README.md

---
license: apache-2.0
base_model: "black-forest-labs/FLUX.2-klein-base-4B"
tags:
  - flux2
  - flux2-diffusers
  - text-to-image
  - image-to-image
  - diffusers
  - lora
  - standard
  - macos
  - liquid-glass
  - icon-generation
  - apple-design
pipeline_tag: text-to-image
inference: true
widget:
- text: 'LIQGLASS, macOS Liquid Glass style icon, translucent frosted glass squircle shape, Safari browser app, compass symbol, blue tinted glass, specular highlight, inner glow, light refraction at edges, depth layering, soft shadow, Apple macOS design language, high detail'
  parameters:
    negative_prompt: 'ugly, flat, matte, no glass effect, blurry, low quality, old style, skeuomorphic, glossy blob, vista style'
  output:
    url: ./assets/image_0_0.png
---



# LIQGLASS — macOS 26 Liquid Glass Icon LoRA

<div align="center">

**A FLUX.2 Klein 4B LoRA fine-tuned to generate macOS Tahoe 26 Liquid Glass style app icons.**

[![Model](https://img.shields.io/badge/Base%20Model-FLUX.2%20Klein%204B-blue)](https://huggingface.co/black-forest-labs/FLUX.2-klein-base-4B)
[![License](https://img.shields.io/badge/License-Apache%202.0-green)](https://www.apache.org/licenses/LICENSE-2.0)
[![Made by](https://img.shields.io/badge/Made%20by-Andy--ML--And--AI-purple)](https://huggingface.co/Andy-ML-And-AI)

</div>

---

## What is this?

LIQGLASS is a LoRA adapter for FLUX.2 Klein 4B, trained on 542 hand-curated macOS 26 Tahoe Liquid Glass icons scraped and filtered from the community. It generates app icons that follow Apple's macOS 26 Liquid Glass design language — translucent frosted glass squircle shapes, specular highlights, layered depth, and clean minimal symbols.

The trigger word is `LIQGLASS`.

---

## Training Details

| Parameter | Value |
|-----------|-------|
| Base Model | FLUX.2 Klein 4B Base |
| Training Framework | SimpleTuner |
| Steps | 800 |
| LoRA Rank | 8 |
| LoRA Alpha | 8 |
| Learning Rate | 1e-4 |
| Optimizer | AdamW BF16 |
| Mixed Precision | BF16 |
| Dataset | 542 hand-curated macOS 26 Liquid Glass icons |
| Captioning | Qwen2.5-VL-7B (local, on Kaggle T4 x2) |
| Resolution | 512px |
| Trigger Word | `LIQGLASS` |

---

## Usage

### Recommended: Via app.py 

```python
python app.py
```

### With Diffusers + PEFT

```python
import torch
from diffusers import DiffusionPipeline

model_id = 'black-forest-labs/FLUX.2-klein-base-4B'
adapter_id = 'Andy-ML-And-AI/Andy-ML-And-AI/LIQGLASS-klein-lora'
pipeline = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.bfloat16) # loading directly in bf16
pipeline.load_lora_weights(adapter_id)

prompt = "LIQGLASS, macOS Liquid Glass style icon, translucent frosted glass squircle shape, Safari browser app, compass symbol, blue tinted glass, specular highlight, inner glow, light refraction at edges, depth layering, soft shadow, Apple macOS design language, high detail"
negative_prompt = 'ugly, flat, matte, no glass effect, blurry, low quality, old style, skeuomorphic, glossy blob, vista style'

## Optional: quantise the model to save on vram.
## Note: The model was quantised during training, and so it is recommended to do the same during inference time.
from optimum.quanto import quantize, freeze, qint8
quantize(pipeline.transformer, weights=qint8)
freeze(pipeline.transformer)
    
pipeline.to('cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu') # the pipeline is already in its target precision level
model_output = pipeline(
    prompt=prompt,
    negative_prompt=negative_prompt,
    num_inference_steps=20,
    generator=torch.Generator(device='cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu').manual_seed(42),
    width=512,
    height=512,
    guidance_scale=None,
).images[0]

model_output.save("output.png", format="PNG")

```

## Prompt Template
LIQGLASS, macOS Liquid Glass style icon, [APP NAME], [SYMBOL DESCRIPTION],
[COLOR] tinted glass, translucent frosted glass squircle shape,
[COLOR] gradient background layer with lighter frosted overlay,
specular highlight on upper left rim, inner glow, light refraction at squircle edges,
soft inter-layer shadow, 2.5D flat glass design,
Apple macOS 26 Tahoe design language, 1024x1024, high detail

### Recommended Parameters

| Parameter | Value |
|-----------|-------|
| Inference Steps | 20–28 |
| Guidance Scale | 3.5–5.0 |
| LoRA Strength | 0.8–1.0 |
| Resolution | 1024x1024 |

---

## Negative Prompt
ugly, flat matte, no glass effect, blurry, low quality, old style, skeuomorphic,
glossy blob, Vista style, Windows Aero, circular shape, two separate floating objects,
hyper realistic CGI, neon glow, particle effects, photographic background,
distorted symbol, warped, morphed, watermark, signature, border, frame

---

## What to Avoid

- ❌ Circular icons — must be Apple squircle
- ❌ Vista/Aero style glossy blobs
- ❌ Hyper-realistic 3D CGI rendering
- ❌ Skeuomorphic textures
- ❌ Symbol distortion or warping
- ❌ Two physically separated floating objects
- ❌ Neon glow or particle effects

---

## What Works Well

- ✅ Single clean symbol centered in the squircle
- ✅ One dominant color family per icon
- ✅ Frosted glass layering with subtle depth
- ✅ Apple-style gradients (blue, teal, green, purple, amber)
- ✅ Metallic or white symbols with subtle bevel
- ✅ Soft specular highlights and inner glow

---

## License

This LoRA is released under **Apache 2.0** — same as the base FLUX.2 Klein 4B model. Free for personal and commercial use.

---

## Credits

- Base model: [FLUX.2 Klein 4B](https://huggingface.co/black-forest-labs/FLUX.2-klein-base-4B) by Black Forest Labs
- Dataset: 542 macOS 26 Liquid Glass icons curated from [macosicons.com](https://macosicons.com)
- Trained by: [Andy-ML-And-AI](https://huggingface.co/Andy-ML-And-AI)