• Results depend on the LoRA weight, sampler, seed and prompt structure.
• The model may reproduce visual characteristics present in its training dataset.

Installation with Diffusers

import torch
from diffusers import DiffusionPipeline

base_model = "stabilityai/stable-diffusion-xl-base-1.0"
lora_model = "klibaner/fluent-emoji-style-sdxl"

pipe = DiffusionPipeline.from_pretrained(
    base_model,
    torch_dtype=torch.float16,
    use_safetensors=True,
)

pipe.load_lora_weights(lora_model)
pipe.to("cuda")

prompt = (
    "msemj style, a friendly robot holding a laptop, "
    "3d emoji icon, smooth rounded shapes, "
    "isolated on white background"
)

image = pipe(
    prompt=prompt,
    num_inference_steps=30,
    guidance_scale=7.0,
).images[0]

image.save("fluent_emoji.png")

License

The model is distributed under the MIT License.

Users are responsible for ensuring that generated content and downstream uses comply with applicable copyright, trademark and platform rules.

Download them in the Files & versions tab.