Howdy guys! I've been messing around a bit with some various AI programs, trying to see how they might enhance our capabilities in modding, since our community is pretty small these days (though our existing content creators are freaking amazing!). I've been playing with Elevenlabs to generate voicepacks and such and have had some success, but one of the things that I've really been wanting for my Saturday Morning Adventures mod is an image I can use for the main menu background. I have in mind a cool, big group shot of some of the heroes and villains from the various shows battling it out, in classic comic crossover fashion.
I started experimenting with OpenArt, and I made some progress, but I can't manage to get it to produce something that doesn't end up with some hybrids and monstrosities. And, for some reason, I suppose because of the name in common with the lead Silverhawk, it kept adding Marvel's Quicksilver into the scene, haha. I was wondering if there is anyone out there in the community who has more skill and experience with this kind of thing than I do and might be willing to give it a try for me. Alternatively, I wonder if there might be some folks who have enough Photoshop experience that they could weave a whole out of the various broken bits that I managed to produce.
Here's my (often times unexpectedly hilarious) poor efforts.
(http://i.imgur.com/Ij9lLtim.png) (https://imgur.com/Ij9lLti) (http://i.imgur.com/iQobf2Ym.png) (https://imgur.com/iQobf2Y)
(http://i.imgur.com/474HwAjm.png) (https://imgur.com/474HwAj) (http://i.imgur.com/x0zzv1Wm.png) (https://imgur.com/x0zzv1W)
(http://i.imgur.com/VoewRT2m.png) (https://imgur.com/VoewRT2) (http://i.imgur.com/yVz9R0Ym.png) (https://imgur.com/yVz9R0Y)
The most logical approach would be to create the character designs one at a time. This means finding and/or generating the desired image for each character and positioning it. Then, use a relatively good AI to match them up. Midjourney is generally considered the best, but I've never used it. They always have errors, like Cheetara's bo staff that shoots from the tip, Donatello with a machine gun and sword instead of a bo staff, Leonardo with a sai, Quicksilver in a Silverhawks version, and even a Rocksteady with a turtle torso.
You should generate a prompt (a good one, to give the AI ��instructions that, without specifying the character's identity, ensure it accurately represents each one). That's what I would suggest. If you leave it up to the AI ��to search for and interpret the character's design, even if you tell it which series or cartoon it's from, you'll have design discrepancies. You must ensure the design is consistent regardless of the character. For example, instead of saying, "I want Bebop from TMNT fighting Dakota Dude from Moo Mesa," you should specify the character. To say, I want this attached character model (attached image 1) fighting this attached character model (attached image 2) in the same line work and style, respecting 100% fidelity to the original design, on a background similar to attached image 3. This way, you tie the character design together regardless of whether you created it or it's from 90 years of comics. When using generative image AI, you should always construct the prompt well, based on what you want to achieve, and limit creativity and errors by adhering to the attached models. This way, you'll have less chance of things going creatively wrong.