AI voices in a Steam game: what to disclose and which voices you may ship
Steam accepts games with AI voice acting. Since January 2024 Valve asks developers to describe their use of AI in the Content Survey they fill in before release, and shows much of that disclosure on the store page. What it does not accept is content that is illegal or infringes someone’s rights, and with voices that question is usually about whose voice it is.
This guide explains the two kinds of AI content Valve distinguishes, what each means for voice lines, and how to choose voices you can defend, with the free SuperClonica engine as one way to make them.
Pre-generated or live-generated
Valve splits AI content into two kinds. Pre-generated content is made with AI tools during development, such as voice lines you generate, review and pack into the build. Live-generated content is created while the game runs, such as an NPC that voices text the player typed or a chat model wrote. Both must stay legal and free of infringement; for live-generated content you must also tell Valve what guardrails stop the AI from producing illegal content, and players can report such content from inside the game.
What to write in the disclosure
Say plainly what is AI and how it was made: for example, “Dialogue of the NPCs is voiced with AI text to speech from our own script; voices are stock voices of an openly licensed model and a voice trained from our actor’s recording with her written consent.” Players read this on the store page, so a clear sentence works better than a vague one. If lines are voiced at runtime, describe the filters on what text can reach the voice.
Voices you can ship
The safe choices are voices you have the rights to. The ten stock voices of Supertonic 3, M1 to M5 and F1 to F5, may be used commercially under the OpenRAIL-M model license. A voice trained from the recording of an actor who agreed in writing is yours to use under that agreement; our consent form guide lists what it should cover. Do not ship a clone of a real actor, a celebrity or a character from someone else’s franchise without their permission: on SuperClonica such character voices are for fan and parody use only.
Making the lines
Install the engine with python -m pip install superclonica-cpu and voice your script in batches: superclonica --text-file line.txt --voice M2 --lang en --normalize --output line.wav. Everything runs offline on an ordinary processor, so the whole team and the build server can regenerate lines whenever the script changes. The same text, voice and seed give the same audio, which keeps builds reproducible.