Image To Audio

Overview

The Image To Audio node accepts an image and produces a spoken audio description by passing the image through a multimodal model that generates a caption or narrative, then synthesizing that text to speech. Use it in battle workflows where visual content must be accessible via audio, or as a pre-processing step before audio-based judging. A valid multimodal model credential and a TTS voice selection are required; if the model call or synthesis fails the node emits on the error port rather than halting the workflow. Output audio is returned as a base64-encoded buffer with a configurable MIME type.

Configuration

Field	Type	Required	Description
model	enum	Yes	Multimodal model used to generate the image description (e.g. gpt-4o, gemini-1.5-pro, claude-3-5-sonnet).
ttsVoice	enum	Yes	Text-to-speech voice used to synthesize the generated description (e.g. alloy, nova, echo).
descriptionPrompt	string	No	Custom system prompt appended to the model request to steer the style or focus of the image description. Defaults to a neutral accessibility-style caption prompt.
outputFormat	enum	No	Audio encoding format for the synthesized output: mp3, opus, or wav. Defaults to mp3.
maxDescriptionTokens	number	No	Maximum tokens allowed for the generated description before TTS synthesis. Limits cost and audio length. Defaults to 300.

Inputs

Port	Type	Description
image	string or binary buffer	The source image. Accepts a URL, base64-encoded data URI, or raw binary buffer. Must be a supported format (JPEG, PNG, WebP, GIF).
promptOverride	string	Optional runtime override for the description prompt, taking precedence over the static descriptionPrompt config field.

Outputs

Port	Type	Description
audio	audio object	Emitted on success. Includes base64-encoded audio (data), MIME type (mimeType, e.g. audio/mpeg), and the model-generated caption (description) before synthesis.
error	error object	Emitted when the model call or TTS synthesis fails. Includes message, code, and stage (description or synthesis) so downstream nodes can branch on the failing step.

Example

{
  "nodeType": "image_to_audio",
  "config": {
    "model": "gpt-4o",
    "ttsVoice": "nova",
    "descriptionPrompt": "Describe this image concisely for a visually impaired listener. Focus on subject, composition, and mood.",
    "outputFormat": "mp3",
    "maxDescriptionTokens": 250
  }
}