Comment by xelia

I wonder if this can be optimized by letting GPT provide multiple instructions per screenshot instead of just one.

For example in the twitter screenshot, it could use just the one image.