Update README.md
Browse files
README.md
CHANGED
|
@@ -1,10 +1,87 @@
|
|
| 1 |
---
|
| 2 |
title: Image Upscale Lab WebGPU
|
| 3 |
-
emoji:
|
| 4 |
-
colorFrom:
|
| 5 |
colorTo: green
|
| 6 |
sdk: static
|
| 7 |
pinned: false
|
|
|
|
| 8 |
---
|
| 9 |
|
| 10 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
title: Image Upscale Lab WebGPU
|
| 3 |
+
emoji: 👀
|
| 4 |
+
colorFrom: purple
|
| 5 |
colorTo: green
|
| 6 |
sdk: static
|
| 7 |
pinned: false
|
| 8 |
+
short_description: WebGPU image upscale
|
| 9 |
---
|
| 10 |
|
| 11 |
+
|
| 12 |
+
## **TIER 1: TOP PERFORMERS** ⭐⭐⭐
|
| 13 |
+
|
| 14 |
+
1. **4x UltraSharp V2 FP32** (51.8MB) - The gold standard for general-purpose upscaling
|
| 15 |
+
- Superior detail preservation across all content types
|
| 16 |
+
- Excellent handling of both photos and artwork
|
| 17 |
+
- Minimal artifacts, natural texture enhancement
|
| 18 |
+
|
| 19 |
+
2. **4x PurePhoto RealPLSKR** (29.9MB) - Best for photographic content
|
| 20 |
+
- Exceptional on portraits and realistic images
|
| 21 |
+
- Advanced noise handling while preserving skin textures
|
| 22 |
+
- Recommended by multiple 2026 benchmarks
|
| 23 |
+
|
| 24 |
+
3. **4x ClearReality V1 FP32** (1.7MB) - Best lightweight option
|
| 25 |
+
- Surprisingly excellent quality for its tiny size
|
| 26 |
+
- Fast processing with minimal resource usage
|
| 27 |
+
- Great balance of quality vs. performance
|
| 28 |
+
|
| 29 |
+
## **TIER 2: EXCELLENT SPECIALISTS** ⭐⭐
|
| 30 |
+
|
| 31 |
+
4. **4x Nomos8k DAT** (89.9MB) - JPEG compression specialist
|
| 32 |
+
- Specifically trained for compressed image restoration
|
| 33 |
+
- Excellent on heavily compressed photos
|
| 34 |
+
- Recommended for social media/web images
|
| 35 |
+
|
| 36 |
+
5. **4x PurePhoto SPAN** (1.7MB) - Another excellent lightweight
|
| 37 |
+
- Very similar quality to ClearReality
|
| 38 |
+
- Great for mobile/edge deployment
|
| 39 |
+
- Recommended for real-time applications
|
| 40 |
+
|
| 41 |
+
6. **4x LSDIR** (67.2MB) - Large dataset trained
|
| 42 |
+
- Good general performance across content types
|
| 43 |
+
- Solid choice for batch processing
|
| 44 |
+
|
| 45 |
+
## **TIER 3: SOLID PERFORMERS** ⭐
|
| 46 |
+
|
| 47 |
+
7. **4x UltraSharp FP32** (67MB) - Original UltraSharp
|
| 48 |
+
- Still excellent but superseded by V2
|
| 49 |
+
- Good for legacy compatibility
|
| 50 |
+
|
| 51 |
+
8. **RealESR-General WDN x4v3** (4.9MB) - Compact and efficient
|
| 52 |
+
- Recommended in multiple 2026 reviews
|
| 53 |
+
- Good balance of speed and quality
|
| 54 |
+
|
| 55 |
+
9. **4x Nomos2 HQ DAT2 FP32** (52.9MB) - High-quality alternative
|
| 56 |
+
- Similar performance to top tier but larger file size
|
| 57 |
+
|
| 58 |
+
## **TIER 4: SPECIALIZED/NICHE**
|
| 59 |
+
|
| 60 |
+
10. **4x Upscayl UltraSharp** (67.2MB) - Upscayl version
|
| 61 |
+
- Good performance but redundant with standalone UltraSharp
|
| 62 |
+
|
| 63 |
+
11. **4x Upscayl High Fidelity** (67.2MB) - Upscayl variant
|
| 64 |
+
- Quality varies by source image
|
| 65 |
+
|
| 66 |
+
12. **2x Real-ESRGAN x2plus** (67.2MB) - Only 2x scaling
|
| 67 |
+
- Limited by scale factor but good quality
|
| 68 |
+
|
| 69 |
+
## **TIER 5: AVOID/OUTDATED**
|
| 70 |
+
|
| 71 |
+
13-42. Most remaining models fall into these categories:
|
| 72 |
+
- **Outdated architectures** (original ESRGAN variants)
|
| 73 |
+
- **Overly specialized** (JPG restoration models for specific compression levels)
|
| 74 |
+
- **Redundant** (multiple similar Upscayl variants)
|
| 75 |
+
- **Limited scale factors** (1x models for restoration only)
|
| 76 |
+
|
| 77 |
+
## **KEY RECOMMENDATIONS:**
|
| 78 |
+
|
| 79 |
+
**For General Use:** Start with **4x UltraSharp V2 FP32** - it consistently ranks #1 in 2026 benchmarks
|
| 80 |
+
|
| 81 |
+
**For Mobile/Limited Resources:** Use **4x ClearReality V1 FP32** or **4x PurePhoto SPAN** - tiny files, excellent quality
|
| 82 |
+
|
| 83 |
+
**For Compressed Images:** **4x Nomos8k DAT** handles JPEG artifacts better than anything else
|
| 84 |
+
|
| 85 |
+
**For Photography:** **4x PurePhoto RealPLSKR** is specifically optimized for photos and portraits
|
| 86 |
+
|
| 87 |
+
The field has matured significantly, with newer architectures like DAT2 and RealPLKSR clearly outperforming older ESRGAN-based models. File size doesn't always correlate with quality - some of the best models are surprisingly compact.
|