For developers, the most pressing question is: "Can AI actually be used for content creation?" rather than just producing some impressive-looking but ultimately messy "useless images" or "useless models".
Under this proposition, AssetHub from Japan and Kapnetix from San Francisco, USA, and the UK, respectively, offer quite pragmatic answers from the two dimensions of "static modeling" and "motion capture." Simply put, they don't pursue the perfect magic of AI "one-click generation," but rather use AI to intervene in the most time-consuming traditional "manual" processes, shortening 3D production from weeks to days, and even enabling motion capture that previously required millions of dollars in a photography studio with just a mobile phone.
AssetHub: AI modeling should not only aim for "likeness," but also for "usability."
There are many 3D generative AIs on the market that can generate a 3D model simply by inputting text. However, in the eyes of AssetHub CEO Takuya Goto, these models often have fatal flaws: messy meshes, textures stuck to the model and cannot be modified, and unnecessary details (such as ghostly hair) that cannot be removed.
"For game developers, uneditable models are garbage." AssetHub proposes an "AI-Native 3D Workflow" solution.
Unlike traditional photogrammetry, which is prone to noise, AssetHub's core technology lies in "part decomposition." When AI reads a 2D character design, it doesn't directly generate "a 3D object." Instead, it first understands "this is hair," "this is shoes," and "this is a skirt," and then breaks them down into individual components.
In his presentation, Takuya Goto demonstrated how this process allows them to first "style transfer" and "remove backgrounds/shadows" from 2D anime characters, and then use AI to generate a preliminary 3D geometric structure. AssetHub emphasizes the concept of "Human-in-the-loop"—AI is responsible for generating 80% of the content structure, and the final 20% is refined by professional 3D artists (Retopology).
This solves the problem of AI-generated models being "only for viewing and not for use," allowing the manual modeling process, which originally took two weeks, to be completed in just a few days. Moreover, the resulting models are productivity-level assets that are "clearly segmented, can be bound to a skeleton, and can be used in the engine."
Kapnetix: Say Goodbye to Millions in Studio Setups, One iPhone Handles AAA-Level Motion Capture
With the model in place, the next step is to bring the character to life. Traditional motion capture technology is almost exclusively the domain of Hollywood and AAA games. It typically requires renting a high-ceilinged studio, setting up over 100 optical lenses, and having the actors wear bodysuits covered in reflective beads. The cost can be as high as $5000 per day.
Kapnetix co-founder Johny Darkwah's goal is straightforward: "We want to bring motion capture costs close to zero."
The core technology showcased by Kapnetix lies in using AI combined with physics calculations to achieve "single-camera 3D skeleton recognition." This means that developers don't even need professional cameras; they can simply use an iPhone to film real-life movements, upload the footage to Kapnetix's cloud platform, and the AI can calculate the corresponding 3D skeleton data (FBX format) within minutes.
Johny Darkwah shared an interesting experience of failure: Initially, they trained the AI purely with massive amounts of data, resulting in animations full of jitter and slippage, which was clearly unacceptable to professional animators. Later, they changed their strategy, no longer relying on brute force with big data, but instead introducing "physical constraints" and "ergonomic logic."
For example, when the camera captures an arm that is obscured by the body (occlusion), pure AI might guess the position of the arm and cause clipping; but Kapnetix's algorithm incorporates the physical constraint that "the elbow cannot bend in the opposite direction," allowing the AI to calculate a reasonable motion path.
In the case study presented at the event, Johny Darkwah pointed out that what originally required 5 days of filming and post-production of NBA players' dribbling moves could be processed in just 10 minutes using Kapnetix's mobile phone filming solution. The smoothness of the movements was such that they could be directly applied to Maya or Blender for fine-tuning, and the cost was reduced from thousands of dollars to the price of a cup of coffee (about 2 euros/30 seconds).
Analysis: AI is reshaping the "middle layer" of the 3D industry.
The sharing from AssetHub and Kapnetix shows that the current 3D industry is undergoing a "decentralized" revolution.
In the past, the production of 3D content had extremely high barriers to entry, making it affordable only for large companies. But now, AssetHub has solved the problem of rapid construction and structuring of "static assets," while Kapnetix has solved the problem of the cost of acquiring "dynamic performances."
What these two companies have in common is that they are not trying to replace artists with AI, but rather to replace cumbersome workflows—including tedious topology work, key cell repair, and noise cleanup.
When a mobile phone can capture motion and a few images can generate a usable 3D model, future game development or VTuber content production will no longer be limited by technology and budget, but will return to the most core essence: creativity and storytelling. This may be the greatest contribution of AI to the content industry.
