Adobe Firefly

[KeegansCreation]

Adobe has made their own AI called Firefly. They very pointedly say they trained it only on public domain, Adobe Stock and images they licensed, so that's good. Today I got access to the beta. Any Adobe subscriber can request access (although they are slow rolling it, I requested access several days ago).

My verdict: as good as DALLE-2 (but there will be a new DALL E soon) and not quite as good as Midjourney but then it's in beta. Their plan is to integrate it into Photoshop and Illustrator but currently (in beta) it's a standalone website.     https://www.adobe.com/sensei/generative-ai/firefly.html 

Below is my test for cover photo background. I can't quite plunk a dinner party invitation down between the fork and spoon until the fork tines get a little tweaking. But it's very close. This has real potential for backgrounds on cover photos. As you can see in the lower left corner, I can't use it yet. Commercial use is not allowed while in beta but they do encourage non-commercial sharing to spread the word so there ya go. I encourage any Adobe subscribers here to request access to the beta via the above link.

[Alissanos · Accepted solution]

Ive already requested access but am still waiting, and I cant wait to check it out, apparently they are working on text to editable vectors, which would be very interesting!

[KeegansCreation]

Oh yes. Text to editable vectors would be amazing. They say they plan to integrate it into Illustrator and it wouldn't make any sense to do that if it couldn't do vector. I am also intrigued about inpainting, which they are working on. Hopefully they let you in soon.

• Text to image is pretty good, easily as good as DALL-E. Along with your initial prompt there are also side panels where you can add in coloring (pastel colors, muted colors etc.), lighting (low lighting, backlighting etc.) and composition (camera angle). There are also panels where you can choose an art movement, technique or material. But you are not limited to these panels. They are merely add-ons for whatever your prompt is and to get newbies used to prompting.
• There is no upscaling currently but square downloads as 1024 pixels square and rectangular is 1408 x 1024.
• no commercial use while in beta. They add the watermark in the corner as a reminder.
• Your images are not saved in this beta. You'd have to download them.
• Text Effects are... meh. They are very intricate and full of flourishes which is great if your desired effect is flowers or lace but looks awful if you are going for metal or anything smooth. Their attempt at a bread text effect is horrifying.
• They encourage but do not require you to rate the images by thumbs up or down and then there is a place for you to fill check off what is good (such as being accurate to the prompt) or bad (I asked for two bluejays and got a bluejay with two heads).
• They have a discord where you can share images and also tell them point blank what you think is good or bad. 

All in all it's encouraging. This is a beta so it will get better. And as you can see in my posted example, it's almost ready to generate cover photo backgrounds. The one I posted would be fine to use with a fix of the fork tines and addressing whatever weird reflection is happening in the spoon. If they get inpainting going then boom, done.

They are also working on what they call personalization which is actually just training it on your own images. It looks like their goal is to combine the ease-of-use you get with DALL-E and Midjourney with the features you get with Stable Diffusion. 

[Alissanos]

It would be wouldnt it? It says somewhere on the firefly page that they will be editable

I love illustrator and I use midjourney a lot for inspiration, even for color palettes to use. I even use it to create mood boards which is fun but they are rarely useable for anything bar the color scheme. 

Have you tried midjourney v5 yet? 

Bing ai also has an image generator now too but tos says not for commercisl use and anything you create they own or can use or some such, same for if you upload an image so its not very useful. Also if you want to waste some time ask bing ai to google your zazzle products or store/website. And its pretty good for asking about trends in whatever niche youre in..

Personalization would be awesome 

I see they are also working on creating images out of sketches, which would be great, midjourney can kind of do that but not with any reliable kind of accuracy. At the moment I sketch and then create vectors of my sketches so that would be great. 

