What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Game developers use generative AI at three separate points in character and prop work: generating concept art and sample 3D assets, assisting with animation, and running characters that speak and react during play. Each job uses a different kind of tool. Asset generators produce visual material that artists then refine, animation tools generate or drive motion, and runtime character systems govern dialogue and behavior once the game is running. The published evidence documents these uses and survey-reported adoption. It does not show that AI produces finished, production-ready characters or props without artist direction and review.
Three jobs that get lumped together
Most confusion about “AI-made characters” comes from treating three different workflows as one. A character’s appearance, the way its body moves, and the way it talks and makes decisions in a live game are handled by different tools, often by different teams.
| Workflow stage | What the tool produces | Example described in the sources | Source |
|---|---|---|---|
| Concept and asset generation | Concept images, sample characters, props, and landscapes | Scenario, as described in an AWS customer example | AWS 2025 guide |
| Animation assistance | Base animation sets adapted to a character’s style; facial blendshapes driven by audio | Base animation generation; Audio2Face-3D | AWS 2025 guide; NVIDIA ACE for Games |
| Runtime character behavior | Speech, decision-making, and animation during play | NVIDIA ACE for Games examples in shipped or announced titles | NVIDIA ACE for Games |
The runtime row does not create a character’s model. Nothing in the reviewed NVIDIA documentation indicates that ACE generates character meshes or props.
Concept art and asset generation
Concept exploration is the most established use. In its 2024 gaming report, Unity says survey respondents used AI mainly for rapid prototyping, concepting, asset creation, and worldbuilding (Unity Gaming Report 2024). That describes what respondents said they did, not how well the outputs performed.
#1 Best Overall
Generating characters and props from a workspace or inside a game
The AWS 2025 guide describes Scenario, an API-first generative tool, as producing characters, props, and landscapes. Teams can generate these from a shared workspace or through an integration inside a game (AWS 2025 guide). The guide is written by the cloud provider, so treat its description as vendor material.
Two executives quoted in the guide explain why studios choose this route. Hervé Nivon, Scenario Co-Founder & CTO, says the company “has served and generated millions of images with only three people.” Wang Yu, CEO of iFUN.COM GCR, says generative AI on the cloud “allows us to quickly obtain the materials we need and does not require us to operate and maintain AI-related infrastructure ourselves.” Both statements are the executives’ own accounts in vendor case material. They are not independent measurements of labor savings.
Rank #2
Trade-offs of cloud-hosted generation
Cloud generation removes the need to run model infrastructure, but it adds a dependency on the vendor’s service, pricing, and terms. The sources do not state per-asset costs, and they do not settle who owns generated outputs or what training data sits behind a given model. Those questions belong in your legal and procurement review, not in an assumption that a generated prop is cleared for use.
Animation: from base motion to facial performance
Base animation sets
The AWS guide lists generating base animation sets and adapting them to a character’s style as possible uses (AWS 2025 guide). This is a described workflow. The guide does not provide evidence of finished animation quality, so a studio would still need animators to judge the motion.
Facial animation driven by dialogue
NVIDIA’s Audio2Face-3D converts streaming audio into facial blendshapes and is documented for Unreal Engine and Maya workflows (NVIDIA ACE for Games). It is useful for bringing a voiced line to an animated face. It does not generate the character’s underlying appearance or props.
Characters that talk and react at runtime
NVIDIA’s ACE for Games provides cloud and on-device models for speech, intelligence, and animation, along with Unreal Engine plugins and integration SDKs (NVIDIA ACE for Games). NVIDIA names several examples: PUBG Co-Player Characters, inZOI Smart Zois, MIR5 bosses, and a Total War: PHARAOH advisor. The documented patterns include natural-language AI teammates, adaptive enemies, and character agents.
These are NVIDIA’s own descriptions of in-game interaction and behavior, not independent evaluations. NVIDIA’s developer page lists plugin versions and model access that can change, so confirm current details there before planning a project around them.
What the survey numbers show, and what they do not
Survey figures help show adoption, but each comes from a different sample and measures a different thing. Read them individually:
Best Value
- Unity 2024: 62% of surveyed studios said they used AI in their workflows, and 63% of surveyed AI adopters used generative technology for asset creation (Unity Gaming Report 2024). The second figure is a share of adopters, not of all developers.
- Unity 2025: 79% of developers polled said they felt positive about using AI in gaming (Unity Gaming Report 2025). This describes the people who responded to that poll.
- Google, AI Meets The Games Industry: 36% of respondents were using AI for dynamic level design, animation and rigging, and dialogue writing (Google report PDF). The report groups these tasks together, so the figure should not be read as a separate rate for each one.
Do not combine these into one trend line. The three reports use different questions, populations, and years.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Cloud or local inference
Where a model runs changes what your team has to provision. NVIDIA describes its models as optimized for gaming hardware and documents an on-device inference path. Some of its models can run across GPU, NPU, and CPU hardware. Hardware requirements depend on the specific model and the project, so check the model’s own documentation before buying or specifying machines.
| Factor | Cloud inference | On-device inference |
|---|---|---|
| Where the model runs | Vendor-hosted service | Developer or player hardware |
| Local hardware needed | Not required for the model itself, per the AWS guide’s description | Depends on the model and project; NVIDIA documents GPU, NPU, and CPU options for some models |
| Per-use cost | Not stated in the sources | Not stated in the sources |
| Data leaves your environment | Yes, to the vendor’s service | Not stated in the sources for every product |
Questions to answer before adopting a tool
- Output type: Does the tool return 2D images, 3D assets, rigging or motion, text, or speech?
- Integration: Is it standalone, an engine plugin, an API, or a local SDK?
- Consistency and editability: Can you regenerate the same character with the same look, and can artists edit the result in your existing tools?
- Rights and provenance: Can you document the training data and the terms that govern generated outputs?
- Latency and compute cost: Does the workflow stay fast enough for iteration, and who pays for inference?
- Human review: Who checks each output against your art direction and technical limits before it reaches a build?
What the evidence does not settle
- No independent cross-vendor comparison of generated character or prop quality appears in the reviewed material.
- No source establishes per-asset or total production cost, or labor outcomes beyond vendor-reported accounts.
- Rights and provenance for generated characters and props are not resolved by the sources.
- Vendor referral or affiliate arrangements for the tools named here were not verified, so this article does not recommend any purchase path.
Where the evidence is strongest, it shows that AI is used for concepts, sample assets, base animation, facial motion from audio, and runtime character behavior. Where it is weakest, it leaves quality, cost, and ownership for each studio to test against its own pipeline.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteQuick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
