Media Generation

The desktop app includes a media generation service that lets agents create images, videos, and audio using multiple AI providers.

This page covers media the agent creates for you. To browse existing self-hosted libraries or stock catalogs, see Cloud Storage.

Image Generation

Supported Providers

How It Works

Image generation is instant -- the agent requests an image and receives it when ready. The agent can:

1. Describe the desired image in natural language
2. Specify aspect ratio, style, and mood
3. Receive the generated image directly
4. Save it to the workspace or Library

Using Image Generation

Images can be generated in two ways:

• Through the agent: The agent uses media generation tools during task execution
• Directly: Use the image generation feature in the web platform

Video Generation

Supported Providers

New BytePlus setups default to Seedance 2.0 Fast for quicker turnaround at 720p. Existing Seedance 1.5 and 1.0 models remain available in Settings > Providers > BytePlus if you prefer them.

DesignMode may show HyperFrames as a video option for HTML/Motion workflows, but it is marked unsupported until the renderer is available for that project.

How It Works

Video generation runs in the background because it takes longer:

1. The agent submits a video generation request
2. You're notified immediately that it's being processed
3. The system checks for completion automatically (up to 1 hour)
4. When ready, the video is saved and available in the Library

Reference Images

Supply a single reference image to preserve identity and style, or supply both a first-frame and a last-frame image to have Seedance 2.0 animate the transition between them. Providing only a last-frame image is not supported and will be ignored with a warning.

Direct DesignMode Media Projects

In DesignMode, choose Media, then pick Image, Video, or Audio. The media form shows the available model groups for that surface, along with aspect ratio, duration, references, and prompt controls.

For audio projects, the current direct picker supports Speech and Voiceover. Music, Sound effect, and Ambience appear locked until the configured provider supports those output kinds.

Provider Configuration

Configure media providers in Settings > General:

• Enter the API key for your preferred provider
• The app automatically detects which providers are available
• Tools are registered based on configured providers

You can use multiple providers simultaneously, and switching providers requires no other changes.

Image edits that use a reference image can take longer than simple generation. The app now waits longer for OpenAI image requests and keeps reference-image fetching bounded so failed sources do not block indefinitely.

Graceful Degradation

If no media provider is configured:

• Media generation tools are not shown to the agent
• The agent can still perform all other tasks
• No errors are thrown -- the feature is simply unavailable

Voice I/O

The desktop app also supports voice interaction:

Text-to-Speech (TTS)

Hear agent responses spoken aloud:

• Sentence-by-sentence streaming for natural pacing
• Multiple providers: ElevenLabs, Azure, Google Cloud, OpenAI, MiniMax, and local models
• Works offline with bundled local models
• MiniMax supports a language boost setting for clearer multilingual pronunciation

Speech-to-Text (STT)

Dictate your prompts:

• Real-time streaming transcription
• Multiple providers: Azure, Google Cloud, OpenAI Whisper, and local models
• Automatic language detection

Local Speech Models

For offline use, the app includes built-in speech models:

• No API key required
• Works without internet
• Bundled with the desktop app

Learn More

• Agent System -- How agents use media tools
• Cloud Storage -- Browse and attach existing media from connected libraries
• Publish Pipeline -- Publish generated media to configured destinations
• Voice and Speech -- Transcription and text-to-speech workflows
• Desktop Application -- Overview and setup