It’s incredible this is out already. It’s about 40% of the way to DALL-E but you can mostly tell the prompt from the image.

Is there a Dall-E Ultra Max in the works?

There's this

https://replicate.com/jack000/glid-3-xl

Check out MidJourney, which is in closed beta but expanding rapidly. I think it uses a different approach (different models), but also stunning results.

He's currently training Dall-E mega: https://twitter.com/borisdayma/status/1532071560558759937

Careful, the moral police will try and take you down. They decided we are not ready for this yet.

"Too much traffic"

Just try again after a few seconds. HN always does this to smaller servers.

The logo of the project is not good, and i suggested to them a better one in the forum. They will use it in the future. The avocado armchair which i created, i did not stated simply "A logo of an avocado armchair" like i am talking to a human being. I expressed it as declarative as possible, like a computer program, and the machine understood it in a much more correct way, what is the command to execute.

As i have stated in the past, i consider natural language inputs to the machine, to be a very poor substitute of programming inputs, and i think that experiment has failed already.

https://huggingface.co/spaces/dalle-mini/dalle-mini/discussi...

Can you elaborate what you mean by "programming inputs" because the only thing you did in your example is be more descriptive in your prompt...

The prompt is the programming, I'd guess.

Can you share exactly how you phrased it? Might help with making your point.

Yours do indeed look substantially better than the authors.