Milestone #2 Voting Results!

[Eric.B]

Can't you somehow use generic AI for creation of car models? I believe it would save loads of time and provide some great results if it is a possibility.... could be even integrated into the game therefore couple clicks "generate new body" could be an option for click click click until you like what you see... just an idea 

The New Vehicle Designer entirely removes the need to place parts on the car, etc. It's a completely new system. There isn't even a need for the "predesigned" anymore. It can create entirely random designs with a single click. You can dig deep into the design process, but you'll only interact with the UI. There is no more interaction with the 3D model.

---

If you're talking about Generative AI, there are a number of issues with your idea.

Firstly, Generative AI has not advanced enough to generate good quality 3D vehicles with realistic morphable animations that allow players to alter the vehicles to their liking. It is almost to the point where it can do humanoids like this, but there is much more source material for such systems to learn from with humanoids than car creation systems. So, any system would likely only generate static models. The player wouldn't be able to alter them. I'm not sure how many people want to hit "generate" a few dozen times to get their vehicle design. (If we remove the aspect of building your own designs, then I should remove the entire system from the game.)

Secondly, if you included such a system in the game, the minimum system requirements would go up from Geforce 4 to Geforce RTX 3080.

Thirdly, there is the copyright issue. All Generative AI has to be trained in order to produce their work. A lower court in the US has already said that Generative AI training data is clear of copyright issues, but there are challenges in higher courts. The EU has just started challenging the copyright legality of Generative AI. Many expect these systems to lose in the EU. This has prompted companies like Valve to ban ALL games with Generative AI content unless they can prove you own the copyright to the training material. https://www.eurogamer.net/valve-says-ai-...t-on-steam Considering the rarity of morphable vehicle artwork, I would spend many of thousands, even up to tens of thousands, to secure the artwork to train the model. I would have to spend thousands of dollars in cloud GPUs to train it. And then hope that the results are decent. For 3D artwork, at this time, it isn't.

Fourthly, it's not really an issue for me, but it's uncertain that AI can generate copyrightable content. The US said it could, but I think the EU said it can't. Again, it doesn't matter much to me. I made many of the plain text files of the game CC0 already.

Fifthly, Serghy is a talented artist. He deserves the work. I won't be giving him that much to do during AeroMogul. I'm not pocketing anything for Artwork-related bounties.

---

Summation, Generative AI isn't ready yet for 3D morph animated vehicle bodies that fit specific criteria. Any generative content of this type would have to be 2D. The only way to guarantee quality would be to produce the generations ourselves. Even if the program could eliminate the pure crap on its own, shipping a Generative AI system with the game would cut out about 90% of GC's user base because of system requirements. If the designs are not user-editable, such a system would alienate many current and potential users.

The major distributor that is 75% of our current income is blocking most Generative AI content. To abide by their rules would be more expensive than just creating the content ourselves, and that's assuming good results. The legality of Generative AI is murky at the moment.

Sergey works for cheap. He produces much better quality content than what you've seen in GC. He knows the engine and how the system works. The new designer can generate many variations of vehicle designs with the new artwork. So, there will no longer be a need for premade designs in the game. The AI will have its own vehicle bodies too. (Again, they don't have to use predesigns either.) However, we still have to make the base artwork for that system to use.

---

I use Generative AI with licensed training data for portraits in AeroMogul. I will eventually seek out public domain/open-sourced trained text data for use in AM/GC. I am not against the concept, but I understand the limitations. It's not there yet for many things. And until the legal issues are cleared up, you're unlikely to see this kind of content outside of the AAA companies who can afford to own the copyrights to the training data.