asteria ai test
Added 2023-03-26 01:38:51 +0000 UTCTHIS IS NOT REAL CONTENT! And no I'm not turning into an AI artist. That's an oxymoron. Just learning. Itβs controversial, polarizing, and people hate it so naturally I tried training a model on my own work using some renders from almost a decade ago. edit: and rie
Why? It's fun to play with. But there will be specific uses in the future of 3D art, especially for asset creation and postwork to address anatomical limitations using daz figures. What impresses me? The lighting and the ability to generate realistic materials (e.g. the floor stones). What doesnβt? Almost everything else. Will it replace me some day? Most definitely π
Comments
Lost another creator to AI
CocaFristyHoi
2025-01-21 17:53:44 +0000 UTCTalking of people being away, has anyone seen The Creator on here lately? Iβve joined recently and am hoping to get access to the Google Driveβ¦ if anyone can help my email is kissella.love at Gmail. Thank you! π₯°π
Kissella Love
2023-10-10 14:17:15 +0000 UTCAs with most AI right now, itβs great for creating initial drafts and options (for discussion?), but the final edit requires your God-like hand to form the most perfect creations!
Kissella Love
2023-10-10 05:52:21 +0000 UTCwas this an app or a actual site or something and how do you find stuff like that xD
Ben Mayers
2023-04-22 07:26:41 +0000 UTCCould you also try to do some realism ai ones? I mean with your work you are even now pretty close, but I wonder how that would look like going a few steps further.
xshwigermaker
2023-03-31 03:16:35 +0000 UTCI think the anatomy of frame 5 is kind of exciting
M
2023-03-28 10:20:21 +0000 UTCI "tried" training a model using about 100 images in the dreambooth extension for automatic1111 and merging the checkpoints with an existing futa-focused model. I also tried making a LoRA. Using prompts indeed resulted in some nightmarish abominations, so yes these are img2img. It's not clear if that tiny amount of training added anything to the results. Even if I trained every piece I've every done it isn't near enough for a standalone model by any stretch of the imagination. At any rate this test is not what I'll be using it for. I just linked an example for one possible use in a typical 3d workflow in the post.
nothing
2023-03-28 00:26:05 +0000 UTCA couple of reasons. 1) It takes away work from real artists and illustrators. 2) It may not look great now - but what about next year, after it's been better trained? 3) What's stopping people from creating "deepfakes"?
rtpoe
2023-03-27 23:29:04 +0000 UTCI always wanted to see some of the older characters return and maybe run into the newer ones so this was a fun test.
DatAsuna
2023-03-27 18:10:29 +0000 UTCFrames 5 and 6 was your work?
gp
2023-03-27 15:31:48 +0000 UTCFor textures I suppose, but how SD has been made is deeply flawed. A lot of copywritten work was funneled into a commercial product under the pretext of research, and now it's being used to push actual talent out of the industry.
MC3DX
2023-03-26 21:05:43 +0000 UTCKatie3DX is now making art that has the exact same appearance as every other lewd AI art because they're all using Stable Diffusion models. If everyone has the same style, why even be an artist? There's no soul to any of these works.
MC3DX
2023-03-26 21:03:23 +0000 UTCAn image prompt is still attempting to train the diffusers. Using an existing model's NN and then trying to fit an image requires training data, the image, to be revealed to the model with each layer pass. Undoing Stable Diffusion's biases is very difficult, that's why it's haphazardly mashed the typical anime style that SD generates onto Nothing's image. It's attempting to fit what it's interpreted from his image. Training data by necessity gets leaked into the model when you force SD to try and match an artist's style. He's not remaking the SD model when he does this, he is forcing it to be biased which is still in essence training that copy of the SD model.
MC3DX
2023-03-26 21:01:05 +0000 UTCI'm pretty sure these weren't generated by training an AI on his works and then simply giving it a text prompt. Ironically you wouldn't get results that were so close to the original by doing that. Pretty sure these were generated by using the original 3D art as an image prompt for an existing model, hence the complete change in art style since the AI hasn't actually been trained on nothing's art style at all.
