AI NPC voice generator
AI NPC voice generator for lines, barks and full dialogue
Write the script, pick a voice for each character, direct the delivery with tags like [whispering] or [urgent], and get audio you can drop straight into your dialogue system.
The short answer
Summer generates NPC voice lines and multi-voice conversations from text using ElevenLabs voices. You choose a voice per character, add delivery tags for emotion, and export MP3 or uncompressed audio. In a Summer project the agent can also import the lines and wire them to your dialogue or trigger system.
Voicing even a small game used to mean casting, recording sessions and a lot of file wrangling. Placeholder text-to-speech was fast but flat, so most indie games shipped with text boxes and no voices at all.
Modern voice models change the math. With Summer you can voice a whole cast of NPCs, iterate on a line until the delivery is right, and hear it in the game minutes later. Shopkeepers, guards, quest givers, enemy barks and radio chatter all come from the same panel.
Because Summer is also a game engine, the job does not end at an audio file. Ask the agent to attach each line to the right dialogue node or trigger and it writes the code and places the audio in your project.
What it does
Single lines and barks
Generate one-off lines such as greetings, combat callouts and pickup reactions. The length follows the text you write.
Multi-character dialogue
Write a conversation line by line, assign a different voice to each speaker, and get one scene of dialogue back.
Direct the delivery
Bracketed tags such as [excited], [urgent] or [whispering] steer the emotion of a line, and stability, style and speed controls fine-tune the read.
A library of voices
Pick from narration, forceful, soft and character voices, and keep the same voice for a character across every line.
Clean formats
Export MP3 at 128 or 192 kbps, or uncompressed PCM at 44.1 kHz for editing and mastering.
Wired into the game
The Summer agent imports lines into your project and connects them to dialogue nodes, triggers or subtitles.
How it works
- 1
Step 1: Write the script
Paste the lines for one character, or a full conversation with the speaker named on each line.
- 2
Step 2: Cast the voices
Choose a voice per character. Keep a short note of which voice belongs to whom so every future line matches.
- 3
Step 3: Direct and regenerate
Add delivery tags, adjust punctuation for pacing, and regenerate single lines until they sound right.
- 4
Step 4: Drop into the game
Download the audio or let the agent import it and hook it up to your dialogue system and subtitles.
Two ways to run it
In Studio, or from the agent you already use
In Summer Studio
Open the Audio tab in Studio and choose Voice for single lines or Dialogue for multi-speaker scenes. Pick voices, set the output format, and generate. Everything lands in your asset library for reuse.
Open the Audio tab in StudioFrom Claude, Cursor or Codex via MCP
Through the Summer MCP your agent calls summer_generate_audio with the text_to_speech or text_to_dialogue capability, then imports the files and writes the dialogue code. It can batch a whole quest script in one request.
Example prompt for your agent
Voice every line in dialogue/blacksmith.json with a gruff older male voice using Summer, save the audio under res://audio/voice/blacksmith/, and play the matching clip when each dialogue line is shown.Set up the Summer MCP
Pricing
What it costs
- Voice line, about 5 seconds
- $0.01
- Speech or dialogue, per minute of audio
- $0.14
Voice is charged by the length of the audio produced, so a short bark costs a fraction of a long monologue.
Prices are charged from your Summer usage balance. Generating assets needs a paid plan or a top-up: Pro is $20 a month and includes $20 of usage, and top-ups start at $10. The desktop engine itself is free to download.
See all plansExamples
Prompts that work, and what comes back
A few character archetypes and how to prompt them. Delivery tags go at the start of the line.
Village blacksmith
Prompt
[gruff] Another sword? Fine. Leave it on the anvil and come back at sundown.
What you get
A single line in a low, weathered voice, ready for a shop dialogue.
Guard bark
Prompt
[urgent] Intruder at the east gate! Sound the alarm!
What you get
A short, shouted callout for an alert state. Generate three variants so repeats do not grate.
Quest giver
Prompt
[warm] You found it. I did not think anyone would come back from the marsh.
What you get
A softer delivery for a story beat, paired with subtitles.
Two-character scene
Prompt
Captain: [calm] Hold the line. / Scout: [breathless] They are coming over the ridge!
What you get
One dialogue clip with two distinct voices, timed as a conversation.
Radio chatter
Prompt
[flat, clipped] Unit four, proceed to checkpoint bravo. Over.
What you get
A neutral line you can run through a radio filter in the engine.
Narrator
Prompt
[slow, reverent] Long before the kingdoms, there was only the forest.
What you get
A narration line for an intro or loading screen.
Consistency across hundreds of lines
The main risk with generated voices is drift, where a character sounds slightly different in every scene. Keep one voice per character, write lines in the same register, and regenerate only the lines that stand out. With the agent, keep a small cast file in your project that maps each character to a voice so every new line uses the right one.
Steam and AI voices
Steam asks developers to disclose AI-generated content in the content survey. Its category for pre-generated content covers "any kind of content that ships with your game and is consumed by players that is created with the help of AI tools during development", which includes generated voice lines. Valve reviews that content against the same rules as any other content.
If your game generates voices live while it runs, Steam treats that as live-generated content and asks what guardrails you put in place. Summer's generator is built for pre-generated lines that you review before you ship.
Voices of real people
Do not use a generated voice to imitate a real actor or public figure without permission. Use the library voices, or voices you have the rights to, for anything you ship.
Frequently asked questions
Can I make different voices for each NPC?
Yes. Pick a voice per character. In dialogue mode each line carries its own voice, so a whole conversation comes back in one clip.
Can I control emotion and delivery?
Yes. Put delivery tags such as [excited], [urgent] or [whispering] at the start of a line, and use punctuation to control pacing.
What formats can I export?
MP3 at 44.1 kHz and 128 or 192 kbps, or uncompressed PCM at 44.1 kHz. Both import directly into Summer and other engines.
How long can a line be?
The length follows the text. Split long monologues into several lines, which keeps the delivery consistent and makes them easier to subtitle.
Can I use the voices in a commercial game?
Summer's terms state that you retain ownership of the content you create with Summer Engine. Disclose AI-generated audio where a store requires it, such as Steam's content survey.
Can it voice a game made in Unity or Godot?
Yes. Download the audio and import it into any engine. The automatic wiring into a dialogue system is specific to Summer projects.
Is there a free option?
The Summer engine is free. Voice generation needs a paid plan or a top-up and is charged by the length of the audio produced.
Make your game. Grow your game.
Even if it is built in another engine.
Free to download for macOS (Apple silicon) and Windows.