Update: Klein 9B version added.
Dat gap. Z-Image Turbo needed some help with this one, so here it is. Use "thighgap" as the trigger word. Best around 0.9 strength. Works well enough most of the time.
For a discussion of realistic AI image/music/video generation, join us at our new Discord server: https://discord.gg/B6BSCbdAJX
Description
Same dataset as the ZiT version but for Klein 9B. More pronounced effect now since K9B trains better than ZiT.
FAQ
Comments (4)
Qwen 2512 please!
Turns out with Klein and Z-Image that removing half the model does actually cause issues.
Removing half the model What do you mean? 0.5 weight?
@RaymondLuxuryYacht I think he means that Qwen is 18-20 Billion parameters and that Klein and ZIT are 8-9B.
HOWEVER.... With the news today about Qwen2. Don't waste any time one Qwen1/2512.
Finally, I can turn a girl into an AT-ST from Star Wars.