Visual Text Generation on CreativeBench
28AccuracyGlyphControl
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| GlyphControl#Params=1.3B, Text Encoder=CLIP(354M), Training Dataset=LAION-Glyph-10M2023.05 | 28 | 34 | 2.4 | |
| GlyphControl#Params=1.3B, Text Encoder=CLIP(354M), Training Dataset=LAION-Glyph-1M2023.05 | 26 | 30 | 2.47 | |
| GlyphControl#Params=1.3B, Text Encoder=CLIP(354M), Training Dataset=LAION-Glyph-100K2023.05 | 19 | 24 | 2.58 | |
| DeepFloyd (IF-I-XL)#Params=6.0B, Text Encoder=T5-XXL(4.8B), Training Dataset=LAION 1.2B2023.05 | 1 | 21 | 3.09 | |
| DeepFloyd (IF-I-L)#Params=2.6B, Text Encoder=T5-XXL(4.8B), Training Dataset=LAION 1.2B2023.05 | 0.7 | 17 | 3.37 | |
| SDXL 1.0#Params=5.8B, Text Encoder=CLIP & OpenCLIP(817M), Training Dataset=Internal Dataset (>100M)2023.05 | 0.5 | 8 | 6.3 | |
| DeepFloyd (IF-I-M)#Params=2.1B, Text Encoder=T5-XXL(4.8B), Training Dataset=LAION 1.2B2023.05 | 0.1 | 11 | 3.86 | |
| Stable Diffusion v2.0#Params=865M, Text Encoder=CLIP(354M), Training Dataset=LAION 1.2B2023.05 | 0 | 2 | 5.01 |