Do you hate the repetition between models? The scammy merges with no listing of sources? And how loras trained on base SDXL don't often work too well with newer merges?
Then consider running a collaborative community finetune to base SD3! This could be done using something like GitHub, where a team approves proposed merges into a common base. That way the approved common finetune could be collaborated on by a team. More cooperation means a better community base to build on, as well as faster improvements overall!
This would also solve the "loras not working" issue as existing loras would continually be tested against that common finetune and it would quickly be identified where a deviation needs to be addressed.
Please consider this, and feel free to suggest other solutions to encourage the community to work together to achieve these goals.
Thank you!