I have been doing 3D art for over a couple of decades now to the point where I am kind of a pro. I started doing AI art a couple of months ago, as the field didn’t even exist a year ago. I have written quite a few posts about the 3D art community and the Daz Studio vs Poser stuff. I have recently posted a 2 part essay about the controversies surrounding AI art from a middle ground perspective. My general take on AI was that some of the controversies are overblown, but also real.

The affect of AI on art in general is not going to be as huge as everyone fears. At the current technology, it is a lot more difficult to get really great results than people think, and artists that know what they are doing are creating far superior renders than amateurs, meaning for now, the artists are winning.

But there is one sector of art where AI is currently having a major effect and that is in the 3D art community — My community.

The pictures at the top are ones I made myself. On the left is the “cover” of Date Ariane Remastered produced in Poser 12 superfly using the Victoria 4.2 model that I have been using for years. Poser Superfly is basically the Blender Cycles engine adapted to Poser, and represents the state of the art in 3D rendering. Most 3D art and animation you see today is done with cycles or a variation of cycles.

Over the past month I have been tinkering, trying to create an AI model using a custom Dreambooth model created with 20 renders of Ariane at different angles and expressions enhancing the model with various look alike celebrity images so as to create a prompt I can use that looks consistently like 3D Ariane as much as possible but cannot be recognizable as one of the prompt celebrities. Think I finally found such a prompt and dubbed it my “AI Ariane” model prompt.

I used a tool called image2image using the left image as a beginning step, and the “AI Ariane” model prompt to generate the image on the right.

The result is a realistic image of what the 3D character would look like if they were a real person.

Realism as a goal?

Date Ariane was never meant to look realistic. When people started imitating my Ariane Date Simulator online game using photos instead of artistic renders, it came off as “creepy”. The goal is to make beautiful characters that are on the good side of the Uncanny Valley. Realistic characters in an adult gaming setting should not be a goal. Beautiful pictures rather than realistic pictures are definitely the way to go.

Several popular adult visual novels are done using hand drawn images, and many people prefer them to the 3D variety. I kind of wish I had that drawing talent to match. I do what I do because it is what I am best at, but I appreciate the other artists and developers in this community.

I have no plans to go AI in my visual novels and avoiding realism is one of the many reasons I’ll discuss more later, but let’s get to the elephant in the room first:

AI People are more aesthetically pleasing than 3D People

I primarily do 3D to make games and tell stories. I am in the minority. The vast majority of amateur and hobbyist 3D artists (and we are talking 90% or more), are most interested in making 3D stills that are pretty to look at. So big is this interest, that the 3D community, currently dominated by Daz3D and Renderosity, are designed to cater to the making of original pretty pictures. It has been going on for 20 years now, and making these hobbyist sites millions of dollars in the sale of textures and 3D assets to hobbyists like me.

Enter in AI art generation. If your only goal is to generate “pretty pictures”, AI tools are faster cheaper easier and better than 3D.

The 3D art community is no longer growing as a result of the rise of AI, and hobbyists with little money to spend and little patience to get over the learning curve of 3D, are dumping 3D for AI fully.

Online store fronts of 3D assets are likely going to suffer the worst. I subscribe to several e-mail lists from these store fronts, and the frequency of promotions, sales, and coupons has gone way up in the last few months. It would not surprise me if some consolidation happens, and some of the lesser profitable sites are going to disappear completely.

Why “Storytelling” using AI is not currently possible

When I produced the AI Ariane picture at the top, I actually generated a bunch of them, using the same starting picture and prompt, just changing the seed, then picked out the best of the lot for the top picture. Here are some of the rejects. At first glance you might see these as a whole bunch of photos of the same woman in the same place, but look closer… (click on image to see a big version) the backgrounds are different, the purple dress is different, the necklace is different, hair and makeup are not consistent, even the size of her boobs are different. One even has an undershirt instead of a necklace.

Telling stories with images requires consistent sets, consistent lighting, consistent characters, consistent clothing, consistent body types, and especially consistent faces. This example was done in ideal settings, to make them as consistent as possible, and the resulting images are anything but consistent.

The truth about AI art is that it is not as easy as it looks in the finished images. AI artists will generate dozens even hundreds of images until they get what they are looking for. More truths:

• Results are extremely random and cannot be adjusted much at all within the AI program. You can move txt2img pictures to img2img or use masking or large canvases to make some adjustments, but these adjustments are still randomly applied.
• Every AI generated image you see will either have one person with one face, or if there are multiple people, they will have variations of the same face. AI Images with multiple different people are ALWAYS photoshopped or generated with masks with multiple prompts. Go to the Stable Diffusion Reddit to see what I mean. Most pictures will be single characters, or multiple characters with similar faces.
• Most AI art is test generated at small resolution because results are too random to waste time with large resolution, then use different AI enlargement tools to increase the resolution. Finding the right prompt to generate the perfect 720p image, then using the exact same prompt to make a 2160p doesn’t work, it will generate a completely different image. Every AI image in this post was generated at 800×800, and the seed 3D renders were shrunk to that resolution too, because larger either gives me memory issues, or produces multi person weirdness issues.
• Minor changes to an image are also impossible in the AI program. Ultimately, AI art can’t escape photoshop.
• As it currently stands, AI art really isn’t a threat to most kinds of artists. The multiple head problem prevents it’s use in comic books, the enlargement problem and minor change problem currently prevents it’s use in commercial art. These may be problems fixable in the future, but they are currently issues now.

AI can produce pretty pictures of individual people, but storytelling requires showing multiple different people interacting, meaning you have to AI create each person, then use and inpainting tools or good old photoshop to combine them separately.

It is way easier to just do it with another medium than with AI.

Why making games with AI, is frought with legal minefields that are best to avoid.

Besides the difficulty of storytelling with AI art, there is the problem of ethics and legalities, many of which will likely change soon, Here’s a simple video explaining what I mean.

Bottom line, I have no plans to use AI for making any of my games. I might use it to create one off backgrounds for individual pictures, but characters in the foreground will always be 3D renders, at least until the technology gets better or until the legalities get ironed out.

So I am staying firmly in the 3D market place to create my games. My fear is that the 3D market will collapse to the point that I can’t find what I am looking for. I’m seriously considering learning Unreal 5 and Metahumans, or Realillusions Character Creator which are the latest professional quality gaming creation tools for making future games, but I’ll have to research licensing and other things to see if I can even make adult oriented games with these tools.

Lots of unanswered questions with no existing answers. For now I am sticking with what I know, and hope it doesn’t go extinct too soon.