Prepare a Jira work item for the Rovo agent

To get a good video from the Storyboard AI agent in one pass, write the Jira work item's description as a scene list.

The agent reads the work item before it does anything else — the description, the title, and the list of attachments. A few minutes spent on the description is the difference between one clean pass and a round of "no, that image belongs on scene 3".

None of this is required. The agent works fine on a rough ticket. This page is for when you want the first draft to be the final one.

The short version

One attachment per scene, numbered in the filename.
Say which file goes with which scene if the order isn't obvious.
Write the narration you care about verbatim; leave the rest blank on purpose.
State the voice, template, and caption style once, in the description.
Give every scene a distinct title.
Keep it to 20 scenes or fewer.

Getting the attachments in the right order

This is the single most common thing to get wrong, because the agent never guesses from the content of an image — it won't look at a screenshot and decide it belongs to the intro. It uses one of three rules, in this order.

An explicit mapping in the description — always wins. Write a line per scene:

Frame 1: Wireframe - 3.png
Frame 2: Wireframe - 1.png
Frame 3: dashboard-empty.png

Scene 1: ... works too, as does Frame 1 — attachment: .... The filename has to match the attachment exactly (capitalization doesn't matter), and it has to be unambiguous — if two attachments share a name, the agent stops and asks rather than picking one.

Map either all of the scenes or none of them. A description that maps scenes 1 and 2 but goes quiet on 3 and 4 makes the agent stop and ask you, which costs you a round trip.

Numbers in the filenames. With no mapping in the description, the agent sorts by a number in each filename:

Filename	Scene
`Wireframe - 1.png`	1
`Frame 2.png`	2
`Scene 5.png`	5
`Slide 3.png`	3
`01_intro.png`	1
`3-header.png`	3

Either a wireframe / frame / scene / slide keyword followed by a number, or a number at the very start of the name.

This rule is all-or-nothing. Every single attachment needs a parseable number. Add one dashboard.png to a set of nine numbered wireframes and the agent abandons filename order for the whole set and falls back to upload time.

Jira's duplicate suffix is ignored, which is what you want. Upload a second file called Wireframe - 3.png and Jira stores it as Wireframe - 3 (4).png. The agent reads that as scene 3, not scene 4 — the (4) is Jira's bookkeeping, not your numbering.

Upload order. The fallback: oldest attachment first. Reliable only if you uploaded the files one at a time in the order you want. Dragging a folder in at once gives you no control over this, so don't lean on it.

If you're unsure which rule the agent used, ask it to list the attachments — it reports which one it applied.

Narration: you get exactly what you write

The agent will not paraphrase your description into a script, and it will not invent narration for a scene you didn't write one for. That's deliberate — it's far easier to add a line in the editor than to notice and strip out a sentence you never approved.

You have three options per scene, and you can mix them freely in one description:

Write the script out and it's used verbatim. Nothing is rewritten, and no AI writing step runs for that scene.

Frame 2: Filters
Script: "Narrow the list by owner, status, or date — the counts update as you go."

Give a one-line brief and the agent writes that scene. Use this when you know what the scene must cover but don't want to draft it.

Frame 3: Cost savings
Cover: why caching cuts the S3 bill, keep it under 15 seconds

Say nothing and the scene stays silent. It holds on screen for its default duration with no voiceover. This is a valid, intentional outcome — not a gap the agent will helpfully fill. If you actually wanted narration there, give it a brief.

If you'd rather the agent draft the whole thing, just say so ("write the scripts") and skip the per-scene detail entirely.

State the settings once, up front

If the description names a voice, template, or caption style, the agent applies them while it builds. Mentioning them afterwards in chat means the narration audio gets thrown away and regenerated — an extra step and an extra confirmation for the same result.

Voice: Sarah (Confident)
Template: Product Walkthrough
Captions: TikTok

Use the voice names exactly as they appear — George (Warm Storyteller), Sarah (Confident), Alice (Clear Educator), Liam (Energetic), Matilda (Professional), Roger (Laid-Back). "A female voice" also works; the agent picks one and tells you which. Caption styles are TikTok, Subtitle, Karaoke, Minimal, or none.

Give every scene a different title

Scene titles are how the agent pairs an attachment to a scene, and how you refer to a scene later ("redo the narration on Filters"). Two scenes called "Overview" and it can't tell them apart — attachment mapping fails on both. Distinct titles cost nothing.

Size and count limits

20 scenes is the most the agent can create in one go. Beyond that, build the first 20 and add the rest in the editor, which has no limit.
Five scenes or fewer is the sweet spot: the agent can do the whole job — build, attach, narrate, render — with a single confirmation from you. Six or more splits into two.
Video clips must be 1920×1080 or smaller. The agent can't resize, so it refuses an oversized clip and tells you which scene is missing its asset. Adding that clip from the editor instead resizes it automatically.

What the agent can't see

Files you drop into the Rovo chat window. They go to the chat, not to the work item, and the agent's tools can't read them. Drag the file onto the work item's Attachments section instead.

Large files you added in the editor. The editor copies an upload to the work item's Attachments only if it's under roughly 4 MB. A bigger file — most video clips — stays in the storyboard, where the agent can't see it. If you want the agent to work with a clip, put it on the work item's Attachments section directly.

Anything on another work item. The agent works on one work item at a time, and a storyboard belongs to that item.

Zoom targets and per-scene caption styling. Editor-only, by design. Let the agent build the storyboard, then open the editor to add emphasis.

A worked example

Demo video for the new filter panel.

Voice: Alice (Clear Educator)
Template: Product Walkthrough
Captions: Subtitle

Frame 1: Overview
  Attachment: 01-overview.png
  Script: "The filter panel now lives beside your list, not behind a menu."

Frame 2: Filters
  Attachment: 02-filters.png
  Cover: how the three filter types combine, mention the live counts

Frame 3: Empty state
  Attachment: 03-empty.png
  (no narration — silent scene)

Three scenes, so it runs as a single confirmation: scenes created, the right image on each, Alice narrating scenes 1 and 2, scene 3 silent, rendered. Then open the Storyboard AI panel on the work item to watch it.

Last updated: October 05, 2026