ManyMotions
Media ForgeGenerate images & video, fastMedia StudioNewBuild images & video by chatting with AIWorkflowsAutomate pipelines at scaleClipperNewBreak a video down & cut the highlightsMarketing AgentDone-for-you UGC video ads
MCPPricingFAQ
Log inStart for $1
AI Models/Sora 2/Prompting

How to prompt Sora AI

Sora 2 is the model you prompt like a screenwriter with a shot list attached. It excels at longer, physically coherent clips — up to 20 seconds — and rewards a hybrid format: a prose paragraph describing the scene, then labeled blocks for cinematography, action beats with timing, and dialogue. Its golden rule is one camera move plus one subject action per shot, with actions anchored in counts ("pedals three times, brakes at the crosswalk"). Vague verbs produce mush; counted beats produce cinema.

Anatomy of a Sora 2 prompt

  1. 1

    Style spine

    Declare the format before anything else: "IMAX-scale", "16mm documentary", "handheld phone footage". Reuse the exact same spine across every clip of a project — it is the cheapest consistency tool Sora has.

  2. 2

    Prose scene

    A 100-150 word paragraph, 3-6 sentences: subject with physical detail, evocative setting, what is at stake in the shot. This is where Sora's world simulation gets its material.

  3. 3

    Cinematography block

    A labeled list: camera framing and angle (wide establishing, low angle, over-the-shoulder), lens (35mm wide, 50mm natural, 85mm compression), movement, and mood.

  4. 4

    Action beats with timing

    Number the beats and anchor them in counts or seconds: "takes four steps to the window, pauses, pulls the curtain in the final second". Specific verbs plus counts are what keep 12-20s clips from drifting.

  5. 5

    Dialogue block

    Short attributed lines: Speaker: "line". Sora generates native audio and lip-syncs quoted dialogue — keep each line brief.

  6. 6

    Light & palette anchors

    Light quality plus direction, and 3-5 named palette colors: "soft window light, warm lamp fill; palette anchors: amber, cream, walnut brown". Palette anchors hold the grade steady across clips.

Template

[style spine]. [prose scene: subject, setting, stakes — 3-6 sentences]. Cinematography: - Camera: [framing, angle, lens, movement] - Mood: [tone] Actions: - [beat 1 with count or timing] - [beat 2 with count or timing] Dialogue: - [Speaker]: "[short line]" Lighting: [quality + direction]. Palette anchors: [3-5 colors].

10 example prompts that work

Long narrative take

16:9 · 20s · Sora 2 Pro · 1080p
Documentary-style handheld footage. A skateboarder in a faded green jacket carves through an empty city plaza at sunset, narrating his ride as golden light cuts between the buildings. He pushes twice, glides past a fountain, ollies a two-stair, lands clean and rolls toward camera grinning. He says over the ride: "Everyone asks why I skate alone. This is why — the whole city goes quiet." Lighting: low golden hour sun with long shadows. Palette anchors: amber, concrete grey, deep teal sky.

Twenty seconds is where Sora separates from every other model — one continuous, physically coherent ride with narration. The counted actions ("pushes twice", "ollies a two-stair") anchor the timeline so the clip does not wander.

Structured ad shot (block format)

16:9 · 8s · Sora 2 Pro · 1080p
Cinematic product commercial, 50mm at f/2.8. A frosted glass bottle of cold brew stands on a walnut counter in a sunlit kitchen, condensation beading on the glass.

Cinematography:
- Camera: medium close-up, eye level, slow push-in
- Mood: calm, premium, morning-fresh

Actions:
- A hand lifts the bottle from the counter, tilts it 30 degrees toward camera
- Light catches the condensation droplets rolling down the cold glass
- The hand sets it down with a soft thud in the final second

Lighting: soft window light from the right, warm lamp fill. Palette anchors: cream, walnut brown, deep coffee black.

The full block format from the Sora cookbook — prose first, then labeled sections. "Tilts it 30 degrees" and "in the final second" are the kind of measurable beats Sora executes on time.

Physics and materials showcase

16:9 · 12s · Sora 2 · 720p
16mm film look, handheld tracking. A cyclist in a wet nylon jacket rides through light rain across an old iron bridge, an 8-10 mph crosswind from camera left pressing the fabric against her arm. She pedals three times, stops pedaling, and coasts as the bridge cables sweep past. Footfalls of a jogger splash in shallow puddles in the background. Lighting: overcast softbox sky, cool and even. Palette anchors: slate blue, wet iron grey, muted red jacket.

Coding materials and forces explicitly — "wet nylon", "8-10 mph crosswind", "splashing in shallow puddles" — is the Sora technique that kills floaty, weightless motion. Name the physics you expect and the simulation delivers it.

Stylized world

16:9 · 8s · Sora 2 · 720p
Handcrafted paper-craft animation style, everything visibly made of folded and layered paper. A tiny paper fishing boat rocks on a sea of overlapping blue paper waves, a paper whale surfaces in three folding movements, water "splashes" as white paper confetti. The boat's lantern glows warm through the paper. Camera: wide shot, slow crane down toward the waterline. Mood: gentle, storybook. Palette anchors: indigo, cream, lantern amber.

