Content
@
0 reply
0 recast
0 reaction
horsefacts
@horsefacts.eth
Waifu technology is rapidly accelerating. Good job @eggman.eth!
20 replies
3 recasts
121 reactions
eggman 🔵
@eggman.eth
oh man, you got my favourite gen yet!!
1 reply
0 recast
6 reactions
horsefacts
@horsefacts.eth
It's impressive, I've tried similar generations before and it's very difficult!
2 replies
0 recast
4 reactions
eggman 🔵
@eggman.eth
Thanks man! Yeah, this took a while to get right - lots of custom agent pipelining going on, and two external endpoints hitting the higher end generative stuff which ain't fitting on any consumer hardware. Think it's why we're seeing some delays on gens atm, looks like we're getting backoff warnings from demand 😂
1 reply
0 recast
2 reactions
christin
@christin
please tell us more about how to do this if you are open to it!! 🙏 tried to make an AI your pfp frame with @sohey.eth and @jrf and we couldn’t stop AI from going off the rails 🥲
1 reply
0 recast
2 reactions
eggman 🔵
@eggman.eth
Yeah of course! So honestly, the most intensive part of the process was really just the actual prompt engineering side of it. I wrote up three standalone startup/system prompts which are about 7000 chars a piece. This was hugely a trial/error thing as usual, until I started getting individual outputs that I liked. From there then it pipes through to Claude-Sonnet (Haiku worked well too!) & an SDXL instance running a model fine-tuned for anime! Demand is high atm tho, causing said SDXL provider to send me backoff requests 🥲
4 replies
0 recast
5 reactions
