Video generation — short AI clips for Reels, Stories, and TikTok

You’re writing a Reel for tomorrow. You need 8 seconds of pasta being plated, with steam rising and a hand sliding the bowl onto a marble counter. Phone shots from last week look amateur. Stock footage looks like everyone else’s. The AI image you generated earlier is great for the cover, but Reels need motion. This is what video generation is for: a short editorial clip, generated to match your brand voice, attached to the same draft, ready to publish through the same flow as your image posts.

What it does

Inside the Composer, alongside the existing Generate image button, you’ll now see a Generate video button. Clicking it opens a small picker:

What to show — a short description of the moment (the subject, the action, the mood).
Duration — 5, 10, 15, or 30 seconds. Shorter is cheaper and arrives faster; longer reads better on Reels.
Quality — 480p (lower cost, fine for feed playback) or 720p (sharper, the right choice for Reels and Stories).
Style — Editorial (the warm restaurant baseline), Cinematic (slow and deliberate), Dynamic (kinetic kitchen action), Handheld (intimate, friend-filming-you look).

The picker shows the exact credit cost for the combination you’ve selected, and an estimate of how long the generation will take (3–6 minutes depending on duration). Click Generate and a progress card appears in the preview column — the rest of the Composer stays editable, so you can keep tweaking the caption while the video is being made.

When the video is ready, the card swaps to a playable preview with an Attach to post button. Click it and the video joins your draft alongside the image (or replaces it, your choice). Save the draft as usual; the video publishes through the same scheduling and approval flow as your image posts.

Where it costs credits

Video generation is an order of magnitude more expensive than image generation — a single 30-second 720p clip costs roughly the same as twenty AI images. The picker shows the exact credit cost before you confirm, and the Generate button only fires once you’ve seen and accepted that number. There are no surprise charges.

If the generation fails (provider hiccup, network bump), your credits are refunded automatically and the progress card switches to an amber-tone “didn’t work” message with a one-click retry. You only pay for clips you successfully receive.

When to pick which option

Use this table to pick duration + quality based on what the clip is for:

Destination	Recommended	Why
Instagram Reel	10–15s · 720p · 9:16	Reels reward 10–15s clips that play out; sharp enough to read at full screen.
Instagram Story	5–10s · 480p · 9:16	Stories autoplay-skip after 5–7s anyway; 480p is fine at story chrome size.
Facebook feed	5–10s · 480p · 1:1	Feed plays muted on autoplay — short and square reads best.
TikTok	15–30s · 720p · 9:16	TikTok rewards longer dwell time; viewers will tap to full-screen.
LinkedIn / Google Business	10–15s · 720p · 1:1	More patient audience; quality matters more than duration.

The defaults inside the picker (5 seconds, 720p, Editorial style) are a safe starting point for any feed post. Stretch the duration when you have a real arc to tell — a dish being plated step-by-step, a glass being poured to the brim.

The rule

Generate when the post needs motion. Edit the image when the post needs polish. Different tools, different costs, never interchangeable.

A static dish photo is the right call for 80% of restaurant social. Video earns its higher cost when motion is the message — flames, steam, hands, pour, slice. Don’t generate a video just to have one. Don’t generate a static “video” that could have been an image at one-twentieth the cost.

How to use it

Open Growth → Composer. Pick a dish (or write a free-form idea), pick a channel (Instagram, Facebook, TikTok…), pick a post kind (Post, Story, Reel). Hit Generate post as usual — this produces the caption + hashtags + still cover image.

Now look at the form column on the left. Below the Image section you’ll see a new Video section with a Generate video button. Click it. The picker opens.

Why your dish photo matters

When you pick A dish as the source, the system passes the dish’s primary photo to the AI as the first frame. The video then animates around that photo — the plate, the lighting, the composition you already chose — instead of inventing an unrelated scene from the prompt text.

The rule: the video is only as good as the still it starts from. A sharp, well-lit photo of carbonara becomes a sharp, well-lit video of carbonara. A blurry phone snap becomes a blurry video.