Sora's creative range is the reason to pick it for non-photoreal work — committed style worlds hold consistently for the whole clip. Describe the material logic of the world ("splashes as paper confetti") and it stays coherent.

Dialogue two-shot

16:9 · 8s · Sora 2 Pro · 1080p
Cinematic drama, 85mm portrait compression. Over-the-shoulder two-shot in a late-night diner booth, neon from the window edging both faces.

Actions:
- The man in a corduroy jacket slides a folded map across the table
- The woman opens it, scans it for two seconds, looks up

Dialogue:
- Man: "It is a six-hour drive."
- Woman: "Then we leave tonight."

Lighting: warm practical lamps with cool neon rim. Palette anchors: amber, teal, diner-red vinyl.

Keep each line under a breath and attribute it — Sora lip-syncs quoted dialogue to the described speaker. The map is a prop beat that gives the actors business between lines, which reads far more natural than two people trading lines statically.

Documentary vérité

16:9 · 12s · Sora 2 · 720p
Documentary-style handheld, slight breathing movement, natural imperfect framing. A potter in a cluttered sunlit studio centers a mound of wet clay on the wheel, hands glistening. The wheel spins; the clay rises into a cylinder over four seconds; she narrows it at the neck with two fingers. Ambient: wheel motor hum, wet clay slip, a radio murmuring in the next room. Lighting: hard window light from the left, dust in the beam. Palette anchors: terracotta, bone white, denim blue.

"Documentary-style handheld" plus natural ambient sound is a reliable realism recipe on Sora. The clay rising "over four seconds" paces the transformation so it neither snaps nor drags.

Vertical hook for social

9:16 · 8s · Sora 2 · 720p
Handheld phone footage, vertical framing, energetic. Extreme close-up, shallow depth of field, 85mm: a barista snaps a shaker shut and shakes it twice hard, ice rattling. Whip to a medium shot as she pours a layered iced matcha in one continuous ribbon into a tall glass. She looks at the lens and says: "Stop paying seven dollars for this." Lighting: bright cafe daylight. Palette anchors: matcha green, cream, warm oak.

Front-load the loudest beat — the shaker snap — in the first second; the pour is the payoff. One camera change ("whip to a medium") is the maximum a short vertical clip should carry.

Image-to-video from a still

i2v · 16:9 · 8s · Sora 2 · 720p
Animate this frame with documentary restraint: the mist drifts slowly across the valley, the hiker's jacket flutters at the hem, a hawk crosses the sky far in the background left to right. Camera holds nearly static with a barely perceptible push-in. Ambient: wind over the ridge, fabric flutter, one distant hawk cry.

In image-to-video the still fixes composition and light; the prompt directs only motion and sound. Keep the camera nearly static and the motion secondary ("flutters at the hem") to preserve the source frame's fidelity. Note: input images with real human faces are rejected by the safety system — use generated or faceless stills.

Retro film look

16:9 · 12s · Sora 2 Pro · 1080p
Shot on 16mm, visible grain, slight gate weave, 1970s road-movie feel. A dusty coupe rolls down a desert two-lane at dusk, telephoto from a low angle beside the road so heat shimmer bends the horizon. The car passes camera; the shot holds on the taillights shrinking toward the mountains. Ambient: engine fading, cicadas rising, wind. Lighting: last direct sun, hard and orange. Palette anchors: burnt orange, dust tan, fading violet sky.

Sora commits to period film emulation when the spine names the stock, the artifacts and the era together. The hold-on-taillights ending is a single sustained composition — Sora keeps it stable where shorter-clip models drift.

Draft-to-final iteration pair

16:9 · 8s · Sora 2 · 720p
Same shot as previous generation, one change: switch to an 85mm lens and tighten to a medium close-up. Keep the action, timing, lighting and palette anchors identical. — Base prompt: Cinematic kitchen scene, a chef plates a seared duck breast in three precise movements, wide shot at eye level, soft overhead key light. Palette anchors: copper, slate, herb green.

The iteration discipline that makes Sora productive: change exactly one variable and name the change. Run these iterations on Sora 2 at 720p, and only send the locked prompt to Sora 2 Pro at 1080p for the final render.

Settings that matter

  • Duration

    4-20s in steps (4, 8, 12, 16, 20). 8s is the control sweet spot for ads — and two 8s clips cut together beat one 16s clip for reliability. Reserve 16-20s for continuous takes that genuinely need the length: that is Sora's signature move.

  • Model

    Sora 2 (720p) for iterating — generate 3-5 variants per shot cheaply. Sora 2 Pro for the final render once the prompt is locked.

  • Resolution

    Sora 2 outputs 720p. On Sora 2 Pro pick 1080p for hero shots and finals; the prompt transfers unchanged.

  • Aspect ratio

    16:9 for cinematic and YouTube work, 9:16 for vertical social. Match your framing language to the ratio before writing.

  • Mode

    Text-to-video by default. Image-to-video to lock an opening frame or chain clips (feed the last frame of clip A as the start of clip B) — but input images must not contain real human faces.

