As of today Adobe has released a beta version of Photoshop that has Firefly AI integrated into it. Access this on your Cloud app by going to Beta apps and installing it. It's still in beta so still no commercial use.
The new thing it now has (as of today) is inpainting and outpainting. That puts it in competition to DALL-E2. The edge it has over DALL-E2 is it's integrated into Photoshop. No downloading here and uploading there and bopping between apps. It's a seamless workflow.
It's still meh at generation but it's in beta. Adobe will work that out. I asked it to generate a table set for a celebration. I got this:
The forks are awful so I used inpainting hoping for better forks. Not any better.
I don't think DALL-E2 is any better.
I'm actually confused about whether I made these with DALL-E or DALL-E2 - It's whatever you get when you go to: https://labs.openai.com/
I tried outpainting. Bad. I also tried removing the plate so that this would work as a mockup for a favor box. No dice. It kept adding new and different plates with different napkins. Is DALL-E2 any better? No. It did the exact same thing. "Empty plain tablecloth" was not an option.
It's not ready to use commercially here (want to use it for mockups) but that it isn't yet allowed. Because it's beta they ask for feedback so I requested the ability to have negative prompts such as "no plates".
By 2024, Firefly will be breathtaking, I have no doubt. The forks suck today. In several months they will be fine.
I don't think DALL-E2 is any better.
I'm actually confused about whether I made these with DALL-E or DALL-E2 - It's whatever you get when you go to: https://labs.openai.com/
I use NightCafe and was working last night on trying to get something I could use for door mat mock ups. I got a lot like this one for example
that I thought were really pretty and surprisingly photo-realistic (enough for a cover that's not going to be analyzed in excruciating detail) but I can't get the angle / composition right to be closer in and have more porch floor and less door/wall.
I think there is so much potential in this AI generation but it's not the easy magic some think it is. You (and I mean that in a general sense not you specifically 😉 ) can't just enter a few words and presto have exactly what you want. There's a learning curve for the promptist, it takes time and practice and trial & error, with the randomness of AI thrown in. I liken it to tie-dying. You can follow best practices according to what you have both read and learned first-hand, but you still never know just what the outcome will be until you unwrap and rinse it .
Adobe's Firefly sounds interesting and convenient but I already have issues with Photoshop (I do pay the monthly subscription for the legit licensed version) so for now my AI experimenting will have to remain web-based.
That's true. I started off with prompts that specifically included a door mat in the scene so there'd be the space there to overlay my own. Then I evolved a couple of them, using inpainting to get rid of the door mat and got things like this:
 |  |
Not bad but still not good for a cover as the mat would only be like 1/3rd of the image instead of the focus. (The one on the left I re-did countless times with inpainting and it kept insisting on making the wall red and the larger pot all warped.)
That's actually not bad, but I totally get what you're saying in terms of the angles and perspective. AI doesn't seem to understand prepositions! At one point I had a bright idea that I'd use the outpainting feature to create some mockups - uploading the cutout mockup from Zazzle. I could only laugh at the results!
