darkman111a

Bro. These are raw models. Finetuning is needed. For raw output. I considered that pretty amazing. I do wonder why they dind't use Flan-T5 instead, these wouldn't be issues, even...

Yes, could expertiment with finetuning once you swap it out. I'm 1000% sure FLAN-T5 would result in higher fidelity output, better composition, way better spatial awareness. I think "tango model"...

I wonder if it has to do with image dimensions? It seems that the support_noise tensor has a different shape than expected.

Just repurpose the imagen trainer script. Until they release their paper, just use random search and figure the best hyperparameters to you. Currently trying to figure out myself, I'm beyond...

I'm interested in this as well. Appreciate you OG. Stay blessed!

@lucidrains you're really the chosen one. Looks good to me. Appreciate all your hard work.