vae/config.json
| 1 | { |
| 2 | "_class_name": "AutoencoderKLQwenImage", |
| 3 | "_diffusers_version": "0.39.0.dev0", |
| 4 | "_name_or_path": "Qwen/Qwen-Image", |
| 5 | "attn_scales": [], |
| 6 | "base_dim": 96, |
| 7 | "dim_mult": [ |
| 8 | 1, |
| 9 | 2, |
| 10 | 4, |
| 11 | 4 |
| 12 | ], |
| 13 | "dropout": 0.0, |
| 14 | "input_channels": 3, |
| 15 | "latents_mean": [ |
| 16 | -0.7571, |
| 17 | -0.7089, |
| 18 | -0.9113, |
| 19 | 0.1075, |
| 20 | -0.1745, |
| 21 | 0.9653, |
| 22 | -0.1517, |
| 23 | 1.5508, |
| 24 | 0.4134, |
| 25 | -0.0715, |
| 26 | 0.5517, |
| 27 | -0.3632, |
| 28 | -0.1922, |
| 29 | -0.9497, |
| 30 | 0.2503, |
| 31 | -0.2921 |
| 32 | ], |
| 33 | "latents_std": [ |
| 34 | 2.8184, |
| 35 | 1.4541, |
| 36 | 2.3275, |
| 37 | 2.6558, |
| 38 | 1.2196, |
| 39 | 1.7708, |
| 40 | 2.6052, |
| 41 | 2.0743, |
| 42 | 3.2687, |
| 43 | 2.1526, |
| 44 | 2.8652, |
| 45 | 1.5579, |
| 46 | 1.6382, |
| 47 | 1.1253, |
| 48 | 2.8251, |
| 49 | 1.916 |
| 50 | ], |
| 51 | "num_res_blocks": 2, |
| 52 | "temperal_downsample": [ |
| 53 | false, |
| 54 | true, |
| 55 | true |
| 56 | ], |
| 57 | "z_dim": 16 |
| 58 | } |
| 59 | |