Rolls-Royce / README.md
fantaxy's picture
Update README.md
5f357ae verified
---
title: Rolls-Royce FLUX LoRA
emoji: 🐠
colorFrom: yellow
colorTo: green
sdk: gradio
sdk_version: 5.35.0
app_file: app.py
pinned: false
---
I'll create comprehensive documentation for this Flux LoRA Rolls-Royce image generation code in both English and Korean.
## English Documentation
### Flux LoRA Rolls-Royce Image Generator
This application is a specialized image generation tool that uses the FLUX.1-dev diffusion model with a custom LoRA (Low-Rank Adaptation) fine-tuned specifically for generating high-quality Rolls-Royce automobile images.
#### Key Features
1. **Advanced AI Model Integration**
- Utilizes the state-of-the-art FLUX.1-dev diffusion pipeline from Black Forest Labs
- Incorporates a custom LoRA adapter (`seawolf2357/flux-lora-car-rolls-royce`) specifically trained on Rolls-Royce vehicles
- Supports both CUDA GPU acceleration and CPU fallback for broader compatibility
2. **Persistent Image Storage**
- Automatically saves all generated images with unique timestamps and UUIDs
- Maintains a metadata file tracking prompts and generation times
- Provides a gallery view for browsing previously generated images
3. **User-Friendly Interface**
- Built with Gradio for an intuitive web-based interface
- Features two main tabs: Generation and Gallery
- Includes pre-written example prompts showcasing various Rolls-Royce models in luxurious settings
4. **Customizable Generation Parameters**
- **Seed Control**: Option for randomization or manual seed setting for reproducibility
- **Image Dimensions**: Adjustable width and height (up to 1024x1024 pixels)
- **Guidance Scale**: Controls how closely the image follows the prompt (0.0-10.0)
- **Inference Steps**: Number of denoising steps (1-50)
- **LoRA Scale**: Strength of the Rolls-Royce-specific adaptation (0.0-1.0)
#### Technical Implementation
The application leverages several key technologies:
- **PyTorch** for deep learning operations
- **Diffusers** library for the diffusion model pipeline
- **Gradio** for the web interface
- **Spaces GPU** decorator for optimized GPU usage (120-second duration limit)
The generation process:
1. Loads the base FLUX.1-dev model with bfloat16 precision for memory efficiency
2. Applies the Rolls-Royce LoRA weights for specialized car generation
3. Processes user prompts with specified parameters
4. Saves generated images automatically with metadata
5. Updates the gallery view with new creations
#### Example Use Cases
The included examples demonstrate various scenarios:
- Classic Rolls-Royce models in architectural settings
- Modern vehicles in urban environments
- Luxury SUVs in natural landscapes
- Vintage cars in historical contexts
- High-performance models in exclusive locations
Each prompt includes the `[trigger]` keyword to activate the LoRA adaptation effectively.
---
## ν•œκΈ€ μ„€λͺ…μ„œ
### Flux LoRA 둀슀둜이슀 이미지 생성기
이 μ• ν”Œλ¦¬μΌ€μ΄μ…˜μ€ FLUX.1-dev ν™•μ‚° λͺ¨λΈκ³Ό 둀슀둜이슀 μžλ™μ°¨ 이미지 생성에 νŠΉν™”λœ μ»€μŠ€ν…€ LoRA(Low-Rank Adaptation)λ₯Ό μ‚¬μš©ν•˜λŠ” μ „λ¬Έ 이미지 생성 λ„κ΅¬μž…λ‹ˆλ‹€.
