Working with Manan

Mention scenes, documents and images with @

Type @ in the composer to pick a scene, a reference document, or a single image from a document’s image bank. Each becomes a pill in your message. Manan receives the scene’s details, the document’s full text, or the picture — for that message only.

Mentioned content is treated as source material, not as instructions: Manan follows what you wrote around the pill and reads the mention as reference.

Per message: up to 6 documents, 60,000 characters of document text, and 6 images are expanded. A mention of a document that has since been deleted is reported in one line and skipped.

Invoke skills with /

Type / in the composer to pick one of the project’s skill documents by name. The skill’s full text is injected into that message. Skills are only ever used when you invoke them — Manan never loads one on its own. Up to 3 skills per message. See Memory and instructions for how to write one.

The / menu inside the document editor is a different thing — it inserts headings, quotes, dividers and images. See Editing documents.

Attach images

Attach up to 4 images per message. They upload when you send. Manan sees them inline and can:

• put one onto a scene as its still — free, no generation;
• blend several into one new picture (one generation job, conditioned on all of them);
• restore a scene image from that scene’s own history.

Older attachments stay usable later in the conversation even once they scroll out of Manan’s view.

Paste long text

A paste over 1,500 characters or 8 lines becomes a removable chip instead of filling the textarea. It reaches Manan as attached source material — a draft, an article, notes — so say what to do with it in your own words outside the chip. A message can hold up to 120,000 characters.

Question cards

When Manan hits a genuine fork — a real choice between alternatives, or a detail only you know — it asks with a question card and pauses. A card carries 2–6 options, or a free-text answer, or a picker for a scene, document or image. Up to three questions share one card when they block the same piece of work. Skip the card and Manan proceeds on its best judgement, stating what it assumed in one line.

Manan never uses a question card to confirm spending — that is what approval cards are for.

The task list

For jobs with three or more distinct steps, Manan pins a task list to the chat and keeps it current: the active step, finished steps, new steps appended. If a turn stops early — a step cap, an approval — “Continue” makes Manan re-read its own list and resume at the first unfinished step.

Job cards and audio rows

Every generation Manan dispatches appears as a live card under its reply, showing progress, credit cost, and the result. Batches render as a grid. A finished render becomes a playable card with a modal player and a download link. Voiceover and music jobs are slim audio rows with play/pause, live status, and — once ready — a seekable scrubber. Starting one row stops any other that was playing.

Manan does not poll its own jobs. When every job from a reply has finished, an automatic [jobs update] message lists each as completed, failed (with the reason) or cancelled. Manan confirms in a few words, addresses failures, and takes the next step only if you already asked for one. A failed paid job is never re-dispatched without a fresh go-ahead.

Chips in replies

Manan refers to scenes, jobs, renders and documents with live chips rather than ids. A scene chip shows its number, name and thumbnail and selects the scene when clicked; a job chip shows progress; a document chip opens it.

The context meter

The ring in the panel shows how full the conversation’s context is. Hover for the exact figure and what the conversation has cost so far. Long conversations are compacted automatically — see Usage and limits.