311_code

Ok, so the SC fine tuning tab always looks at the examples folder for the toml file. I will edit the toml with images path. Trying this out again today!

Ok got it, I see it under SC Finetuning tab > folders > Dataset toml path (it looks for .json by default) "selected all filetypes" then chose the model files...

I had some luck with slightly higher quality outputs and a few issues, here are some samples:     The first one has a prompt censorship joke if...

That first one is much better likeness than I ever got. I really had to fight with text encoder unet model combinations. Maybe increasing text encoder learning rate a bit...

Ok, I spent I spent a couple more days testing. Tried a few things, no captions, changing classes, general tokens, training self with 60 photos like I would do on...

If devs here add Cascade support, can you allow for selecting different stage c and clip text encoder model checkpoints? I'm having issues with the official comfyui workflow doing this...

I would also love to see this extension working in this, It's so underrated. It's really awesome being able to control at what step a too strong lora kicks in...

I'm very confused by this project, all the advertising of video generation and I finally got it working and got the models. I assumed the pipeline.py was doing some sort...

Yeah I saw it before I downloaded, but I assumed even a little code would at least be there in the pipeline.py, there is nothing there for video at all....

> @christiandarkin not sure something will be published, @brentjohnston doesn't have any repo. Copyright law is the concern of model makers, not people sharing workflows. Wall of text, but yes...