nupurkmr9

Hi, I have not tried training the runway in-painting model with custom diffusion. But it's one of the things on my to-do lists. I will let you know when I...

Hi, with the diffuser training script mentioned [here](https://github.com/adobe-research/custom-diffusion#training-using-diffusers-library) for multiple concepts, I get the following results.  Our results in the paper were with the CompVis codebase,...

Hi, can you mention the exact training command used to train the model .bin? Thanks.

Hi if you are using [this](https://github.com/huggingface/diffusers/tree/main/examples/custom_diffusion) to train the model, you would need to update the model load command as below: ``` ldm_stable = DiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", torch_dtype=torch.float16).to("cuda") ldm_stable.unet.load_attn_procs("path-to-save-model", weight_name="pytorch_custom_diffusion_weights.bin") ldm_stable.load_textual_inversion("path-to-save-model", weight_name=".bin")...

Hi, We don't use the validation set in our method. Thus, it can be either the same as training or entirely skipped. Hopefully, this clears your doubt. Thanks.

Hi, Since the first start of the sentence token is always fixed, I noticed a small improvement when detaching it during the training. I guess this helps in better association...

Hi, you can provide your own regularization images as well. E.g., if using `diffusers`: 1. If your regularization images have unique prompt for each image. Save images with path `class_data_dir/{'_'.join(class_prompt.split())}`...

Hi, can you provide the exact command used to train the model? This will help me in debugging the issue. Thanks.

Hi, our [diffusers code](https://github.com/adobe-research/custom-diffusion/tree/main#training-using-diffusers-library) can be used to fine-tune SD-2.0 models as well. I haven't tried the latest SDXL models, but will post an update when I do so. Thanks!!

It seems to be working now. Otherwise you can also use generated images for prior preservation. Thanks!