- This topic has 18 replies, 5 voices, and was last updated 11 months, 1 week ago by
.
-
Posts
-
filled everywhere with the possibilities of AI - But I steadfastly ignored them. And then an idea came to mind.... (I used DeepSeek - ChatGPT disappointed me with its rules and prohibitions)
I ignore G9 - I use G8.1, G8. I will give only positive results and my ideas - based on the experience I received.
1- With the help of AI, I turned the character Mika7 (for G8) - into Scotlyn (for Victoria4) - I fed the AI ​​screenshots of Mika and Scotlyn - it gave a recommendation for correction - and in ~5 minutes I got the desired result - and the most problems were caused by the shape of the eyes (I should have taken Mika8 as a basis). AI is good for correcting characters in the direction you need......
2-But this is more complicated and is good for those who sculpt characters themselves. With the help of Deepseeks we create a promt - specifically for a specific site. And then I used the services of these sites: tensor.art and seaart.ai - there they replenish points daily - enough for 20-30 generations - enough to solve the problem - creating reference images.
Tensor.art - there is a technique: we take the original image of the original head - and with the help of promts we generate the necessary references (I do them separately - profile, then front, etc.)
Seaart is even more gorgeous in this regard.... When generating on both sites - there is an option to choose a LoRA model - which gives an excellent result when using:
This is Tiffiny Towers(NFSW) - one of the generated images on seaart ... Why did I like seaart? - it allows you to create a LoRA model yourself and generate references with it (I haven't done this yet - I generated enough for a couple of months... 🙂 ) Take 15-30 photos of the character - put them in and at the output we get a model that will generate the necessary angles - Just Follow the Rules of the resource.... For example, generate in an a-pose, full-length - I couldn't do it - it's blocked as NFSW....
Here on the resource there are those who use Stable Diffusion on their computer - Ask the AI ​​- because you can download the tutorial and LoRA - and make the necessary references yourself without any censorship.UPD1: Advice for those who will use seaart - I got the best results using - Flux1Dev.
UPD2: If you used other sites - to generate reference images - please share!!!
and
UPD3: If there are willing users among the resource users - maybe we will open another type of Requests - only not models - but references for 3D sculpting??? For those who can generate on their computer - without censorship.I'll continue: maybe someone will be interested...
With stable diffusion, the results were disappointing. However, the Flux model using LoRA yielded interesting results – they could be used to create a decent model: for example
Love how you are not afraid of experimentation between AI and 3D. For detailed prompt generation I use -
Florence-2
docsbot.ai
joy-caption-pre-alpha (uncensored)Frank21 Thanks! I've bookmarked the sites - they'll come in handy!
.....
Here are some more examples - they're from another resource: Tenzor.art
Flux and the LoRA software for it (available on the website) were used.
One of the generation options: StarLord
Two variants of Nemesis:
Evil old woman: (add glasses and hair..)
...........
Using AI, you can generate references for creating your own character or editing an existing one. The number of "heads and their projections" can be controlled using the prompt. Users of previous generations of figures were left without content—and this use of AI capabilities will allow them to be given a second, and a third, and... life. 🙂
Also, chatgpt and deepseek claim it's possible to generate textures using the UV base. I don't know when I'll have the free time to try it out.UPD: Such references can be used not only for sculpting in ZBrush or Blender, but also for generating a 3D model and creating a morph using Wraps.
Where do you go for models?
I browse Civitai most mornings but find that it's getting harder to wade through the tsunami of cartoon, manga and anime stuff that I have no interest in. I've enabled every tag I can think of to block this stuff but they have no effect. I get the feeling that blocking is there just to make people think they're filtering, while still jamming this stuff in everybody's faces. Blocking creators also has zero effect.I haven't used civit.ai, they only give free points once. Although, yes, there's a lot of LoRa there... I'm "afraid" to use it.
I use seaart.ai and tensor.art. Tensor has paid LoRa, but I haven't encountered any on seaart yet—it just limits the number of downloads per day.
upd: And the best results are obtained with models for flux... for stable diffusion and its models - a good result was obtained less often... Although maybe I couldn’t find a good promt??
upd2: talk to deepseek or chatgpt - they will explain: How to make textures with LoRA, models using stable diffusion - I can’t check the computer is weak."I get the feeling that blocking is there just to make people think they're filtering.."
