Verified: 2 years ago
SafeTensor
Type | |
Stats | 1,366 57,651 295 |
Reviews | (129) |
Published | Sep 28, 2023 |
Base Model | |
Training | Epochs: 10 |
Hash | AutoV2 4563CAE543 |
NextPhoto is my new favorite model for photorealism. It can generate high-quality photos with very simple prompt. However, personally I find the default output of the model is a bit "bland", not eye-pleasing enough. So I create this LoRA to give images a more cinematic look, in terms of color, lighting, and depth of field.
You can see the effect of the LoRA in the sample images and the comparison grids.
I think the effect is relatively subtle in v1; you may need to raise the weight more than 1 in some instances to see a more significant effect. I might train another version for more dramatic results.
Trained on SD V1.5, and tested on NextPhoto V3.0. It might work on other SD1.5-based models, but there is no guarantee.