Top Tweets for #textencoder

over 1 year ago

Here's a #CLIP #finetune w/ a #Modality #Gap of: 0.474. The TEXT *is* the IMAGE! 👊😈 Geometric Parametrization Register Tokens Gated ReLU MLP Fusion #AI - what?!🤣 Ah, just download a #TextEncoder and use it for #AIart. Or.. u want sum CODE? Sure! 👇🧵 https://t.co/KdRKGjXZrW

956

zer0int (it·its) @zer0int1

almost 2 years ago

Using #LongCLIP-L (248 #tokens input max) vs. #CLIP-L (77 tokens max) - Long-CLIP is the perfect match for T5 with regard to prompts being looooooooooooooong. Original images below. 👇🧵 #flux #flux1 #AIart #AI #textencoder #mod #finetune

zer0int1's tweet photo. Using #LongCLIP-L (248 #tokens input max) vs. #CLIP-L (77 tokens max) - Long-CLIP is the perfect match for T5 with regard to prompts being looooooooooooooong.

Original images below. 👇🧵

#flux #flux1 #AIart #AI #textencoder #mod #finetune https://t.co/tsgFyOwPVW

310

Jixian Wang

@hellojixian

about 3 years ago · Ukraine

硬核科普：#stablediffusion 的img2img生成图片的时候种子图片只是先传递给了VAE用于提取“像素”信息，每次循环采样的时候并没有参考种子图片的语义信息，所有的语义信息依然是来自 #TextEncoder 处理过的 prompt 信息。所以开源还是很重要的，不要想当然，一定要看一遍源代码再去用。代码源 #diffusers

Top Tweets for #textencoder

Last Seen Hashtags on Sotwe

Trends for you

Most Popular Users