Stable Diffusion ?

Stability.aiが新しいモデルを公式発表するまでの情報置き場

もちろん仮名

Stable Diffusionなのかもしれないし、そうじゃないかもしれない

Emad acc/acc(@EMostaque)

#SD3 preview really is almost ready to go, 8b Turbo model just done.

We are tuning parameters and picking candidates for the preview so we can get the right feedback for the tuning.

Also team has been cleaning up inference code for partners.

Phew.

🔜

> Lykon(@Lykon4072)

> #SD3

> https://pbs.twimg.com/media/GIfxGb7WkAAfMOQ.jpg https://pbs.twimg.com/media/GIfxHu8XIAEuTE7.jpg

Emad(@EMostaque)

Testing something out (not #SD3 ), give me some prompts plz

Emad(@EMostaque)

@mysticaltech

https://pbs.twimg.com/media/GHA170HXkAAn5FJ.jpg

Emad(@EMostaque)

Thanks all, will return later with another spin. This is a fast model being tested. Also feel free to label prompts easy to hard hehe

2024/2/22

多分Stable Diffusion 3だった

Emadが情報を出すたびに上に追加していくwogikaze.icon

Emad(@EMostaque)

Waiting

https://pbs.twimg.com/media/GG1hmmKbMAALVc4.jpg https://pbs.twimg.com/media/GG1hmmOaAAAlmbR.jpg

https://pbs.twimg.com/media/GG1hmmHawAAEb_2.jpg https://pbs.twimg.com/media/GG1hmmEbIAAWrU3.jpg

Emad(@EMostaque)

Image is satisficing rapidly with zero shot already getting towards human preferences out of the box without ensembling/pipelining (which make it even better)

Some things like a few fish in both this and Sora output can be handled with existing methods with minimal overhead

https://pbs.twimg.com/media/GGi1U2zWIAAl9w1.jpg https://pbs.twimg.com/media/GGi1VkJXEAAQaH1.jpg

https://pbs.twimg.com/media/GGi1WAPXMAA2OKx.jpg https://pbs.twimg.com/media/GGi1Xh_WwAAOJrE.jpg

Emadさんや、新しいモデルをバカスカ出しても、先にX-Adapterが一般的にならないとControlNetやLoRAは全然使えないんだぞwogikaze.icon

共有できる財産ほぼないのにオープンソースLLM界隈はどうしてるんだろうかnomadoor.icon

データセットは結構共有されてるかなwogikaze.icon