Introducing the Kitsch Labs API

Introducing the Kitsch Labs API. You can now bring the AI voices you choose in the app into games, animation tools, AI character chat, and other products. Generate a voice for the moment a character first says hello, or for a line that changes with a player's choice.

The API lets you list available voices, turn text into speech, and receive a streaming-format response. Voice access and request limits are checked against your account. See the full request contract in the Kitsch Labs developer documentation. Here's how to create an audio file from your first line.

Find a voice and generate a line

Start with GET /v1/voices to see the voices available to your account. Choose a voice ID that suits your character, then send it and your line to POST /v1/text-to-speech/{voice_id} to receive audio.

Create a key in the app's API key page, then replace YOUR_API_KEY and VOICE_ID below with your values. Keep the API key on your server.

curl -X POST "https://api.kitschlabs.com/v1/text-to-speech/VOICE_ID?output_format=mp3_44100_128" \
  -H "xi-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text":"You made it. I was waiting for you.","language_code":"en"}' \
  --output character-line.mp3

This request saves an MP3 file. In a product, your server can call the API and deliver the resulting audio at the right point in a game dialogue or conversation.

Receive audio as a streaming response

To receive a chunked HTTP response, send the same request body to POST /v1/text-to-speech/{voice_id}/stream. Add /stream to the end of the generation URL; the supported output formats and authentication are the same.

The current streaming route generates the full utterance before sending audio chunks. It therefore does not guarantee a faster first sound than the regular request or play words as they are generated. Measure latency with your actual line lengths and network conditions before choosing a route. See the developer documentation for the request and response format.

Shape the delivery for each scene

The same words should sound different depending on the character's relationships and the moment. Use language_code to set a supported spoken language and instruct to give a natural-language direction for the delivery. You might keep a character calm in everyday dialogue, then adjust the direction for an excited reunion.

Listen to several lines before settling on a voice. A single sample is only the beginning; the character should still sound like the same person across consecutive game lines, animation cuts, and longer character-chat exchanges.

Before connecting a live product

Store API keys on your server, never in browser or client code. Available voices depend on your account's access, and requests are subject to usage and credit limits. Check your current plan and each voice's terms before commercial use.

Start with a short representative line and check audio quality, response time, and what people see when a request fails.

Give your character a first line

The Kitsch Labs API is a starting point for bringing a voice into a scene in your product. Check the current request format and supported features in the developer documentation, then create an API key and generate your first line.