Vices
2023-03-26 19:31:35 +0000 UTCThis looks incredible, and I suspect that with experience with this tool you will be able to create works that capture your style effectively while saving you a lot of time.
Shadow Shaman
2023-03-26 14:33:32 +0000 UTCThere isn't enough of your artwork to truly train a robust model that will generate unique combinations. The reason why these look so much like your old work is because from a statistical standpoint your dataset is far too small to produce anything that won't simply reflect the training dataset. Honestly this is a really good example of how these models are intentionally overfitting the training datasets to look more like the original works, rather than something truly new.
MC3DX
2023-03-26 13:37:33 +0000 UTCI think that's the right mentality, play to the AI's strengths to complement your own skills. Rather than using AI as an inferior replacement for your own strengths.
Vices
2023-03-26 09:35:59 +0000 UTCWhat, the "modded" series isn't big enough for you? :)
nothing
2023-03-26 06:31:26 +0000 UTCWould be great if you could make an app or something tthat people could subscribe to that let users generate images that looked like your work. Something like https://pornpen.ai/ but exclusively trained on your art. I know that if people could save the images they generate from such an app it could be an issue for you, but I think if you could figure it out there could be a real market for something like that.
Yooki
2023-03-26 05:32:46 +0000 UTCtwin, you are not alone, dear brother. I've been a fan of nothing since 2015. I remember all his futa were Large.
Case Green
2023-03-26 04:07:06 +0000 UTCgot to admit, nothing, i loved you more then in 2015 when your girls were HUGE.
Case Green
2023-03-26 04:05:04 +0000 UTCwhoa, nothing getting back to basics. i recognise that background from 2015, wtf
Case Green
2023-03-26 04:03:24 +0000 UTCI don't really know why everyone is talking about ai art. I doesn't look that great.
org123
2023-03-26 03:47:39 +0000 UTCI love that you're trying this. This looks like a small training set. If you do more you should be able to start getting new works that won't be so familiar. Don't be afraid of it!
Mike
2023-03-26 03:31:27 +0000 UTCI vastly prefer your original work!
TwinArchangels Gaming
2023-03-26 02:41:53 +0000 UTClove no balls β₯
Xforcen
2023-03-26 02:30:39 +0000 UTCIt's not Summer, it's not a substitute or replacement here for content, and by very definition it's derivative and flawed as you mentioned. But it will have specific uses in the future like generating procedural textures for 3D materials. I'll be left in the dust if I'm not continually learning. It may be incorporated behind the scenes in workflows without being the ENTIRE workflow.
nothing
2023-03-26 02:29:00 +0000 UTCI genuinely don't understand the rush of so many of my favourite 3DX artist's to embrace AI. First Beren144, then Katie3DX, Miro, and now, possibly, you. AI art looks like AI art, devoid of any of the personality the original artists imbued in their own work. Not to mention the little glitches that AI always seems put somewhere in the image. Whether it's mangled fingers, random backgrounds, or in Summers case a twisted ankle. ππ
Kraig
2023-03-26 02:20:08 +0000 UTCInferior to your non ai work. But still 5/10
Hyporea
2023-03-26 02:19:08 +0000 UTCLooks fantastic! Itβs a great addition to your existing work. Iβm not gonna lie, sheβs no Summer, but this model looks close to your art style.
Steve Osborne
2023-03-26 01:54:24 +0000 UTCStable Diffusion. I tried training a LoRA on some old renders and merging checkpoints with futagen2 and truefuta1.5 models, also futacumAnything LoRA
nothing
2023-03-26 01:49:14 +0000 UTCThis looks good! Which ai generator did you use?
Moebiusatplay
2023-03-26 01:44:14 +0000 UTC
