Base

Winner // 1 ETH: @keccers.eth with Trendcoins

It uses ai to coin trending google searches every 2 hours on Zora, that are pushed to a miniapp where you can get the full scoop.

https://trendcoins.replit.app/

i did it! 🎉 

thank you @base for the opportunity to enter my project, i'm proud 🥰 

the repl for this whole project — everything! the mini app! the auto coin script! — is public 

https://replit.com/@katherinechampa/Trendcoins?v=1#README.md

my sincere hope is that people play with it  — an updated readme in the works to help make this as easy as possible for you

Congrats and thanks for open-sourcing your work!

I have been working on sth similar, but instead of trending google searches my bot checks for large prediction market price spikes or finalised markets. I am happy with the description for the coins but the images could be more fun, will check how you generate them

The Farcaster formerly known as Kchamp. Books, beats, bodybuilding, bright ideas and beautiful things.

katherinechampagne.com

Congrats and thanks for open-sourcing your work!

I have been working on sth similar, but instead of trending google searches my bot checks for large prediction market price spikes or finalised markets. I am happy with the description for the coins but the images could be more fun, will check how you generate them   
https://zora.co/coin/base:0xe6b2a504bb04d28e973431d1211ea86e9102efe4?referrer=0x159a514cba2941d3fb5f3bc06237ea3531200748

Based builder of TrueCast | BasedGiving | Xonin 🔵

Rad!!!! In mine Gemini makes an image prompt along with making the token details and sends it to OpenAI’s gpt-image-1!

Gemini at this usage level is totally free bc they give you so many tokens. I am even using 2.0 Flash which is a powerful model! 

So I offloaded as much as possible to Gemini to keep it cheap and pass as little as possible to open ai 

Gemini afaik also the only one to reliably return JSON which you need for token metadata 

The fx to generate prompts is in lib/gemini.js

I would prob never write these prompts directly ‘cause the bot is totally automated! 

There’s definitely room for improvement here. I had to add that last line cause a LOT of jobs were failing due to OAI safety rejection

Interesting, did you find it to give better results when you let another model write the image prompt rather than prompting it directly? So far only tried dalle but the images are not that great, might have to change to gpt-image-1 too