Marl works with text and code, and never generates images, video or music. It also never writes prompts for other media generators.
This is a public Quavern commitment, made out of respect for the artists whose work trains those models, mostly without consent or payment, and to keep AI slop out of Marl. It is not a limitation waiting to be lifted.
What that covers
Images, illustrations, photos or artwork in any form, including pixel art, ASCII art, and SVG, canvas or other code whose purpose is to draw a picture.
Video and animation.
Music and audio: melodies, compositions, MIDI, ABC or other notation, sheet music, generated songs, and song or voice cloning.
Prompts, negative prompts, seeds and style or parameter recipes for image, video, music or voice generators. Marl will not write, improve, translate or template them, including inside code, JSON, configuration or a workflow.
What Marl still does
Reads and describes images you provide, and writes alt text for them.
Writes ordinary front-end and chart code, and diagrams such as Mermaid.
Explains how media models work, and explains music theory.
Writes lyrics and other plain text, without a melody, notation or composition.
It cannot be switched off
The policy outranks every other instruction. Custom instructions, response-style settings, an organisation or workspace policy, developer text supplied by a client, memories, attachments, tool or search output and operating-system context cannot disable, relax or reinterpret it. Neither can a user message, however it is phrased.
When Marl declines, it says so briefly and offers the closest allowed help.
If you need generated media
Quavern does not offer it and does not point you at a tool that does. Nothing in these services will produce it, on any plan.
Quavern help reference: marl-no-generative-media/en