Skip to content
Auteryn

Multi-Modal Content Generation

Auteryn agents can produce visual and structured outputs — not just text — when the relevant capabilities are enabled and the run uses Deep mode (for image/video tools).

What you can generate

Section titled “What you can generate”

Images

Text-to-image via the generate_image tool (Deep mode). The agent picks a quality tier — lite, standard, or pro (higher fidelity, up to 4K, plan-gated) — to fit each image.

Videos

Text-to-video with native audio via Gemini Omni Flash (gemini-omni-1.1-flash). Generate a clip, extend it, or edit one the agent already made. Plan-gated to Pro and up, with strict per-run and per-day limits.

Presentations

JSON slide decks via presentation builder tools — create, edit, export, and publish without a sandbox.

Diagrams & charts

D2/Python diagrams and UI artifacts (charts, tables, documents) emitted as artifacts in the workspace.

Image generation

Section titled “Image generation”
  • Available in Deep mode for Internal agents (and other types that receive Deep toolsets)
  • Quality tiers — the agent chooses a model per image to match the need: lite (backgrounds, decorative fills), standard (most content imagery), or pro (hero/brand images, accurate in-image text, up to 4K — Pro plan and up). Each tier has its own credit rate; see Pricing.
  • Configurable aspect ratios and output resolution; outputs upload to cloud storage (GCS) and appear as artifacts
  • Rate limits apply per run and per hour (the pro tier has a tighter per-run cap)

Example instruction:

Generate a hero image for our product launch: modern tech aesthetic,
blue gradient, abstract shapes. Use 16:9 aspect ratio.

Video generation

Section titled “Video generation”

Video uses Gemini Omni Flash (gemini-omni-1.1-flash). Every clip comes back as an MP4 with its own audio track — ambience, music, or spoken dialogue you asked for in the prompt.

Three tools work together:

Tool What it does
generate_video One 3–10 second beat from a prompt. The model decides the exact length; you choose the resolution (360p, 720p, 1080p, 4k) and the aspect ratio (16:9 or 9:16).
extend_video Continues a clip the agent just made, 10 seconds at a time, to a 40-second total.
edit_video Adds or removes an object, restyles, or relights — either a clip the agent generated or one you attached to the conversation.
  • Steer the look with up to 3 reference images, or supply a first and last frame and let the model interpolate the motion between them
  • Available on Pro and up, with limits (defaults: 2 videos/run, 5/day, 50/month per org — configurable), plus a daily cap on total generated seconds
  • The agent confirms the estimated cost with you before rendering, and progress streams live in the workspace while it renders
  • Every generated video is marked as AI-generated

Presentations

Section titled “Presentations”

Presentation builder tools work without a sandbox:

  • create_presentation, edit_presentation_slide, add_presentation_slide, etc.
  • Narrate, export (PDF / video), and publish to a shareable link

For the full lifecycle — generate, refine, narrate, export, and publish — see the Presentations deep dive.

Diagrams & artifacts

Section titled “Diagrams & artifacts”
  • D2 / Python diagrams — rendered in ephemeral diagram_renderer task containers
  • Artifacts — charts, tables, markdown documents, HTML via artifact tools (streamed to the workspace UI)

Audio

Section titled “Audio”

Agents can also generate audio — text-to-speech narration and voiceover, and multi-speaker podcast-style conversations. Outputs upload to cloud storage and appear as playable artifacts. Speech costs 50 credits per 1,000 characters in a preset or designed voice, or 60 in a cloned voice, per the pricing page.

For the full breakdown — including how audio differs from real-time Voice conversations — see the Audio generation deep dive.

Tips

Section titled “Tips”
  • Enable Deep mode (or Auto with SmartRouter selecting Deep) for image/video tasks
  • Be specific about style, aspect ratio, and brand constraints in instructions
  • Review generated media before publishing — models can misinterpret prompts
  • Video generation is slow and credit-intensive; iterate at 360p and only render the final shot at 720p or above
Section titled “Related”