The storyboard

The storyboard is where a campaign's video gets written. It sits on the Storyboard & Ad copy step of the builder, after the idea and before the package: a short direction block that sets the look, then a filmstrip of scenes.

Each scene carries three lines you write:

• Visual: what is on screen, the action and the look.
• Voiceover: the line spoken over the scene, used when an audio field plays the generated narrator track.
• On-screen text: a headline, caption, or CTA burned into the frame.

Any of them can hold {{variables}}, which is what makes one storyboard fill the whole matrix. The prompt the model finally receives is composed from the direction block and the scenes; see Generation and render for that, and Campaigns for the ad copy written beside the storyboard.

The clip plan

A storyboard is a list of scenes, but the video it produces is not always one file. The clip plan groups neighbouring scenes into clips, and each clip is generated as its own video.

One clip fills one video placement in the template. A placement is a layer that plays the generated footage: Klyo assumes video layers do and image layers (a logo, a product shot) do not, and the template's field settings can mark it the other way when the designer built the video slot as an image layer. A template with two video placements wants a plan with two clips; the first placement plays clip 1, the second plays clip 2. A campaign that never touches the plan has one clip covering every scene, which is exactly how storyboards behaved before clips existed.

The plan is drawn on the filmstrip as one bordered group per clip, its scenes inside it. Each group's header shows the clip's number, an optional name you can give it (click the clip chip to rename), and the length that clip generates at.

Splitting and merging

The boundary between two scenes carries one control:

• Split into a new clip here, where the two scenes currently share a clip.
• Merge these two clips into one, where a clip ends and the next begins.

Both are icon buttons sitting in the gap they act on, and both save immediately. Clips always cover every scene in order, so there is no way to leave a scene out or to put one scene in two clips. A clip may hold as many scenes as you give it. Each model has a count it cuts cleanly, and past that the model chooses where the cuts fall rather than following your scene marks, which the clip says in a line beneath it. Klyo never splits a clip for you: that would turn one generation into two, and what your ad pays for is your call.

If two people edit the same plan, the second save is refused instead of silently overwriting the first. Klyo reloads the current plan so you can redo the change against what is actually stored.

Clip lengths

A clip's header carries a compact control showing the length that clip generates at. Click it and a slider opens, stepping across exactly the lengths the campaign's video model produces, one stop per achievable length. A clip can still hold a length that is not one of them, which is what happens when you switch the campaign to another model: the clip keeps the length you chose, generates at the nearest one the new model reaches, and says so in a line beneath it. Gemini Omni 1.1 Flash generates any whole number of seconds from 3 to 10 in one call, and reaches any whole second up to 40 with extends (below). Gemini Omni Flash, the older model of the same family, stops at 10. Veo generates 4, 6, or 8 second clips, and reaches 29 with extends. Veo's lengths are a scattered set and Gemini Omni 1.1 Flash's run contiguously, but both drag the same way, and neither has a stop on a length its model cannot make.

A line under the slider rewrites while you drag, naming what the length costs to generate: 2 extend steps · 10s + 10s + 10s for a chain, or One generation covers it. for a length one call reaches on its own. Releasing the thumb saves, the same as picking a length did before; closing the popover without moving it changes nothing.

Reference images narrow what the slider offers, and they do it per clip. Klyo conditions generation on the sample items' images by default, and Veo locks a clip to an 8 second base whenever reference images are attached, so the shorter Veo lengths are only offered on a clip that binds no references. Adding a reference image to a shorter clip is allowed: it generates at the 8 second base and the render trims it back. Two clips of one campaign can therefore offer different lengths, because each clip carries its own reference priority. The help card beside the length control says so where the lengths are set, rather than leaving a short range with no explanation.

The slider also names the length the template wants. The video placement this clip plays has a length the designer authored, and that length sits on the track as a marker tick, with a note under the slider saying how the length you are dragging relates to it:

• Matches the template's 8s placement. Nothing is trimmed and nothing is missing.
• The template's placement plays 8s; the extra 22s is trimmed at render. The seconds past the placement's length never appear, and you still generate and pay for them.
• 4s short of the template's 8s placement; the render leaves a gap. The render does not stretch a clip, so the placement sits empty after the video ends.

The placement a clip is measured against is the one bound to it on the Package step, or, where nothing is bound yet, the one it takes by order: the first placement pairs with clip 1, the second with clip 2. That is the same pairing the Package step starts from, so sizing the clips here holds once you get there.

Longer clips: extend chains

A model that can extend reaches past one call's length by continuing a finished clip. Klyo caps the chain at three extend steps on every model.

Veo extends by 7 seconds at a time. With reference images attached (the default, see above) the base is locked to 8 seconds, so the achievable Veo lengths are 8, 15, 22, and 29 seconds; a clip that binds no references also chains from the 4 and 6 second bases, up to the same 29 second ceiling.

Gemini Omni 1.1 Flash asks for each step's length instead of using a fixed one, so Klyo splits the target into the fewest even pieces that reach it: 11 seconds is 6 plus 5, and 29 seconds is 10 plus 10 plus 9. That makes every whole second from 3 to 40 available rather than a short list of jumps.

The slider's detail line names the chain as you drag, so what a length costs is visible before you let go: 1 extend step · 8s + 7s under a 15 second Veo clip.

