When I’m editing a section of the image, I don’t want to retype the entire prompt. I just want to type what I want in that section, fitted to the existing style of the generated image.

Curated works / posts of user best practices on prompt design, optimisation etc. Right now I’m relying on random people I follow on Twitter and blog posts.

I wish I could get direct access to all 50000 potential logit logprobs as I generate 1 token at a time, or upload a function to work with them.

the logical way to use DALL-E right now is to write a query and test it on a free service like crayion ai to have a glimpse of what the image would look like to not waste credits... DALL-E could have a “draft” section generating low res images before the full final drawing

Ability to fine tune codex models would be nice. Fine tuning for Dall-E would open quite a few doors for us.

Placing limits on DALLE2 during a new user’s first day adds scarcity and kills the previously magical onboarding experience. Most people I knew exhausted their first daily limit in a flurry of wonder and amazement. New users try two pictures and save their credits for later.

Cheaper embedding. Other than that it’s my absolute go to tool for most things. Epic product.

Probably hard to pull off but (even) faster turnaround time + more results would be amazing

Too many restrictions.

As a (non-paying, so I really can’t complain) user of the Codex API: rate limits of 20/minute are way too low for many tasks. Totally understand though, having now seen firsthand how expensive inference is.