>>25630865Also even more bizarre when you realize it tried to generate a T-rex on the neck of an already existing T-rex.
>>25631187Oh I know, it's just painful because the lagtime for this is so great, if it even goes through half the time.
With something like NeuralBlender it was a case of waiting a few seconds at the most before trying another prompt. I was able to bend that thing to my will pretty easily because of the quick generation.
This model seems to have a better grasp of natural language, by quite a large margin, because of its GPT foundations, trying to brute force interpret how IT interprets and applies weights to prompts is... pain.
I gave up trying to get HR Giger to generate interesting things, but it shows some promise. I think the resolution is too low for Giger though, Gigers shit is filled with intricate details and contrasting texture, it just ends up noisy as fuck in the low res images.
