ShokiQ: Hi !
I got a couple of question :)
1 - As dreambooth is available for stable diffusion, is it something supermachine would get a some point ?
2 - If it ever get implemented into supermachine, how much credit you think it would cost per image. (This will maybe change my mind in buying a fourth or a fifth code)
3 - Would you consider integrate a crop image feature for img2img ?
Thank you !
Rexxar
May 15, 2024A: Thanks for the question. We're looking at dreambooth, if we do add it, it will probably be a cost to do a trained model but after that it would be 1 credit to generate an image. Unfortunately dreambooth is kinda expensive to train a model as it locks up a GPU pretty much entirely for 40-60 mins and then the filesize of these models is upwards of 2GB so if we do add it, we'll keep it reasonable but there will be a small charge to train a model.
The actual making images from dreambooth would be a standard 1 credit per image with possible a more expensive 1.5 or 2 credits if we auto upscale and face enhance what comes out.
For image 2 image, we're working on a way to auto crop to make the process more simple. Thanks!