Info regarding prompts

[thakral-kartik]

Hi,

I really liked your work!!!
I have a few questions regarding the repo, it would be really helpful if you could address the below.
1.Do you have pre-computed concept inversion prompts you can share ?
2.Also, could you help me understand if concept inversion prompts are model dependent ?
3.Additionally, in each erasure sub folder a train.sh script is provided where DATA_DIR is a required argument, but as no details in readme has been provided what would be this DATA_DIR, could you please help clarify that ?
4. generate_images.py as 1st step for each erasing model is supposed to generate images from standard diffusion model or erasing model checkpoint as the command that has been provided is supposed to generate images from standard diffusion model and if so where are we using the erasing model checkpoint in the entire process ?
5. According to the readme provided, Why MODEL_NAME for ESD should be stable diffusion 1 checkpoint and for every other erasure methods like FMN, SA checkpoint for finetuned model ?

Thanks again for the great work!!

Regards,
Kartik