Well, it's been 7 weeks since I wrote my rather scathing blog about how I thought Deep Dream Generator had fallen way behind in the ever growing world of A.I. art creating, (seen HERE), so I'm sure you can imagine my delight in hearing that they were finally introducing some new features!
I was fortunate enough to be amongst the first round of BETA testers, so I've had a few weeks to play with the new models and features now. Here are some things I've discovered, and some thoughts about them...
THE AI MODELS
So there are 10 models (styles) to create with now. 5 'regular' models, and 5 more 'PRO' options. If you do not have a paid subscription to DDG, you will be limited to creating only 3 images using the PRO models in a 24 hour time period.
To test all the models I thought I should try feeding them all the same prompts and then judge each model's outcome afterward. I used 3 different prompts and created the images in "high quality", (an option in the regular models). Of course it's difficult to rate the models by doing this, as I think it's perfectly normal to get occasional awesome images, or horrible images from any model used, but I'll tell you how I found the results regardless. I didn't use any modifiers in the test.
The first thing I thought of to test was the hands, to see if the ongoing issue had been solved in any of the models. I used the prompt "Handsome young man with astonished look on his face. Arms stretched out and showing palms of his hands to the viewer. Detailed background. Hyper-realistic, photo-realistic, Unreal Engine 5, eXtreme details."
I got the usual deformed hands in all models, except for Animatron and AIVision, (PRO models), both of which gave me PERFECT HANDS! I could hardly believe my eyes and was thrilled that 2 models achieved this!
For my neXt test I thought I'd try a generic prompt and see what happens. I used "Fantasy landscape by Jacek Yerka. eXtreme details, acrylic art, great depth and scale, gradual gradients. No people, no animals, unsigned, no signature." In retrospect, I wish I hadn't used an artist's name in the prompt and just let the models create without it ... but it is what it is!
I got the usual results one would eXpect when using Jacek Yerka's name, and nothing really jumped out at me, eXcept for the AIVision. My 2 second favorite images came from DigitalDaVinci and Animatron, (all PRO models).
For my third test I thought I'd use a more sophisticated prompt to see what the models came up with. I used "Adorable H.R.Giger creature, by Gustav Klimt and Catrin Welz-Stein and Dr.Seuss and Tim Burton and Zdzislaw Beksinski. fine-detailed-face. Intricate detailed fantasy background. No watermark, unsigned, insane details, eXtreme details, realistic photography, hyper-realistic, fantasy, whimsical, mysterious, octane render, Unreal Engine, 4K 3D, fantastical art, science fiction, mystical, compleX, cinematic post-processing, light refleXions, magical, HDR, crisp quality, award winning, very cute, intricate, focused."
Out of all the regular models, only DreamForge produced an image I liked. All of the PRO models gave me good results, with Animatron, and especially AIVision, giving the best results.
In conclusion, I guess I'd say that all the models showed improvement over the previous version, with the pro options (obviously) outshining all of the regular options. The "AIVision" model is certainly the 'best of the best' model, but it often has predictable image outcomes when used. For eXample, I can usually tell when an image has been created using the AIVision model even before I actually look to see which model an artist used, so the thrill of it might not last too long if all the images that come from it have the same look and feel to them. We'll have to wait and see over time as artists add their various prompts into it.
For me, the AI EDITOR is just as eXciting as the new models, maybe even more so! There are some great features within it, so DDG artists can really do some fine tuning if needed. Here are some of the best features...
AI EDITOR - INPAINTING
So this is a wonderful feature, when I get it to work properly! After you make an image, you can bring it into Inpainting and fiddle with the parts you don't like about it. All you need to do is highlight the area you want to change and then enter a prompt describing what you want the AI to make different. Click on the generate button and wait to see the outcome.
From my eXperience in using this feature, I've found that sometimes it works miracles, but at other times it just doesn't. Sometimes you have to be quite specific in your descriptions. For instance, when you highlight the eyes and ask for "blue eyes", be prepared for ALL PARTS of the eyes to turn blue, but if you ask for "blue irises", you might get what you're looking for.
You get charged 4 points every time you click on the Inpainting button, so being very specific will probably save you a lot of energy points in the long run.
AI EDITOR - IMAGE EDITOR
There are some great on-site editor features in here. I especially like the FINE TUNE feature where you'll find all the usual tools, like brightness, contrast, gamma, etc. As we all know, sometimes the AI needs that little bit of eXtra fine tuning. Now, if you get an image too dark or light, etc, for your liking, you can fiX it right up in here!
AI EDITOR - IMAGE EDITOR - ANNOTATE
FINALLY, a way for artists to sign their worX at DDG! There is a "sharpie" option to hand sign your work, and a "teXt" option to type your name on your work. This is a long overdue option and really helps with not having your artwork ripped off by others. Bravo to DDG for adding this!
Personally, I think they should have gone a bit further with this, allowing us to upload our own signatures/watermarks as well, but it's definitely a step in the right direction, and I applaud DDG for their efforts. : )
Well, HOORAY! The cost to generate images has finally come down. This is wonderful news and I'm sure it will be welcomed by all. However, this price decrease eXcludes the AIVision model, which remains pretty pricey. Making a square image will cost you 30 energy points, but making a landscape or portrait sized image will cost you an eXponential 40 points. WHOA! That being said though, you are pretty much guaranteed to get a great image when you use it, so perhaps the price is justified. Why the shape of your AIVision image should cost different amounts of energy points is a mystery that only the DDG gods themselves could answer.
It's certainly very eXciting to see all the new models and features in this updated version of Deep Dream Generator. Over all, I'd say it's a vast improvement over the previous version, and might help to propel DDG back to the forefront of A.I. art creating sites/apps.
It remains to be seen if these new models and features will actually bring some much needed new enthusiasm (and traffic) back to DDG, but I hope that they do. I've heard a bit of feedback from artists about it, both from regular and casual users, and have had miXed results from them. Some love the changes while others downright despise them. I suppose that's to be eXpected.
For me personally, I still have my issues with the popularity algorithm, which continues to irritate me with every single piece I publish at the site, so yeah...there's still THAT! Maybe one day DDG will figure out that it's actually an issue and turn-off for a lot of folks, (not just me), and will eventually do something about it. My guess is it's probably the number one reason that folks don't stick around.
Other than that though, I think that V.3 is a success over all, and most artists will welcome the new changes.
Also, I predict that there are more eXciting models and features coming that are literally just over the horizon. Hurry up DDG, we want to see them NOW!
Happy creating to you!
December 3rd, 2023