Media (image, audio & transcription)

Besides text completions and embeddings, Rig can generate images, generate speech from text, and transcribe audio. Each is a model you get from a provider client, just like a completion model, and you call it with a request:

Task	Client method	Request builder	Response field
Image generation	client.image_generation(model)	ImageGenerationRequestBuilder	image: Vec<u8>
Audio generation (text-to-speech)	client.audio_generation(model)	AudioGenerationRequestBuilder	audio: Vec<u8>
Transcription (speech-to-text)	client.transcription(model)	TranscriptionRequestBuilder	text: String

Every response also carries usage, the provider name, the model the provider reported, and the raw provider response for anything Rig doesn’t normalize.

Before you start

• Know how to build a provider client; see Providers & Clients.
• Image and audio generation sit behind cargo features: rig = { version = "0.44.0", features = ["image", "audio"] }. Transcription needs no feature.
• Set your provider API key (e.g. OPENAI_API_KEY); the examples use from_env().

Image generation

Requires the image feature.

use rig::image_generation::ImageGenerationRequestBuilder;
use rig::providers::openai::{self, OpenAI};

let model = OpenAI::from_env()?.image_generation(openai::GPT_IMAGE_1);

let response = model
    .call(
        ImageGenerationRequestBuilder::new("A futuristic city at sunset")
            .width(1024)
            .height(1024)
            .build(),
    )
    .await?;

std::fs::write("city.png", &response.image)?;

response.image holds the decoded image bytes. Provider-specific options (quality, style, background, …) go in .additional_params(json!({ ... })).

The same code works with Gemini’s image models:

use rig::providers::gemini::{self, Gemini};

let model = Gemini::from_env()?.image_generation(gemini::GEMINI_2_5_FLASH_IMAGE);
let response = model
    .call(ImageGenerationRequestBuilder::new("A flat icon of a banana").build())
    .await?;

Audio generation (text-to-speech)

Requires the audio feature. The builder takes the text and a voice name:

use rig::audio_generation::AudioGenerationRequestBuilder;
use rig::providers::openai::{self, OpenAI};

let model = OpenAI::from_env()?.audio_generation(openai::TTS_1);

let response = model
    .call(AudioGenerationRequestBuilder::new("Hello, how can I help you today?", "alloy").build())
    .await?;

std::fs::write("hello.mp3", &response.audio)?;

Use .speed(1.25) to change the speaking rate.

Transcription (speech-to-text)

Build a request from a file (its name tells the provider the audio format) or from bytes:

use rig::providers::openai::{self, OpenAI};
use rig::transcription::TranscriptionRequestBuilder;

let model = OpenAI::from_env()?.transcription(openai::WHISPER_1);

let response = model
    .call(
        TranscriptionRequestBuilder::from_file("audio.mp3")?
            .language("en".to_string())
            .build(),
    )
    .await?;

println!("Transcription: {}", response.text);

Transcription: Hey, just calling to confirm our meeting tomorrow at ten. Talk soon.

For audio already in memory, use TranscriptionRequestBuilder::new(bytes).filename(Some("clip.wav".to_string())). You can also pass a prompt to guide spelling of names and jargon, and a temperature.

Provider support

Provider	Image generation	Audio generation	Transcription
OpenAI	Yes (gpt-image-*, DALL·E)	Yes (tts-1, tts-1-hd)	Yes (Whisper)
Gemini	Yes	No	Yes
OpenAI-compatible vendors (Azure, Groq, Hugging Face, Mistral, …)	Where the vendor serves it	Where the vendor serves it	Where the vendor serves it

OpenAI-compatible vendors share the OpenAI client type, so they expose the same methods; a call fails with a provider error if the vendor has no such endpoint. Check the provider pages for model names.

Media as input

To send images, audio, video, or documents to a model rather than generate them, add them to a user message with the UserContent constructors. See Sending files to the model.

See also

• Completions: text generation
• Providers & Clients: how clients create models

API Reference (Image Generation)

API Reference (Audio Generation)

API Reference (Transcription)

Next steps

Providers & ClientsSee how the client that builds these media models also creates completion and embedding models.

Model ProvidersCheck which providers support image, audio, and transcription models, and their model names.

LoadersSend documents and images to a model as part of a prompt.

ToolsWrap a media model in a tool so an agent can decide when to generate or transcribe.