Style or Object?

#1
by kaneda2004 - opened
Stable Diffusion concepts library org

Was the intent to train this as an Object or as a Style?

It seems to really really want to reproduce images very close to the training data -- looks like it was trained as a model rather than style

Stable Diffusion concepts library org

thanks, yes it was a test to understand the process, I probably have to choose different sources..

Stable Diffusion concepts library org

thanks, yes it was a test to understand the process, I probably have to choose different sources..

I think your dataset is great --- it just seems like you could re-train it as a style rather than as an object and get the intended result -- I'm assuming the goal would be to provide a prompt like "rough wood depthmap" and receive something usable in a program like blender -- which could be very beneficial

Nice work!

Stable Diffusion concepts library org

great! thanks for the suggestion, the goal it wag to get some sort of style
.. using as a depthmap in 3d sounds interesting, I'll look that

Stable Diffusion concepts library org

I just need to learn how to retrain as a style :)

Stable Diffusion concepts library org

I just need to learn how to retrain as a style :)

Super easy if you're training using colab just swap out the string in the var what_to_teach

#@markdown what_to_teach: what is it that you are teaching? object enables you to teach the model a new object to be used, style allows you to teach the model a new style one can use.
what_to_teach = "object" #@param ["object", "style"]

Stable Diffusion concepts library org

great thanks!!!!

neuralworm changed discussion status to closed

Sign up or log in to comment