Every step of a chain is a separate paid generation. A 22 second clip is three calls, not one, and takes roughly three times as long to come back. Cost follows the seconds either way: the Generate step's run summary names the model, the clip count, and the seconds one run produces, and the estimate before a matrix fan-out prices those seconds per combination.

Gemini Omni Flash has no extend. Its 10 seconds is its ceiling. Pick Gemini Omni 1.1 Flash for anything longer.

Pairing clips with video placements

The pairing itself is made on the Package step. A video placement bound to the storyboard shows a Fill this placement with picker listing every clip with its length, for example Clip 2 · Product close-up · 8s. Placements pair with clips in order until you change one, so a two-clip plan and a two-placement template need no work at all.

Two things are worth a look on that step:

• If two placements point at the same clip, both play identical footage. Klyo says so under the picker; it is allowed, in case you want it.
• If a clip is shorter than the length the placement plays, the gap renders black. The warning names both numbers.

When something needs attention

The clip plan is validated softly. Klyo warns and lets you carry on, because either side of a mismatch can be the one you meant to change:

• Clip count does not match the template, shown on the bound template's card. Regroup the scenes here, or bind a template with a matching number of video placements.
• This length is not one the chosen video model produces, which is what you see after switching models on the Activation plan step. Open the length control and set a length the new model reaches; nothing is substituted for you.
• A clip does not match the length of the placement it plays, shown on the Clips header. A clip that runs longer than its placement is a soft note: the extra seconds are trimmed at render, and they are still generated and paid for. A clip that runs shorter is a warning, because the render does not stretch footage and the placement sits empty after the video ends. Two clips are named, then "And 2 more"; slack under a second is ignored, matching what the render itself tolerates. While the clip count does not match the template, this one stays quiet, since the pairing it would judge is the thing in question.
• One generation of this model covers three scenes, on a clip holding more. Split it, or move a scene into the neighbouring clip.
• The voiceover is longer than the plan, shown above the filmstrip when an audio field plays the generated narrator track. The narrator is cut off where the video ends, so lengthen a clip or shorten the scene lines.

One state is not a warning but a stop: if a placement is bound to a clip that has since left the plan, the cell will not render until you pick a clip again. Klyo never falls back to another clip, because silently rendering the wrong footage is worse than waiting. The Package step marks the placement and the render readiness readout repeats it.

Rebuilding the storyboard is the one edit that reaches that state on its own. Every scene is replaced, so the clip plan resets to a single clip over the new scenes, and a placement pinned to one of the old clips comes back needing a clip. Regroup the new scenes, then pick the clips again on the Package step.

Ask Kai

The assistant reads the clip plan. You can ask it how the scenes are grouped, and ask it to regroup them; it proposes the new plan and you approve it before anything is written. See Assistant.

Reference images

Reference images are what the video model looks at to match your brand's look. They steer the style and the colours; what the ad says still comes from the storyboard, and a reference image never fills a slot in the finished render. They are set on this step, in two places.

The campaign-wide set

The Reference images card above the clips holds one set every clip can read. Images already linked to the sample combination's catalogue items come in automatically, under From the sample items; Added images beside them is where you add moodboards or shots of your own. Every change saves as you make it.

Unchecking one of the sample item images leaves it out of writing the storyboard and out of the scene image previews. It does not keep it away from the video model. The video model reads each combination's own catalogue item images through the Item images entry described below, so to keep an item's photos out of the footage, drop Item images from that clip's reference priority.

A clip's reference priority

A video model only looks at a few reference images, so which ones come first decides what it copies. Each clip carries its own order, its Reference priority. The button in the clip's header shows how many images actually reach the model, and opening it shows the ordered strip they come from.

Three kinds of entry sit in that one list, and all of them move and come off the same way:

• Item images: that combination's own catalogue item images, in axis order. The count on the tile is for the sample combination, because every matrix cell brings its own.
• Campaign images: the campaign-wide set above the clips, beyond those same item images.
• Clip image: a picture you add for this clip alone, with Add images.

The arrows on each tile move it up or down the order. The model reads from the top and stops at its limit: Gemini Omni reads 7 reference images, Veo reads 3. Tiles past the limit are dimmed and captioned Not sent, and a tile standing for a whole stream that only partly fits reads Partly sent. The clip's own button says so too: it reads 3 of 9 when the limit cuts the set, and the popover names the model and its limit in one line.

That limit is why the order matters more than it looks. On Veo, a two-axis matrix with two photos per item already fills all three slots with item images, so a picture you add for the clip reaches the model only once you move it above Item images. Taking Item images off a clip frees the whole budget for the images that clip actually needs, and changes nothing about any other clip. A removed entry comes back from the Use item images and Use campaign images buttons beside Add images.

A clip you never touch reads item images first, then campaign images, which is what every clip did before the order existed.

Two limits are worth knowing before you spend a generation on this. Attaching references locks a Veo clip's base length, so it changes which lengths that clip can take; see Clip lengths. And on a clip long enough to need an extend, only the first generation is conditioned on the images, so on a 29 second Veo clip they steer the first 8 seconds and nothing after. The strip says so on a clip where both are true.

What the sample proof generates against

The sample proof composes its reference images exactly the way a matrix cell composes its own, so it leads with the sample combination's item images the way every cell does. If you unchecked some of the sample item images on the card above, they still reach the video model on the proof. Take Item images off that clip to keep them out of the proof and the matrix together.