Yes, you can ship AI voices in an Unreal game without paying, but only with a few tools. The cleanest free option is an offline plugin that runs an open model inside your build, such as SpeechGen (Apache-2.0, Unreal 5.7, Windows only). Inworld's pay-as-you-go plan also includes a commercial license at $0 a month, while the free plans of ElevenLabs and Fish Audio do not let you sell what you make.
Prices and terms below were checked at each maker's own page on 2026-10-27. They change often, so follow the source link before you build on any line. This is a plain-English summary, not legal advice.
The quick answer
| Tool | How it reaches Unreal | Free to ship? | Main catch |
|---|---|---|---|
| SpeechGen | Unreal plugin, runs on the player's CPU | Yes (Apache-2.0) | Unreal 5.7 and Windows x64 only; one-developer project |
| Runtime Text To Speech (Georgy Dev) | Unreal plugin on Fab, offline | Plugin is paid; voices vary | Each voice model has its own dataset license |
| ReadSpeaker runtime TTS | Unreal plugin on Fab, on-device | No: free trial is watermarked | Publishing license on request; price not public |
| Convai | Official Unreal plugin, cloud | Free plan exists, for testing | "Powered by Convai" credit; every turn costs credits |
| Inworld AI | REST, WebSocket or WebRTC API | Yes, pay as you go | No Unreal plugin; you pay per character spoken |
| ElevenLabs | Export audio files, or call its API | No: Starter ($6/mo) and up | Free plan has no commercial license |
| Fish Audio | Export audio files, or call its API | No: paid plans only | Free output is personal, non-commercial use |
For more voice tools that aren't Unreal-specific, see AI voice and sound tools. For talking NPCs, see AI character and animation tools for Unreal.
First decide: pre-recorded lines or live speech?
This one choice decides which tools fit.
- Pre-recorded lines (a script you write before release): generate WAV files once, import them as Sound Waves, and ship them like any voice-over. Any text-to-speech service works. The only question is whether the plan you generated them on gave you a commercial license.
- Live speech (lines written at runtime, such as an LLM-driven NPC or reading player names): the voice has to be produced while the game runs. You need either an in-engine runtime that runs offline, or a cloud API that bills you for every line every player hears.
Offline runtimes cost nothing per player but add a model to your build and use the player's CPU. Cloud services sound better and do more, but the bill grows with your player count.
Which free options can you actually ship?
SpeechGen: free, offline, Unreal 5.7 only
SpeechGen is an open-source Unreal Engine 5.7 plugin that runs the Kokoro-82M voice model through Unreal's own neural network runtime, on the CPU, with no cloud service. It supports 48 voices across English, Spanish, Hindi, Italian, Brazilian Portuguese and Mandarin, with German, Dutch and Turkish marked experimental. Packaging places the model files beside your executable, so players never download anything. Source: SpeechGen README
The license is clean: the plugin is Apache-2.0, and the Kokoro model and voice weights are Apache-2.0 too. The notices file also lists training-data credits under CC BY 3.0 and CC BY 4.0, so keep the license files it asks you to ship. Source: SpeechGen third-party notices Kokoro's own model card confirms the Apache-2.0 weights. Source: Kokoro-82M on Hugging Face
The catches: it requires Unreal 5.7 and Windows x64, the editor needs internet once to download the runtime, and it is a very young project run by one developer. Prototype with it before you build a game around it.
Runtime Text To Speech (Georgy Dev): wider support, check each voice
This Fab plugin also runs offline, using Piper or Kokoro models, for Unreal 4.27 to 5.8 on Windows, Mac, Linux, Android, iOS and Meta Quest. It streams audio and lets you import your own Piper or Kokoro voice models. Source: Georgy Dev It is sold on Fab; we couldn't read the price this run, so check the listing.
The license catch sits in the voices, not the plugin. Each Piper voice has a model card naming its training data. Some use CC BY 4.0 data (for example the en_US libritts_r voice), while the en_US lessac voice was trained on a dataset released under a research licence. Source: Piper voices, libritts_r model card Source: Lessac Blizzard 2013 licence Pick voices whose data allows commercial use, or use the Kokoro voices.
ReadSpeaker: free to try, not free to ship
ReadSpeaker's on-device Unreal plugin has a free full trial with no time limit, but trial audio carries a watermark. When you're ready to publish, ReadSpeaker removes the watermark and gives you a lifetime license for that game. The price isn't listed. Source: ReadSpeaker on Fab
Inworld AI: commercial license on the $0 plan
Inworld's On-Demand plan costs $0 a month and is pay as you go: TTS-2 is $25 and TTS-2 Flash $15 per million characters, with up to 70 minutes of speech included. The pricing table lists a commercial license on every plan, including On-Demand. Creator is $25 a month. Source: Inworld pricing
The catch for Unreal devs: Inworld is API-first. You call text-to-speech, speech-to-text and speech-to-speech over REST, WebSocket or WebRTC from Unreal. It doesn't name a dedicated engine plugin. Source: Inworld for games Budget the per-character cost against how much your players will hear.
Which paid options are worth it?
Convai: the most Unreal-native, but it's a full NPC platform
Convai has an official Unreal plugin (now in a 4.0 beta) that adds conversation, voice, lip sync and actions to existing characters, with Win64 and Android build targets. Source: Convai Unreal plugin docs The pricing page shows a $0 Free plan "to test out the platform" with 100 interactions a month, and Indie Dev at $29 a month. Source: Convai pricing Its billing docs say plans are moving to credits: each spoken turn is charged for speech recognition, the language model, the voice and a platform fee, and an open session uses credits even when nobody speaks. Source: How Convai credits work
Read the terms before you ship. You own your content and the AI output, but a public release must show Convai's logo and a "Powered by Convai" tagline at the start, and on the Creator tier you grant Convai a license to use your content to improve its services. Source: Convai terms Comparing it with Inworld? See Convai vs Inworld AI.
ElevenLabs and Fish Audio: great for pre-recorded lines, if you pay
Both work fine for an Unreal game if you generate the lines and import the audio. Neither free plan lets you sell the result:
- ElevenLabs Free gives 10,000 credits a month without a commercial license. Starter ($6 a month) adds it. Source: ElevenLabs pricing
- Fish Audio says free users can only use generated content for personal, non-commercial projects; paid subscribers can use voices they own commercially. Source: Fish Audio plans
We compare the two in ElevenLabs vs Fish Audio.
How to choose
- All lines known before release, and budget is zero? Generate them with an offline Kokoro voice (SpeechGen in Unreal 5.7) and ship the WAVs, or pay one month of a cloud plan with a commercial license and generate everything then.
- Live NPC speech, no per-player bill? An offline plugin: SpeechGen on Windows and 5.7, or Georgy Dev's plugin for other platforms and versions (check each voice's license).
- Live speech, best quality, budget per player? Inworld for pay-as-you-go voices with a commercial license, or Convai if you want the whole NPC stack in an Unreal plugin and accept its credit line.
- Before you ship, whichever you pick:
- Save a screenshot or PDF of the pricing and terms page on the day you generated your lines.
- Keep the license files and credits the tool asks for.
- Don't clone a real person's voice without their written consent.
- Check what Steam asks you to disclose: see Steam's AI disclosure.
For other free tiers you can and can't ship with, see which free AI tools you can use in a commercial game.