It's pretty obvious the detractors going on and on about "model collapse" in AI due to the internet getting flooded with AI content have never trained a model. My models are always trained in five steps:

串文

2024-09-06 13:30

讚

回覆

轉發

作者

Bill Meeks
billmeeks

粉絲

串文

61+

讚

回覆

轉發

24小時粉絲增長

無資料

互動率

(讚 + 回覆 + 轉發) / 粉絲數

NaN%

回覆 (BETA)

最先回覆的內容
發文後	用戶	內容
幾秒內	Bill Meeks billmeeks	1. Create original artwork. 2. Train/finetune a model. 3. Use that model to make a bunch of stuff I like. 4. Edit the stuff I like in Photoshop so I can add it to the dataset. 5. Retrain. Using AI images in my dataset always makes it stronger. Naturally, there's a fair amount of human curation/editing going on to fit the aesthetic I'm going for, but model collapse isn't really a thing. StableDiffusion