Media Understanding

Anyy can process image, video, and audio inputs through compatible model or transcription routes. The input surface must deliver the attachment with the matching media kind; uploading a file alone does not establish model support.

Dashboard Attachments

In Dashboard → Chat, attach files before sending. One upload request accepts up to five files, each up to 50 MiB. Pending attachments expire after 24 hours if no message claims them.

Dashboard recognizes JPEG and PNG uploads as images. Other uploads, including audio and video files, are ordinary file attachments in v0.1.107; they do not automatically enter STT or video-understanding processing. Use a supported channel/input path that delivers audio, voice, or video attachments for those routes.

Give images an explicit instruction:

Read this screenshot and list the visible error messages in order.

Runtime Inputs

Medium	Supported route
Image	JPEG or PNG image attachment; native image input or a dedicated image-summary model
Video	MP4, WebM, or QuickTime video attachment delivered by a supporting input path; native video input or a dedicated summary model
Audio	audio/voice attachment delivered by a supporting input path; transcript-first STT

The provider may impose stricter formats, size limits, or duration limits. Audio processing defaults to 25 MiB, 60 seconds of request time, and 12,000 transcript characters; this is separate from the Dashboard upload limit. Some speech protocols/models accept less.

Configure Routes

Use Dashboard → Models to configure the main model and image/video/speech routes. The equivalent image/video Tools pages and CLI setup are also available:

anyy tools image understanding
anyy tools video understanding
anyy tools voice stt

The selected model must declare the input capability. An unknown manually entered model is not automatically treated as vision- or video-capable. See Configuring Models for authentication and routing.

Processing Modes

Mode	Image/video behavior
auto / native_first	Prefer the current model's declared native capability; a configured auxiliary route can produce a summary when needed
summary_first	Use a dedicated understanding route before the main conversation; that route is required
disabled	Disable the auxiliary understanding path; this is not a privacy switch that blocks native-capable models from receiving attached media

Image summaries default to 160 tokens/800 characters and a 60-second timeout. Summary limits can omit fine detail; ask for the evidence you need and verify important text or timestamps. When usable image bytes or a compatible route are unavailable, the model may receive an omitted/unavailable marker instead of image content. Image/video summary failures can also be represented as unavailable; do not assume that an attached file has been understood.

Audio uses transcript_first. media_understanding.audio.on_failure accepts annotate (default), ignore, or fail: respectively add a failed-transcription marker, continue without a transcript, or fail the request. Speaker identification and verbatim accuracy depend on the STT backend; they are not guaranteed by attachment support.

Privacy and Cost

Native media or auxiliary processing can send bytes to its configured provider. A later main-model request may receive a summary/transcript instead. Review every selected provider's retention and regional terms. Large media and transcription can increase latency and cost.

Troubleshooting

Problem	Check
Dashboard video/audio uploads as a file	This is the current Dashboard behavior; use an input path that delivers the matching media kind
Image is omitted or unavailable	Use valid JPEG/PNG bytes and a capable native or auxiliary route
Video is not understood	Check input attachment kind, MP4/WebM/QuickTime format, capability, and route
Audio has no transcript	Check audio/voice kind, STT enablement, protocol, and its smaller limits
Provider rejects size/context	Compress or trim input and check the selected provider's limits

See Voice & Speech for speech protocols.