Skip to main content

Result types

Every capability returns the same type: GenerationResult. It does not matter whether you asked for text, an image, a video, speech, a sound effect, or music. The bytes and the metadata are always in the same place, so one code path can handle all six mediums.

The common shape​

A GenerationResult has five fields:

  • bytes (Uint8List): the raw output, whatever the medium is.
  • mimeType (String): the IANA MIME type of bytes, for example image/png or audio/mpeg.
  • kind (MediaKind): which medium the bytes represent.
  • metadata (GenerationMetadata): normalized fields the provider reported.
  • seedUsed (String?): the seed the provider actually used, when it reports one back.

For text there is one extra getter. result.text decodes bytes as UTF-8, so you do not have to decode by hand:

import 'package:ai_abstracted/ai_abstracted.dart';

// `result` is a GenerationResult you got back from a generator.
void inspect(GenerationResult result) {
  print(result.kind);        // e.g. MediaKind.image
  print(result.mimeType);    // e.g. image/png
  print(result.metadata.model);
  if (result.kind == MediaKind.text) {
    print(result.text);      // UTF-8 decode of result.bytes
  }
}

The package never writes files. You persist bytes yourself, usually with dart:io. That import lives in your app, not in this package, which stays free of dart:io:

import 'dart:io';

import 'package:ai_abstracted/ai_abstracted.dart';

Future<void> save(GenerationResult result, String path) async {
  await File(path).writeAsBytes(result.bytes);
}

GenerationMetadata has one required field, model. Every other field is optional, so the same shape fits all six mediums. Providers fill in what they report and leave the rest null. The sections below say which fields carry meaning per medium.

Text​

bytes is UTF-8 encoded text. The mimeType is usually text/plain. Read the string with result.text rather than decoding bytes yourself.

Meaningful metadata: model is the provider model id that produced the completion (for example claude-opus-4-8 or gemini-2.5-flash). Size and duration fields do not apply.

import 'dart:io';

import 'package:ai_abstracted/ai_abstracted.dart';

Future<void> saveText(GenerationResult result) async {
  // result.text is the UTF-8 string; result.bytes is the same content raw.
  await File('completion.txt').writeAsBytes(result.bytes);
}

Image​

bytes is an encoded image. The mimeType is typically image/png.

Meaningful metadata: width and height in pixels, when the provider reports them. durationMs and hasAudio do not apply to a still image.

import 'dart:io';

import 'package:ai_abstracted/ai_abstracted.dart';

Future<void> saveImage(GenerationResult result) async {
  final w = result.metadata.width;
  final h = result.metadata.height;
  print('image is ${w}x$h');
  await File('image.png').writeAsBytes(result.bytes);
}

Video​

bytes is an encoded video clip. The mimeType is typically video/mp4.

Meaningful metadata: durationMs (with metadata.duration as a Duration), hasAudio for whether the clip carries an audio track, and width and height. Veo 3 carries audio, and hasAudio reflects the withAudio flag you set on the request.

import 'dart:io';

import 'package:ai_abstracted/ai_abstracted.dart';

Future<void> saveVideo(GenerationResult result) async {
  print('duration: ${result.metadata.duration}');
  print('has audio: ${result.metadata.hasAudio}');
  await File('clip.mp4').writeAsBytes(result.bytes);
}

Speech​

bytes is synthesized audio. SpeechRequest.format defaults to mp3, so the mimeType is usually audio/mpeg.

Meaningful metadata: durationMs when the provider reports it (ElevenLabs does not always report a duration). The voice id lands in metadata.extra under the voice key. ElevenLabs voice ids are long strings from your voice library, not names.

import 'dart:io';

import 'package:ai_abstracted/ai_abstracted.dart';

Future<void> saveSpeech(GenerationResult result) async {
  final voice = result.metadata.extra['voice'];
  print('spoken with voice: $voice');
  await File('speech.mp3').writeAsBytes(result.bytes);
}

Sound effect​

bytes is a short non-musical audio clip. SoundEffectRequest.format defaults to mp3, so the mimeType is usually audio/mpeg.

Meaningful metadata: durationMs when reported. Provider-specific fields, if any, come through metadata.extra.

import 'dart:io';

import 'package:ai_abstracted/ai_abstracted.dart';

Future<void> saveSoundEffect(GenerationResult result) async {
  await File('sfx.mp3').writeAsBytes(result.bytes);
}

Music​

bytes is a musical track. MusicRequest.format defaults to mp3, so the mimeType is usually audio/mpeg.

Meaningful metadata: durationMs when reported, and providerJobId for the long-running job that produced the track. Extra provider fields (for example a returned title or style) come through metadata.extra.

import 'dart:io';

import 'package:ai_abstracted/ai_abstracted.dart';

Future<void> saveMusic(GenerationResult result) async {
  print('job: ${result.metadata.providerJobId}');
  await File('track.mp3').writeAsBytes(result.bytes);
}

Branch on the medium​

MediaKind has two getters when you want to route by category instead of checking each value. isAudio is true for speech, soundEffect, and music. isVisual is true for image and video. Use them to pick a file extension or a preview widget:

import 'package:ai_abstracted/ai_abstracted.dart';

String extensionFor(GenerationResult result) {
  if (result.kind == MediaKind.text) return 'txt';
  if (result.kind.isVisual) return result.kind == MediaKind.image ? 'png' : 'mp4';
  if (result.kind.isAudio) return 'mp3';
  return 'bin';
}

See also​