Supporting for multi-concept customization
zzhang2816 opened this issue · comments
Hey, it's really a great job. As you show in the paper, it has the capacibity to composite multiple learned objects("Figure 6: Examples of composing inverted concepts of cat*, chair*, dog*, and barn* to each other."). But I am not sure does the code released supporting multi-concept customization? Thank you!
Thanks for your support!
The code supports multi-concept customization. To train multiple concepts in one model, you can run the following scripts: '/scripts/compositional_custom_diffusion_multi.sh' or '/scripts/compositional_dreambooth_multi.sh'. For now, we haven't support optimization-based approach for multi-concept composition in Custom Diffusion.
I see. Thanks for your prompted reply!