AI and its integration into Spine is indeed a topic of interest and discussion within the Spine community. The developers of Spine have considered the potential of AI in some capacity. However, they also recognize that the results from AI can often be less than satisfactory, although AI technology is rapidly advancing. The Spine team values the detail-oriented nature of animators who desire full control over their animations, suggesting that while AI and automation can be useful, they should be implemented in a way that respects the animators' need for control.

Physics is mentioned as a more immediately useful feature for Spine, as it can automate tedious secondary motion and add dynamics to animations, which might be a more practical approach than full AI automation at this stage. The developers are open to the idea of AI but also cautious about its current limitations and the importance of maintaining the quality and control that animators expect from their work.

In summary, while there is interest in exploring AI within Spine, the focus remains on ensuring that any integration of AI or automation respects the craft of animation and provides real value to animators. The developers are mindful of the rapid advancements in AI but also of the need to maintain the high standards of animation quality that Spine users expect.

That is neat, though the animation is simplistic. Even in the future I think that will be the case: AI can do a thing moderately well and that's good enough for simple stuff (like memes, Xmas e-cards, and one-offs), but beyond that you'll still need a human.

More interesting than getting a finished product from the AI is to have it do tasks for you and then hand over the work so a human can take it from there. That could be rigging or blocking out an animation. Using it as an optional tool that augments rather than as a complete human replacement gets around many of the reliability problems.

The current state of AI is that it can do amazing stuff and it's plausible that in the future it could do literally anything, even going so far as to say it could possibly take over the world and enslave us all. There's not much point in worrying about that worse case. No one knows how good it can get or how the tools will change the landscape for human careers. Most likely AI can replace humans for trivial tasks, raising the bar for what a human needs to do to add enough value to get paid for their efforts. While the bar may be higher, humans will still be needed at some level, if only to push buttons to have the AI do the work. If that is not the case, then we just have to hope our AI overlords are kind to us. Maybe start saying "please" to ChatGPT requests.

Simplify the task. No need for physics, diffusion control, or skeletal control. As an animator, we need a few simple things:

Automatically split images from any source and break them into parts for the spine animation system.
Work references two points: 1. Split major objects into parts. 2. Complete the missing parts of split image sections that are typically obstructed by perspective.

I personally believe AI won't be better than what I can do.

Spinebot

Nate

I think most designers would agree with that. You shouldn’t be left behind in terms of AI. Many programs are left behind because they can’t keep up with AI. However, I think you have to put AI into Spine 2D to even be able to say that it has AI. It might not be an AI that has to do with animation. For example, an AI that can render images with text, or an AI that can detect images and text in layers and automatically create a bone system. So, even if it’s a small update, I think it’s definitely important to show that Spine 2D has AI.

A few suggestions:

Automatic Rigging
The skeletal system (rig) of the characters could be created automatically. The AI ​​would analyze the structure and anatomy of the character and suggest the most suitable skeleton, minimizing user intervention in the process.

Style Transfer
The user could transfer the style of one animation to another. For example, it would be possible to apply the movements of one character to another character in an AI-powered way.

Error Detection and Optimization
AI could detect potential errors in animation projects (e.g., broken movements, incorrectly connected bones) and make suggestions.

Advanced Effects for Animation Quality
AI could make animations look richer by automatically adding shadows, lighting, and details.

I've worked professionally in ML/AI for years, so I'm quite familiar with the benefits and challenges of using such models.

We are not going to chase a fad and put "AI" into Spine, if there is no clear benefit or if the tech isn't there yet.

Currently, there aren't any state-of-the-art models, that would make sense to deploy inside the Spine Editor. And if there was, it is unlikely they'd run on-device on a large percentage of our user's machines due to hardware requirements. Which would require us to build complex and extremely costly server infra. That is not our core business.

Your suggestions sound good on paper, but I challenge you to find a single model that can actually do this well enough or at all and at a quality level that doesn't make animators cry. Did you by any chance let an LLM generate those lofty suggestions?

I did some tests with Gemeni.
It allows importing multiple .json files, and I asked it to adapt the new .json file based on the ones sent, considering the animations.

The result was quite poor, but the idea is to train the .json models provided by animators or royalty-free ones and generate basic animations, for example, to speed up game development in design.
Then, this would ease the transition from placeholder animations to professional ones.

For now, all my sprites can be replaced by Spine in my game engine.
But I'd like to be able to replace them with somewhat rough Spine animations that would give a better preview of the artistic direction and performance before asking professional animators to do the work.

The model has zero context with regards to the actual animation such a JSON file would represent, and that can not be fixed by just feeding it more of the same. Alternative approaches, where you feed the model both JSON and e.g. frames of the actual animation, can not be build using off-the-shelf models, much less large language models like Gemini.

There are startups that try doing something similar for 3D animated models, with massive amounts of venture capital backing them up, and no existing product to maintain and improve. Doing something like that for Spine would mean basically starting up an entirely new business, with a new team specialized in the task of building and productizing machine learning models. This is not something you can create in an afternoon and the ROI from a business side is more than questionable.