| license: mit | |
| base_model: edurnebb/PreciseCam | |
| pipeline_tag: image-to-image | |
| tags: | |
| - diffusers | |
| - controlnet | |
| - stable-diffusion-xl | |
| - precisecam | |
| # PreciseCam middle-only | |
| This is a converted version of [`edurnebb/PreciseCam`](https://huggingface.co/edurnebb/PreciseCam) for standard Diffusers. | |
| The final 1x1 convolution of each ControlNet down-block residual projection is zeroed. The mid-block residual and the ControlNet backbone are unchanged. This bakes behavior equivalent to `middle_res_only=True` into the checkpoint, without requiring a Diffusers fork or monkey patch. | |
| ## Usage | |
| ```python | |
| import torch | |
| from diffusers import ControlNetModel | |
| controlnet = ControlNetModel.from_pretrained( | |
| "piyohogeo/PreciseCam-middle-only", | |
| torch_dtype=torch.float16, | |
| ) | |
| ``` | |
| The checkpoint uses the standard Diffusers `ControlNetModel` layout. PreciseCam's expected PF-US conditioning image is still required. | |
| Conversion details and before/after parameter statistics are recorded in `middle_only_conversion.json`. The conversion and validation tools are available at [piyohogeo/precisecam_middle_only](https://github.com/piyohogeo/precisecam_middle_only). | |
| ## License | |
| The source model is distributed under the MIT license. See the source model repository for its license and attribution details. | |