| --- |
| license: cc-by-4.0 |
| --- |
| |
| # MIID Model Test Results - COMPLETE |
|
|
| Face image variation generation benchmark results on 11 seed faces. |
|
|
| ## 🎉 All Tests Completed! |
|
|
| | Rank | Model | Avg Time/Image | Total Images | Status | |
| |------|-------|----------------|--------------|--------| |
| | 🥇 | **Qwen-Image** | **8.8s** | 66/66 | ✅ **FASTEST** | |
| | 🥈 | **FLUX.1-Dev** | **59s** | 66/66 | ✅ | |
| | 🥉 | SDXL | ~60s | 66/66 | ✅ | |
| | 4 | pulid_flux2 | 91.8s | 66/66 | ✅ | |
| | 5 | flux_klein | 93.2s | 66/66 | ✅ | |
| | 6 | flux_kontext | 172.8s | 66/66 | ✅ | |
| | 7 | pulid | 955s | 66/66 | ✅ | |
| |
| **Total: 7 models, 462 test images** |
| |
| ## Test Configuration |
| |
| - **Seeds**: 11 reference face images |
| - **Variations per seed**: 6 types |
| - **Hardware**: RTX 3090 (24GB VRAM) |
| - **Date**: October 27, 2026 |
| |
| ## Files |
| |
| - `*_results.tar.gz` - Individual model results |
| - `*_metadata.json` - Metadata for each model |
|
|
| ## Winner: Qwen-Image 🏆 |
|
|
| Qwen-Image is the fastest model: |
| - **8.8 seconds per image** (10x faster than FLUX.1-Dev!) |
| - **156 images per hour** |
| - **High quality output** |
|
|
| ## Recommendation |
|
|
| Use **Qwen-Image** for production deployment due to its exceptional speed. |
|
|
| --- |
|
|
| *All tests completed successfully on October 27, 2026* |
|
|