Do and don't

Do

  • Use the hybrid format: a prose scene paragraph, then labeled Cinematography / Actions / Dialogue blocks.
  • Anchor every action in counts or timing — "pedals three times", "pauses two seconds", "in the final second".
  • Hold to one camera move plus one subject action per shot; complexity goes into detail, not into more simultaneous events.
  • Code materials and forces explicitly — "wet nylon jacket", "8-10 mph crosswind", "condensation rolling down cold glass".
  • Keep a constant style spine and 3-5 palette anchors across every clip of a project, phrased identically.
  • Iterate one named change at a time ("same shot, switch to 85mm"), drafting on Sora 2 and finishing on Sora 2 Pro.

Don't

  • Don't rely on Sora for on-screen text — signs, labels and logos come out blurred or wrong. Add all text as overlays in Media Studio.
  • Don't reference celebrities, copyrighted characters, or upload images containing real human faces — the safety system blocks all three.
  • Don't use combat vocabulary ("fight", "battle", "weapon") even metaphorically — it trips false-positive blocks; reframe in cinematic terms.
  • Don't ask one 20s generation to carry a multi-scene story — split it into 8s shots and cut them in Media Studio.
  • Don't write vague motion verbs — "moves quickly" drifts; "takes four steps to the window and pulls the curtain" lands.
  • Don't change three things between takes; you lose the ability to know what worked.

Advanced techniques

Beats and counts: anchoring action to time

Sora's biggest differentiator — long coherent clips — is also its biggest failure mode when the prompt underspecifies time. The fix is counting: convert every action into measurable units. "Cyclist pedals three times, brakes at the crosswalk" instead of "cyclist rides past". "Takes four steps, pauses, pulls the curtain in the final second" instead of "walks to the window". Counts give the model a metronome; without one, a 16-second clip invents filler motion in the gaps.

The style spine: consistency across a whole project

Separate generations share nothing by default. To make a set of Sora clips cut together like one film, repeat an identical style spine in every prompt: the same film-stock phrase, the same lens, the same lighting formula, the same 3-5 palette anchors, and identical descriptive phrases for recurring places and characters ("the potter in the clay-streaked apron" — word for word, every time). For direct continuity between two shots, export the last frame of clip A in Media Studio and use it as the image-to-video start frame of clip B.

  • Same spine sentence in every prompt of the project
  • Palette anchors phrased identically, 3-5 colors
  • Recurring subjects described with the exact same words
  • Frame-chaining via i2v for hard continuity between shots

Working with the safety system

Sora's filters are stricter than most video models and produce occasional false positives. Cinematic terminology passes where casual phrasing trips: describe intensity through craft ("high-contrast chiaroscuro, urgent handheld energy") rather than conflict words. Specify "adult" for human subjects, lean on positive adjectives, and avoid brand names, celebrity likenesses and copyrighted characters entirely. If a reasonable prompt is blocked, reword the flagged concept in film language rather than resubmitting it unchanged.

Which variant to use

  • Sora 2

    Iteration workhorse — 720p, all durations 4-20s. Generate several variants per shot, pick the winner.

  • Sora 2 Pro

    Finals and hero shots — up to 1080p with the same prompt grammar. Render once the prompt is locked.

Frequently asked

How long can a Sora 2 clip be on ManyMotions?

Up to 20 seconds, in steps of 4, 8, 12, 16 and 20 — the longest continuous takes of any video model in Media Forge. For multi-scene stories, 8-second shots cut together in Media Studio give you more control than one long generation.

Sora 2 or Sora 2 Pro — which should I use?

Draft on Sora 2: it renders at 720p and is ideal for generating 3-5 variants per shot while you iterate. Switch to Sora 2 Pro at 1080p only for the final render. Prompts transfer between them unchanged.

Why was my prompt blocked?

Usually one of four triggers: a celebrity or copyrighted character reference, a real human face in an input image, combat/weapon vocabulary (even metaphorical), or content the filter reads as unsafe. Reword in cinematic craft terms — "tense standoff framed in hard side light" passes where "fight scene" may not.

Can Sora render readable text in the video?

No — on-screen text is Sora's known weak spot; signs and labels come out blurred or misspelled. Generate the shot clean and add every headline, price and CTA as an overlay in Media Studio.

How do I keep a character consistent across clips?

Describe the character with the exact same phrase in every prompt, keep the style spine and palette anchors identical, and chain shots by using the last frame of one clip as the image-to-video start of the next. Separate generations share no memory — consistency is engineered in the prompts.

Is there a remix mode for previous Sora videos?

Not currently in Media Forge. The practical equivalent: re-run the same prompt with one named change ("same shot, switch to 85mm"), or grab a frame from the previous output and drive an image-to-video generation from it.

Try Sora 2 in Media Forge →About Sora 2All models
ManyMotions

Tell me what you need and grab a coffee. I'll have it ready before you finish the cup.

Product

Marketing AgentWorkflow StudioMedia ForgeTemplatesAI Models

Open in app

Workflow StudioMedia ForgeProjectsAvatars

Company

ContactPrivacy PolicyTerms of ServiceLegal NoticeCancel subscriptionWithdraw from contract

© 2026 ManyMotions

support@manymotions.com