Genre filtering is badly broken as you say. And with Civit's crack down on NSFW their product filtering is a nightmare. You have to run through each "age" filter separately to see what's there as there is no "show me everything" mode. But in general I think that Civitai LoRas are better made and more reliable compared to the tensor.art ones I've tried.
@blockhead - if you want to use Civitai to render, PM me with your screen name and I'll transfer you some blue buzz points. I don't use them as I render off-line on my PC.Frank21 Hi. Thanks! But for now, I'll pass... I've made reference sketches of the characters I'm interested in, and I'm planning to spend my free time bringing them to life.
....
I'd like to clarify: are the points for generating images also awarded only once? On civit.ai...You can earn blue buzz by clicking on the ads but with my browser I never see them anyway. You can earn 25 by making a new image and posting it - only 25 per day. You can earn others by feedback on posts too. The biggest prize in points is dobbing in images/Loras that may break the TOS for 100 points each. This kind of defines the extreme moral panic that has taken over Civitai earlier this year. They're scared shitless of being blocked by the credit card companies so no celebs or HC and no Anime Loli. They've basically become a Disney world where everything is rainbows and ponies and sunshine.
Thanks for the info. In the meantime... - here's how RKS makes its models:
"Can LoRA be used for 3D?Directly – no
LoRA doesn't output meshes (.obj, .fbx, etc.), only 2D images.
But there are workarounds:
LoRA → Images → 3D Reconstruction
Generate a batch of images (front, side, back views) using LoRA.
Plug them into a photogrammetry tool (e.g., Meshroom, RealityCapture, Luma AI).
You get a 3D model.
Here, the quality will depend on how smooth and consistent the images LoRA generates are.
Use LoRA in a 3D generation pipeline
Models like Stable-DreamFusion, ProlificDreamer, Tripo AI, and Meshy have recently emerged.
They take a text query → generate 3D images. This is where LoRA can be "blended" to control the style (for example, "make it anime-style").
LoRA as a texture style
Even if the 3D mesh is obtained using another method (for example, DECA creates a face), LoRA can be used in Stable Diffusion to generate textures/materials.
In other words, one model provides the geometry, and LoRA colors it in the desired style.Summary
LoRA alone won't create a 3D model.
But it can be used as a tool for generating 2D references or textures → and then assembling 3D data from this (using photogrammetry or 3D-AI)."And then - Zbrush, Blender, or Wraps...
In the adjacent threads there was a link to a site where he sells character heads - which then appear for Daz. I'll have to try it....They're scared shitless of being blocked by the credit card companies so no celebs or HC and no Anime Loli. They've basically become a Disney world where everything is rainbows and ponies and sunshine.
Huh? I see a ton of that shit. The celebs aren't named but it's usually obvious who they are. I don't see that porn has changed much either, there's still a pile of weird hardcore stuff everyday.
The only change has been the loli shit which I never saw much of anyway.I just finished going through the new items for the morning and I saw several celebs, not named, but listed as'Influencers'. There was a blowjob, some disgusting anal thing, a foot job, one named celeb riding a horse as her boobs popped out of her shirt. Another named nude celeb who I wasn't familiar with. Let's see......a couple of tentacle scenes, one realistic and one cartoon. There's a shitload of cartoon or anime game and TV show characters too.
....and one very cute kitten.
So not much has changed.I suspected RKS was using AI since he has no talent whatsoever, well at least now he doesn't steal anymore... hopefully.
Hey everyone. EXPERIMENT!!!! There's a game about AI that's underrated, in my opinion. Nexus: The Jupiter Incident. The AI ​​claims it can make a LoRA based on the ships in the game. Is anyone interested?
Here is an example of a model from the Internet - it is someone else's model - but it shows quality.
I put the ship on the back burner. Two attempts didn't yield the desired result—quality LoRA requires high-quality images—and I'm short on free time.
But there are some interesting results:
The last four images are of actresses styled like Barbie dolls. The result is hilarious.
Even simply using such references - using the existing morphs - gives interesting results.....
- You must be logged in to reply to this topic.
