Michael Schmidt

@joeyballentine Maybe we could dry-run install packages to test whether they are installable? I don't know if that'll work, but it might.

I'd rather not send a 100MB PNG via JSON to the frontend just to write it to the clipboard. This is something the backend has to do.

This has been solved by #920.

> Variable number of inputs is something not possible right now. It's possible using a little trick. We don't actually need to support variable number of inputs, it just has...

Currently, no. While we do store images as float32 numbers, numerous operations convert to uint8. So we could add support for saving an image as 16bit, but it wouldn't achieve...

Joey meant this part: > In chainner's menu, click _Help_ > _Open logs folder_, create a zip with `main.log` and `renderer.log`, and [attach the zip file to a comment](https://docs.github.com/en/get-started/writing-on-github/working-with-advanced-formatting/attaching-files). >...

No, it's in [the issue comment Joey linked](https://github.com/joeyballentine/chaiNNer/issues/747#issuecomment-1219389230). > I haven’t seen any of that guidance. What do you mean by "guidance"? Do you mean [our README project page](https://github.com/joeyballentine/chaiNNer#chainner)? I'm...

I have one major problem with this: whether an upscale runs out of memory depends on the tile size, not the number of tiles/split depth. This means that if Auto...

Also, the auto splitting logic seems to be duplicated for all 3 architectures. We should probably refactor that similar to how I refactored the upscaling implementations (#524). However, that can...

> But keep in mind the maximum tile size that will work changes depending on what model you're using. True. Then how about we treat the tile size not as...