VintageHelper will make your Stable Diffusion 2.0/2.1 generations feel analog. It focuses on bokeh, colorgrading, depth of field, composition and adds grain and imperfection. It's been trained for 500 steps at a learning rate of 0.002, and the next 500 at a learning rate of 0.001; 15 steps gradient accumulation. All 104 training images were captioned with BLIP and corrected.
Included is the .pt for both 600 and 1000 steps, the ones I feel are best, and a .png with the 1000 steps embed into it.
Inference Providers
NEW
This model is not currently available via any of the supported Inference Providers.
The model cannot be deployed to the HF Inference API:
The model has no library tag.