The anatomy of a motion background: what 549 AI-generated backgrounds have in common

What we analyzed

Between July 2 and August 1, 2026 — a 31-day window — 325 creators generated 549 AI motion backgrounds on Scrollytelling, an average of 17.7 per day, peaking at 34 on July 29. Every one left a full trail: the exact prompt, the model, the render time, the retry count, and what visitors later did on the published page.

Why it matters: motion backgrounds have quietly become the default opening move. 88.2% of new stories (462 of 524) created on Scrollytelling in the window now open with an AI motion background. Yet as of August 2026, the search results for “motion backgrounds” are download galleries and stock libraries — Pixabay alone advertises 10,000+ free clips, Videezy over 24,000 — with, as far as we can find, no published data on what people actually generate when they can make their own. So we published ours.

Three of the 549, exactly as they shipped:

Want more? The motion backgrounds gallery shows the backgrounds behind this data, and published stories show them in the wild.

Exhibit A: scroll one yourself

A real generated motion background, served exactly the way the 549 in this study are — scroll it. It doesn’t autoplay; it’s tied to the scrollbar. Scroll inside the frame and the footage advances frame by frame.

Finding 1: the vocabulary of motion

We tokenized all 549 prompts and counted which aesthetic words appear in each. The result is remarkably consistent — a shared dialect. “Slow” appears in 96.7% of prompts and “dolly” in 96.5%. Light itself is the next obsession: “light” (83.1%) and “volumetric” (82.7%) — as in volumetric light, the visible shafts and haze that give AI video its depth.

Read the list top to bottom and a recipe writes itself: slow movement, a dolly camera, volumetric light, drifting dust particles, everything subtle, dark, and soft. That is, almost word for word, the prompt template we teach in how to make a motion background — because it’s what works.

Finding 2: one camera move rules them all

Motion backgrounds are not action footage. They are one camera move, executed slowly. “Dolly-in” appears in 94.5% of prompts (519 of 549) — the camera easing forward into the scene. “Push” reinforces it in 52.8% (290). Alternatives barely register: “parallax” 6.0% (33), “orbit” 2.6% (14). And 53% (291) of prompts go further and explicitly demand “no cuts” — a single continuous shot.

The slow dolly-in is the house style of the AI era. It flatters the models (steady forward motion hides temporal artifacts), it flatters the medium (a background should move without demanding attention), and — on Scrollytelling — it flatters the interaction, because a continuous forward shot scrubs beautifully under the scrollbar via the scroll video widget.

Finding 3: motion backgrounds are dark

We classified all 549 prompts by mood words. 52.5% are dark-mood (dark, night, dusk, moody, shadow, low-key) versus 8.2% explicitly light or bright — a 6.4:1 dark bias. The remaining 38.4% specify neither. In raw counts: 288 dark-only, 45 light-only, 5 mention both, 211 neither.

The bias makes practical sense. A background exists to sit behind white headline text, and dark, low-contrast footage is the safest canvas for legible type. It’s the same reason cinema trailers, product keynotes, and this very website default to black.

Finding 4: the 42-second background

Every background is a two-step render: a still image, then image-to-video. The still is fast — median 5.3 s, p90 10.4 s. The video step takes median 35.9 s, p90 46.6 s, max 107.6 s. Summed per story, prompt-to-motion is a median 41.5 s and p90 72.4 s (n=464 stories).

Reliability is the quieter headline: the video step succeeded on the first attempt for all 549 of 549 renders, and the still-image step hit 99.3% first-attempt (547 of 551), with 99.6% of stories (462 of 464) getting their hero still within the 3-attempt retry loop. For contrast: as of August 2026, imagine.art’s AI background video generator — the top result for “motion background maker” — advertises renders in the 3–5 minute range on a credit meter. The 42-second background is not a niche flex; it changes how casually you can iterate.

Finding 5: how they ship without killing the page

An autoplaying MP4 background is a performance tax — decode, buffer, battery. The 549 backgrounds in this study ship differently: each 20-second clip is split into up to 350 WebP frames (quality 90, max width 1200 px), at min(40, 350 ÷ duration) fps — so a 20 s clip yields 17.5 fps, exactly 350 frames — hosted on a CDN and scrubbed on scroll instead of played.

Frames beat video here for three reasons: seeking a real <video> element stutters, frames decode independently in either direction, and the browser fetches only what the reader scrolls past. It’s the same technique behind Apple-style product pages, packaged as the scroll video widget. (Uploading your own footage instead? Same pipeline: up to 100 MB of video, first 100 seconds used, capped at 350 scroll frames — a full 100 s upload is sampled at 3.5 fps.)

Finding 6: visitors actually land on them

Generation stats would be trivia if the backgrounds never met an audience. They do. 92.3% of stories published in the window have a motion background (48 of 52), and 97.3% of visitor sessions (396 of 407) landed on a motion-background story. Among visited stories with at least 5 sessions, the median story holds readers to a 10.1% average scroll depth, and the best-performing story averaged 25.2% depth over 18 sessions.

Disclosed limitation: a motion-vs-static scroll-depth comparison is not statistically possible yet (only 4 visited stories lack motion backgrounds). We’d rather publish the limitation than a fake comparison; the monthly refresh will revisit it as the static-background sample grows.

Methodology

All figures are first-party Scrollytelling product data. Competitor claims on this page are as observed in August 2026 and may change.

Steal the recipe

The dataset reduces to one sentence: a slow dolly-in through a dark scene with volumetric light and drifting dust particles, subtle and smooth, no cuts. You can apply it two ways:

• Write it yourself — the full prompt template, with variations per mood, is in how to make a motion background.
• Remix one for free — the free motion background maker lets you upload or remix a background in the browser, no signup required.
• Let AI generate it — start a story in the editor and it generates the opening background automatically, at a median 41.5 s of AI time as measured above (uses credits).

Comparing platforms instead? See our background animation software roundup.

Cite or embed this study

Every chart and number on this page may be republished with attribution. Copy the snippet:

Or cite in prose: “96.7% of AI motion-background prompts say ‘slow’ (Scrollytelling, August 2026, n=549).” The charts are plain inline SVG — open this page’s source, copy the <svg> block you need, and keep the source line intact.

Frequently asked questions