//=time() ?>
1/2 Interesting thing about a custom #AI model is that it will either ignore the subject if it doesn't fit with the model, or #stablediffusion's styling takes over.
Prompt: "a 4darchitecture chessboard"
Here the style of the model is perfectly preserved, but not the subject:
dropping out the final value head from #stablediffusion's cross-attention, and replacing it with a duplicate of the penultimate value head
different, but… fine?
https://t.co/3yrsOmw23Z
got #stablediffusion's Unet compiled to CoreML, targeting all compute units (including Neural Engine).
replaced self-attention with the MultiHeadedAttention that Apple optimized for Neural Engine.
not faster yet (need cross-attention too).
https://t.co/9tdKdGIK7X
I'm in 💛with these #bluebird #spaceship's that #stablediffusion's #ArtificialIntelligence created. #mars #tesla #twitter #spacex #elonmusk #dogecoin #doge #text2art
Woah. #StableDiffusion's training data on horses is really good.
Prompt: "a stunning fantasy scene, with a beautiful magic white horse with a sparkling flowing mane in the foreground, highly detailed, with lens flare"