Ive almost given up on cover photos, I dont really like them and I dont like making them but maybe if I get access ill give it a try

[KeegansCreation]

I have used Midjourney v5. It is really quite different from v4. It defaults to a photographic style rather than an artsy style and you have to be super specific in your prompts. It can do art just fine but you have to tell it exactly what kind. Since v5 is good at photography (better than v4) I have been using it for cover photos. I wouldn't chance using Firefly images since they say no commercial use and although you don't sell cover photos, it's still risky. Better to use v5 for cover photos. Some examples of cover photo backgrounds I did with with v5:

Penicillium Humor
by KeeganCreations

25th Anniversary To Husband
by KeeganCreations

[PenguinPower]

I just got in on the Adobe Beta. I gave it a test by doing a few of the same prompts used in the old thread on Dall-e. Adobe gave me some very acceptable painting of Santa Claus - all the eyes were in the correct places and none of them looked like they were about to massacre children in a drunken rage... It still doesn't do great on photorealistic images though.., which is what I was hoping for, to make cover photo images easier to obtain.

People can still be quite horrific.. Asking for a photo of a family posed in front of a Christmas tree gave me some pretty gruesome images and the ones of a kitchen - well, let's just say I'd fire my contractor for not being able to make the countertops level, melting the cabinet doors and putting a faucet on the stove.. 
On a brighter note, it did give me a very nice looking jar of hot dill pickles.. 

[KeegansCreation]

Glad to hear you're doing the beta test now. When it gives me wonky images I do fill out the why-is-this-bad dropdown (appears when you give thumbs down) and hopefully knowing what exactly is wrong will steer them right for the actual release. They come out with new Photoshop versions each Fall so I'm hoping that they can fix the issues over Spring and Summer and bring it out of beta. They do seem reasonably far along; farther along than DALL-E was when Cat started that thread last Fall with nightmare images of people. And of course there were those awful Santas. Progress on AI seems practically logarithmic now so I have high hopes for coming out of beta in the Fall.

What did you think of the text effects? I tried to get it to do smooth metallic finishes but it seemed locked in to intricate detail which just looked creepy and weird.

That it gave you a nice looking jar of hot dill pickles is very positive. I have high hopes of eventually using it to make backgrounds for cover photos. If you had made a Zazzle label for hot dill pickles you could put it on that jar if commercial use was allowed- which it will be when beta is over.  Since cover photos are your intention as well, this could be very good since it seems to do that a lot better than people.

[PenguinPower]

I haven't messed around with it much yet... I just got the invite this morning 😛, so I haven't played with much more than the basic text to image yet.

I do in fact have some hot dill pickle labels. I think I did eventually find enough photos to make covers for all of them (edit... I found enough for regular pickles, but I still need hot ones 🤗), but interestingly, it was one of the harder things to track down in Adobe stock.. I do hope that eventually we'll be able to use them for cover images. I can't imagine that they'd be putting in all this work if they don't intend to go commercial eventually.

[KeegansCreation]

I am confident they will allow commercial use once out of beta. This wording implies that the restriction is just during beta: "During the beta, generative imagery and text from Firefly should not be used for commercial purposes. "

I am particularly excited about things in their "in exploration" section. Inpainting (which DALL-E2 already has) would be a game changer if they can do it better than DALL-E2. I've used DALL-E2's inpainting but it's meh. And text to vector. 😍

[PenguinPower]

Oh yeah - text to vector would be pretty awesome... One of the things I found DALL-E actually useful for was generating woodcut-esque images, then doing a live trace on them in Illustrator and tweaking the heck out of them.. Going right to vector would be nice. 
As far as Inpainting goes... I would bet that they'll have that down pretty quickly - I mean Photoshop already has content aware fill.... which if not perfect does relatively well, especially if what you are filling is either relatively plain or relatively abstract. 

[PenguinPower]

Here's how it did when asked for "a jar with a blue lid containing home canned mini dill pickles with a few red pepper slices on a rustic wooden table, blurred background viewed from the front" ( I had to screen shot it... for whatever reason the download wasn't working)

Strange thing -I had to ask for "mini pickles" to not get some grotesquely blob shaped pickles in the jar...  funny thing - one variation put a beer cap style bottle cap on the pickle jar.. 

[KeegansCreation]

I think that's about ready for you to afix a Zazzle sticker to it (when it comes out of beta, fingers crossed for October release). Text-to-vector is probably a long way off but I think mockups are almost there. It was trained on Adobe Stock after all and that has, among other things, a lot of mockup backgrounds.

[Alissanos]

Still no access for me, so sad! The jar looks great!