Swatcher

goddamnit I'm already coming back. how did i not think of this earlier (see edit)

"I'm prob taking a break after this for a little bit" A promise is a promise. I won't be doing much DALL-E 2 stuff from here on—I'll get only 15 credits a month now, which resets every 18ᵗʰ. I don't know if I'll be checking in too often, but I'm still open to checking requests ITT. (Link) Maybe someone else could pick up the slack. Check some other AIs I'm not overly fond of OpenAI, DALL-E 2's creator, anyhow. DALL-E 2 is good at realistic stuff VS a lot of other AIs(for now). OpenAI's done some other good stuff too like GPT-3. HOWEVER Anyway, go check out Stable Diffusion for free. It's new, open-source, highly-trained, and very good imo, in ways even better than DALL-E 2 (besides maybe at generating photorealism. They have different 'styles'). It also has more options/add-ons since it's open-source, such as 'img2img'. Or, just check one of the other gazillion A.I. models out there. There's definitely no lack of choice. Seriously, A.I. stuff has taken off pretty high, and I don't see it going down any time soon; there's always going to be more and more options in the future. Deuces

... It'll be pretty good. (Future war - Terminator 2)

Kinda gave up halfway through. I tried stitching something better together from the extra images. I added a little orange fire glow to the left ship's mast (only non-AI thing) and moved a couple things around, and besides that this is 100% AI images Full Resolution

And I've run out—The small outpainting of PoE 2's official art. I might've been too hesitant with wordy prompts, but honestly I was expecting some gibberish garbage from my over-detailed prompts! Maybe older AIs made me paranoid I wish this website didn't scale down highres uploads like this to a half-sized .jpeg. The original is a 2944x1024 .png. Is there a good website to direct link images that ISN'T imgur? nvm found a reasonably reliable site. Full resolution Process: Y'know, since I got 4 images to choose from for each square, I could probably stitch together something even better-looking from the extra images.

It's almost expiration time, and no one's coming in. Guess it's time for me to spend the rest on dumb crap no one wants. Here we go! Eh I'll just go halfway on 1 & 2 for the outpainting. I considered a bunch of dumb memes or sh!tposts if I can't even use this thing for portraits, like Aloth getting mugged in a NY subway or something.

Do it, bro! For real, I'm prob taking a break after this for a little bit anyway; maybe lurk and see what other people make. I'll plug this site again: http://lexica.art, if you wanna get a taste of Stable Diffusion images by other people including portraits A good portrait requires the ideal male body (link) Eyyyy, my 50th post and 1,000 views at the same time. Timing. Glad this AI stuff is finally getting some attention here, for the first time.

tbh I wouldn't bother with PoE portrait requests anymore, specifically. You've all seen what they look like in the prev page. Unless anyone has some untapped crazy idea or prompt, the portraits are all going to look stylistically like the results from this thread. I've wrung it out. This is as close as it gets for DALL-E 2; its weakness is not recognizing "Pillars of Eternity". Try again with Stable Diffusion instead. SD recognizing "Pillars of Eternity" will make a huge difference. On top of the difference I noted about the AI 'styles' above

"despite the issues with faces" Ehhhhh for DALL-E 2, the faces haven't really been the issue (but the two inpainting attempts were a little f$cky tho. Don't inpaint faces). Pretty much almost all of them were good, which is extremely impressive for an AI. This year's been ridiculous with AI and deep learning, for real. The real big issue is matching PoE's art style. Super hard to get the game's art style without the AI recognizing the game. Ironically, the weaker AI on page 1 DID recognize PoE, despite having way, waaaay less training data. That's why page 1 matched better. Seems kinda silly in retrospect, but it makes sense. Your prompt requests: [Death Godlike] Painted portrait of a faceless demon paladin with horns in modern digital painting style El diablo [Nature Godlike] painted portrait of a male dryad cleric with antlers in modern digital painting style 6 CREDITS LEFT

ye *painting Watch the spelling. Can't afford to re-run it. I can't fix it once I enter it, either.

btw, a lot of these editing tricks can work with versions of Stable Diffusion too, inpainting and whatnot. It's also worth mentioning that Stable is good but not as good with realistic-looking stuff as DALL-E 2 is. However, with digital art it's just as good. In fact it emphasizes it. Talk about ideal! Doing these digital art portraits is literally one of the best use cases.

This one, right? I could, but in the .gif you saw those 1024x1024 generation squares, they each cost a credit. I'd probably use up my few remaining and then some. High resolution is expensive yo I could INSTEAD scale the img resolution down to fit inside the square (like this) and generate one square around it though, at the cost of resolution. 1 credit only But if I go big with #1, I'd save it for the the last day(17th). I want to see if anyone else wants a request before I destroy my remaining credits.

Anyway, still open to taking any request. Gotta burn through these somehow I did a dumb and tried inpainting on multiple DALL-E Mini faces from one post at the same time, despite how bad faces went last time. Look And the close-up: 8 CREDITS LEFT

Oh yeah, unrelated but I don't think I've ever showed off properly what the built-in Outpainting looks like in DALL-E 2. Here, a .gif: Final Result: Couple more This is how you get the super high-res pics *and also how you burn through all your credits

Anyone, feel free to request the full resolution (1024x1024) for any of these pics if you really want them I don't think they're going to get deleted from my DALL-E 2 account if they get too old, but who knows how OpenAI works?