#### μ£Όμš” κΈ°λŠ₯
1. **κ³ κΈ‰ AI λͺ¨λΈ 톡합**
- Black Forest Labs의 μ΅œμ²¨λ‹¨ FLUX.1-dev ν™•μ‚° νŒŒμ΄ν”„λΌμΈ ν™œμš©
- 둀슀둜이슀 μ°¨λŸ‰ μ „μš©μœΌλ‘œ ν•™μŠ΅λœ μ»€μŠ€ν…€ LoRA μ–΄λŒ‘ν„°(`seawolf2357/flux-lora-car-rolls-royce`) 적용
- CUDA GPU 가속 및 CPU 폴백 μ§€μ›μœΌλ‘œ 폭넓은 ν˜Έν™˜μ„± 제곡
2. **영ꡬ 이미지 μ €μž₯**
- μƒμ„±λœ λͺ¨λ“  이미지λ₯Ό κ³ μœ ν•œ νƒ€μž„μŠ€νƒ¬ν”„μ™€ UUID둜 μžλ™ μ €μž₯
- ν”„λ‘¬ν”„νŠΈμ™€ 생성 μ‹œκ°„μ„ μΆ”μ ν•˜λŠ” 메타데이터 파일 μœ μ§€
- 이전에 μƒμ„±λœ 이미지λ₯Ό 탐색할 수 μžˆλŠ” 가러리 λ·° 제곡
3. **μ‚¬μš©μž μΉœν™”μ  μΈν„°νŽ˜μ΄μŠ€**
- Gradioλ₯Ό μ‚¬μš©ν•œ 직관적인 μ›Ή 기반 μΈν„°νŽ˜μ΄μŠ€ ꡬ좕
- 생성(Generation)κ³Ό 가러리(Gallery) 두 개의 μ£Όμš” νƒ­ 제곡
- λ‹€μ–‘ν•œ 둀슀둜이슀 λͺ¨λΈμ„ κ³ κΈ‰μŠ€λŸ¬μš΄ ν™˜κ²½μ—μ„œ λ³΄μ—¬μ£ΌλŠ” 예제 ν”„λ‘¬ν”„νŠΈ 포함
4. **λ§žμΆ€ν˜• 생성 λ§€κ°œλ³€μˆ˜**
- **μ‹œλ“œ μ œμ–΄**: μž¬ν˜„μ„±μ„ μœ„ν•œ λ¬΄μž‘μœ„ν™” λ˜λŠ” μˆ˜λ™ μ‹œλ“œ μ„€μ • μ˜΅μ…˜
- **이미지 크기**: μ‘°μ • κ°€λŠ₯ν•œ λ„ˆλΉ„μ™€ 높이 (μ΅œλŒ€ 1024x1024 ν”½μ…€)
- **κ°€μ΄λ˜μŠ€ μŠ€μΌ€μΌ**: ν”„λ‘¬ν”„νŠΈ μ€€μˆ˜ 정도 μ œμ–΄ (0.0-10.0)
- **μΆ”λ‘  단계**: λ…Έμ΄μ¦ˆ 제거 단계 수 (1-50)
- **LoRA μŠ€μΌ€μΌ**: 둀슀둜이슀 νŠΉν™” 적응 강도 (0.0-1.0)
#### 기술적 κ΅¬ν˜„
μ• ν”Œλ¦¬μΌ€μ΄μ…˜μ€ λ‹€μŒκ³Ό 같은 핡심 κΈ°μˆ μ„ ν™œμš©ν•©λ‹ˆλ‹€:
- λ”₯λŸ¬λ‹ 연산을 μœ„ν•œ **PyTorch**
- ν™•μ‚° λͺ¨λΈ νŒŒμ΄ν”„λΌμΈμ„ μœ„ν•œ **Diffusers** 라이브러리
- μ›Ή μΈν„°νŽ˜μ΄μŠ€λ₯Ό μœ„ν•œ **Gradio**
- μ΅œμ ν™”λœ GPU μ‚¬μš©μ„ μœ„ν•œ **Spaces GPU** λ°μ½”λ ˆμ΄ν„° (120초 μ‹œκ°„ μ œν•œ)
생성 ν”„λ‘œμ„ΈμŠ€:
1. λ©”λͺ¨λ¦¬ νš¨μœ¨μ„±μ„ μœ„ν•΄ bfloat16 μ •λ°€λ„λ‘œ κΈ°λ³Έ FLUX.1-dev λͺ¨λΈ λ‘œλ“œ
2. μ „λ¬Έ μžλ™μ°¨ 생성을 μœ„ν•œ 둀슀둜이슀 LoRA κ°€μ€‘μΉ˜ 적용
3. μ§€μ •λœ λ§€κ°œλ³€μˆ˜λ‘œ μ‚¬μš©μž ν”„λ‘¬ν”„νŠΈ 처리
4. μƒμ„±λœ 이미지λ₯Ό 메타데이터와 ν•¨κ»˜ μžλ™ μ €μž₯
5. μƒˆλ‘œμš΄ μƒμ„±λ¬Όλ‘œ 가러리 λ·° μ—…λ°μ΄νŠΈ
#### μ‚¬μš© μ˜ˆμ‹œ
ν¬ν•¨λœ μ˜ˆμ œλ“€μ€ λ‹€μ–‘ν•œ μ‹œλ‚˜λ¦¬μ˜€λ₯Ό λ³΄μ—¬μ€λ‹ˆλ‹€:
- 건좕적 배경의 ν΄λž˜μ‹ 둀슀둜이슀 λͺ¨λΈ
- λ„μ‹œ ν™˜κ²½μ˜ ν˜„λŒ€μ  μ°¨λŸ‰
- μžμ—° 풍경 μ†μ˜ λŸ­μ…”λ¦¬ SUV
- 역사적 λ§₯락의 λΉˆν‹°μ§€ μžλ™μ°¨
- 독점적인 μž₯μ†Œμ˜ κ³ μ„±λŠ₯ λͺ¨λΈ
각 ν”„λ‘¬ν”„νŠΈμ—λŠ” LoRA 적응을 효과적으둜 ν™œμ„±ν™”ν•˜κΈ° μœ„ν•œ `[trigger]` ν‚€μ›Œλ“œκ°€ ν¬ν•¨λ˜μ–΄ μžˆμŠ΅λ‹ˆλ‹€.