---
license: creativeml-openrail-m
language:
- en
tags:
- stable-diffusion
- lora
- text-to-image
- gta
- gta-online
base_model: runwayml/stable-diffusion-v1-5
pipeline_tag: text-to-image
---
![example](Без_имени.png)



# 🎮 GTA Online LoRA

A LoRA fine-tune of Stable Diffusion 1.5 trained on screenshots from **GTA Online**.  
Generates images in the visual style of GTA Online — city streets, lighting, atmosphere, and HUD elements like the minimap and money display.

---

## Examples

| Prompt | Result |
|--------|--------|
| `gtaonline, garden with flowers, minimap, hud` | Pink trees, GTA-style lighting |
| `gtaonline, city street, cars, sunny day` | Downtown Los Santos vibes |

---

## Usage

This is a LoRA weight file — you need a base **Stable Diffusion 1.5** model to use it.

### With diffusers

```python
from diffusers import StableDiffusionPipeline
from safetensors.torch import load_file
import torch

pipe = StableDiffusionPipeline.from_pretrained(
    "runwayml/stable-diffusion-v1-5",
    torch_dtype=torch.float16,
    safety_checker=None
).to("cuda")

state_dict = load_file("gta_online_lora.safetensors")
pipe.unet.load_state_dict(state_dict, strict=False)

prompt = "gtaonline, city street, minimap, hud, sunny day"
image = pipe(prompt, num_inference_steps=30, guidance_scale=7.5).images[0]
image.save("output.png")
```

### With Automatic1111 / ComfyUI

1. Download `gta_online_lora.safetensors`
2. Place it in `models/Lora/` folder
3. Use trigger word **`gtaonline`** in your prompt
4. Recommended weight: `<lora:gta_online_lora:0.8>`

---

## Trigger Word

Always include **`gtaonline`** in your prompt to activate the style.

**Good prompts:**
```
gtaonline, luxury penthouse, city view, minimap, hud
gtaonline, beach at sunset, palm trees, sports car
gtaonline, rainy night street, neon lights, puddles
gtaonline, first person view, city, hud, money display
```

---

## Model Details

| Parameter | Value |
|-----------|-------|
| Base model | Stable Diffusion 1.5 |
| LoRA rank | 16 |
| LoRA alpha | 8 |
| Training steps | 3000 |
| Dataset size | ~300 screenshots |
| Resolution | 512×512 |
| File size | 22.9 MB |
| Format | safetensors |

**Target modules:** `to_q`, `to_v`, `to_k`, `to_out.0`, `ff.net.0.proj`, `ff.net.2`

---

## Training

Trained on Google Colab T4 GPU using a custom diffusers training script.  
Dataset consists of ~300 manually captured GTA Online screenshots covering various locations, times of day, and weather conditions.

Training setup:
- Optimizer: AdamW, lr=8e-5, weight_decay=1e-4
- Scheduler: CosineAnnealingLR
- Augmentations: horizontal flip, color jitter
- Mixed precision: fp16
- Gradient clipping: 1.0

---

## License

[creativeml-openrail-m](https://huggingface.co/spaces/CompVis/stable-diffusion-license)

This model is intended for personal and research use.  
GTA Online and all related assets are property of Rockstar Games.

---

*Made with ❤️ and too many hours in Los Santos*