If the dish has no photo on file yet, the system falls through to text-only generation — the model invents a scene from the prompt, which tends to drift (a “steaming bowl of pasta” prompt might come back as a steaming mug). Upload at least one photo of the dish before generating video.

The Free idea source stays text-only — there’s no dish to anchor on, so the prompt does all the heavy lifting. Use this for atmosphere shots, B-roll, or footage of the venue that isn’t tied to a specific menu item.

Writing the prompt

The What to show field is where you describe the moment. Be specific. The model rewards concreteness:

✓ “Pizza coming out of the wood oven, peel sliding it onto a marble counter, steam rising”
✓ “Pouring a cold negroni into a chilled coupe glass, garnish dropping in slow-motion”
✓ “Hand sliding a bowl of carbonara across an oak counter, evening light from the window”
✗ “A nice pizza video” (too vague — the model will pick something generic)
✗ “Italian food” (no subject, no action — you’ll get something cinematic but irrelevant)

Mention the subject (the dish, glass, food item), the action (plating, pouring, slicing, sliding), and the mood or light (warm tungsten, evening light, marble counter). 30–60 words is the sweet spot.

Picking duration

Each duration jumps the cost roughly in proportion: 10s costs about 2× the 5s price; 30s costs about 6×. Start with 5s. If the clip feels rushed, regenerate at 10s. There’s no quality benefit to longer durations — the model uses its budget on more frames, not better-looking ones.

Picking quality

480p is half the cost of 720p and looks fine at Instagram-feed playback size (most viewers won’t notice). 720p is the right pick whenever the viewer might tap to full-screen — Reels especially. Don’t default to 720p out of habit; restaurant social usually plays at small size in a thumb-scroll feed.

Picking style

The four styles modify the motion and light the model is biased toward:

Editorial — the warm restaurant baseline. Subtle motion, candle/window light, hospitable. Default for 95% of posts.
Cinematic — slow, deliberate, documentary-style. Use for “hero” content where the dish is the star.
Dynamic — kinetic kitchen action. Flames, oil, hands, momentum. Use for prep videos, line-cook footage.
Handheld — intimate “a friend filmed this” look. Slight camera sway, natural light. Use for personality-led brands.

Submitting

Once everything’s set, the bottom-left of the picker shows the exact credit cost (e.g. 1,650 credits) and the estimated time (e.g. ~4 min to generate). Hit Generate.

The picker closes. In the right column where the post preview lives, a purple-tinted card appears: “Generating your video…”. A progress bar slides forward as the model works. You can keep editing the caption, hashtags, or the cover image in the rest of the Composer — they don’t conflict.

Cancelling mid-generation

If you change your mind, click Cancel on the progress card. The card closes and your credits are refunded immediately. You can start over without penalty.

When the video arrives

When the generation completes, the card swaps to a video preview with playback controls. Hit play to review it. Three options:

Attach to post — the video is added to your draft. The card closes. The draft now carries both the image and the video; when you publish, the channel-specific publisher (Instagram, TikTok…) picks whichever format makes sense.
Dismiss — discards the video from this draft (it stays in your Library — you can pull it back later). Use this if you don’t like the result.
Regenerate — open the picker again with the same parameters, charge fresh credits, get a different take. (Same convention as images: regenerate means a new charge.)

If it doesn’t work

Occasionally a generation fails — the provider’s queue is overloaded, the prompt triggers a safety filter, the network bumps mid-poll. The progress card switches to an amber-tone “didn’t work” message:

We couldn’t generate your video right now. Our video services are temporarily having trouble — try again in a few minutes.

Your credits are refunded automatically. Click Try again in the card to reopen the picker with the same parameters. If it fails twice in a row, change the prompt (sometimes a single word triggers safety filters) or try a different style chip.

Worked examples

Example 1 — Pizza Reel

You’re posting a Reel for a Margherita launch. You want 10 seconds of a pizza coming out of the wood-fired oven, being slid onto a marble board, with parmesan dust falling.

