Skip to content

Everyday workflows

Separate AI latency from the pace of an explanation

Understand the difference between waiting for a model, preparing speech, drawing a step, and pausing to think. Use playback controls intentionally.

··3 min read

A gap between two visual steps can have several causes. The next idea may still be arriving from the model, speech may be preparing, an animation may be finishing, or the explanation may be waiting for your reply. Those are different kinds of waiting.

Waiting and pacing are differentIllustrative diagram
Separate waiting from pacingThis diagram shows stages, not measured durations. Some preparation can overlap; it is not a latency benchmark.1Screen2ModelWait for context3Voice4ExplanationⅡ ← 1×
  1. 01
    Read the source

    Fresh screen context must be available before a grounded answer.

  2. 02
    Prepare the next idea

    The provider generates the response and its next complete visual step.

  3. 03
    Prepare speech

    The chosen voice needs audio before synchronized playback can start.

  4. 04
    Control the explanation

    Pause, return to a step, or adjust pace once playback is underway.

This diagram shows stages, not measured durations. Some preparation can overlap; it is not a latency benchmark.

Notice where the wait happens

A pause before the first useful response is different from a pause after a question. The first can involve screen capture, a provider request, and speech preparation. The second may be intentional: the companion is waiting for you to answer rather than producing the next part. Looking at the current status helps avoid interrupting a useful pause as if it were a failed request.

For a visual explanation, a step also needs to be complete enough to present. Displaying half an instruction or an unfinished drawing can be worse than waiting briefly for a coherent beat. This does not make every delay necessary; it means that “the model is slow” is not a complete diagnosis for every gap you notice.

Use playback controls for understanding

Pause when you need to inspect a relationship that is already on screen. Return to a previous explanation step when the next idea depends on something you did not catch. Adjust the explanation speed if the material is familiar or if you need more time to connect words with the drawing. These are controls over presentation, not a way to accelerate the provider’s reasoning.

A useful explanation leaves room for those choices without making you restart the entire conversation. Rayito supports live explanation controls so you can follow the current idea at a workable pace. Use the visible controls when you are unsure which keyboard shortcut is active; typing in another app should not be confused with controlling playback.

Try a smaller request before changing every setting

If a response feels slow, compare it with a short question about the same visible source. Asking for one relationship is easier to evaluate than requesting a complete tour, a detailed diagram, several app actions, and a summary at once. This is not a requirement to micromanage the drawing. It is a way to isolate whether the size of the request is part of the delay.

Voice selection can also change preparation behavior. Local models may need an initial warm-up, while cloud services depend on a network request. A changed voice or a missing local package can therefore affect the time before speech starts. Keep the source and question similar while testing one setting; otherwise you cannot tell what caused the difference.

A question to try

“Explain only the first relationship, then let me respond.”

Report a delay with the stage that was visible

If you report a problem, describe whether it happened before the first answer, between narrated steps, after pressing Resume, or after answering a question. Mention the selected provider and voice, but do not include credentials or private screen content. That small amount of context is more useful than a general claim that everything is slow. It helps distinguish request latency, audio preparation, playback pacing, and a task that is intentionally waiting.