Channel: Instagram · Post kind: Reel (forces 9:16)
Duration: 10s · Quality: 720p · Style: Editorial
Prompt: “Pizza coming out of a wood-fired oven on a wooden peel, sliding onto a white marble counter, parmesan dust falling like snow, warm tungsten light, hands visible at the edges”
Cost: 1,650 credits · Time: ~4 min

While it generates, you write the caption (“Margherita, fired live tonight. Book in.”) and pick the hashtags. By the time the video lands, the rest of the post is ready. Attach, save draft, schedule for tomorrow 7pm. Done.

Example 2 — Cocktail Story

You’re posting a 5-second Story showing the bar making a Negroni for happy hour.

Channel: Instagram · Post kind: Story (forces 9:16)
Duration: 5s · Quality: 480p (Stories chrome obscures detail — 480p is plenty) · Style: Cinematic
Prompt: “Pouring a cold negroni into a chilled rocks glass, orange peel twisted over the top, condensation visible on the glass, low key amber bar light, single ice cube”
Cost: 825 credits · Time: ~3 min

Story posts work well as quick proof-of-life — viewers see your bar in action without committing to a full feed scroll. 5s is enough.

Example 3 — Behind-the-scenes for TikTok

You want a 30-second TikTok of your line cook plating a dish from start to finish.

Channel: TikTok · Post kind: Post (9:16 default)
Duration: 30s · Quality: 720p · Style: Dynamic
Prompt: “Line cook hands plating a pasta dish on a white plate, sauce being spooned, herbs torn, parmesan grated, finishing with the chef’s hand sliding the plate forward, evening kitchen light, focused and fast”
Cost: 4,950 credits · Time: ~6 min

This is the priciest cell in the matrix — but a polished 30s TikTok with native motion competes directly with creator content and is worth the spend if it goes onto a paid promotion later. Don’t make this the default; reserve it for posts you’ll re-use.

What it costs

The exact credits per cell — these match what the picker shows you:

Duration	480p	720p
5 seconds	390 credits	825 credits
10 seconds	775 credits	1,650 credits
15 seconds	1,163 credits	2,475 credits
30 seconds	2,325 credits	4,950 credits

A typical month’s video budget for a single venue, posting one Reel per week and two Stories: 4 × 1,650 (Reels) + 8 × 825 (Stories) = 13,200 credits. Compare to image-only social: 5 credits × ~30 posts = 150 credits. Video shifts your AI burn rate by an order of magnitude — budget for it intentionally.

Standard and Premium tiers

The picker today shows one quality level — Fast (the Seedance 2 Fast engine). Two higher tiers are coming:

Standard — better motion fidelity, slightly slower generation. ~2× Fast pricing. Best for hero content.
Premium — top-shelf models (the same engines used by major studios for short film). ~3–4× Fast pricing. Reserved for paid-promotion-grade content.

When these land, the picker will gain a tier selector at the top. Until then, Fast handles every Phase 1 use case at the lowest cost.

Why the wait time

Image generation completes in 10–30 seconds because the model only produces one frame. Video generation is 4–6 minutes because the model is producing hundreds of frames AND ensuring each one is consistent with its neighbours — same dish, same lighting, same composition, moving naturally. That’s a fundamentally harder problem and there’s no way to make it instant short of pre-rendering a library of fake clips (which would look generic).

The progress card subscribes to the live job state and re-renders as the model works — so you’ll see “Submitting…” → “Generating…” → “Almost there…” → “Saving to your library…” → ready. You don’t have to refresh anything. The Composer is fully responsive while the job runs, so you can keep working on the caption.

Where the videos live

Every successfully-generated video is saved into your Library alongside your AI-generated images — same “you pay, you keep” guarantee. If you generate a clip, attach it to a draft, then later delete the draft, the clip stays in the Library. You can re-attach it to a different post anytime without paying again.

The Library will gain Images / Videos / All tabs in the next update — for now videos appear in the same grid as images with a small play-icon overlay so you can tell them apart.

Composer — the surface that hosts the Generate video button.
Image editor — for retouching cover images without re-charging credits.
Library — where every generation lives, even ones you didn’t save into a post.
Prompt tuning — for super admins who want to override the default style and aspect-ratio prompt fragments.