Best Sora Prompts 2026: 30+ Templates and the Anatomy Behind Them

The best Sora prompts share a fixed anatomy: subject, action, camera, lighting, style, and duration, in that order. OpenAI's Sora model will happily accept a one-line prompt and return something watchable, but the difference between a clip you post and a clip you delete is almost always the prompt structure, not the model. This guide gives you 30+ tested prompt templates across six categories, walks through the anatomy each one uses, and shows how to iterate a weak prompt into a strong one. If you already have a script and want a finished narrated video (not a raw clip), skip to the MakeAIVideo pricing section at the end for how these prompts slot into a full pipeline.

Copy the prompts, but read the anatomy first. Every template below follows the same six-part frame. Once you can name each part in your head, you can write a fresh prompt quickly instead of scrolling for a template that almost fits.

What makes a strong Sora prompt

Sora is a text-to-video and image-to-video model. It takes natural language and returns short clips, currently up to around 20 seconds in the general consumer flow, with the exact ceiling depending on which access tier you are on and which model variant is live at the time. OpenAI's Sora product page and the Wikipedia entry both track version changes as they land.

The version change that mattered most for prompting was Sora 2, which OpenAI shipped on 30 September 2025 alongside the standalone Sora app. TechCrunch reported OpenAI's framing directly: earlier video models were "overoptimistic" and would "morph objects and deform reality" in order to satisfy a text prompt, where Sora 2 follows the laws of physics more faithfully. That is the whole reason physical specificity in a prompt now earns its keep. The prompt principles do not change with the version; the model gets better at following them. Our OpenAI Sora review covers the version history and the access tiers in more detail, and the AI video model comparison covers which rivals read these same structures well.

A strong Sora prompt names six things in one sentence or one short paragraph:

• Subject. Who or what is in the shot. Be specific: not "a woman", but "a 30-year-old barista in a green apron".
• Action. What the subject is doing. Verb-first, one motion per clip.
• Camera. Shot type and camera movement. "Medium shot, slow dolly in" beats "cool angle".
• Lighting. Named lighting family. "Warm golden hour side light" beats "nice lighting".
• Style. The visual reference you are borrowing from. "Shot on 35mm film, muted colour grade" is a full style. "Cinematic" is not.
• Duration. How long the shot should feel. Sora reads timing cues like "the full 10 seconds" or "a slow 8-second push in".

Miss two of these and Sora fills the gap with defaults. Miss three and the clip looks like a stock-footage cousin of what you wanted. The 30 templates below all include every part explicitly, so you can see the anatomy sitting on the page. If you would rather skip the prompt craft and get a finished video, MakeAIVideo's prompt to video mode builds the scenes from one description, and image to video handles the case where you have a still you want to animate.

The prompt anatomy in one worked example

Weak prompt:

"A woman walks through a city at night."

Nothing wrong with it, but Sora has to invent everything. The rendered clip will be generic because the prompt was generic. Now the same idea with the six-part anatomy:

"Medium tracking shot of a woman in her late twenties wearing a rain-slick trench coat, walking briskly across a wet Tokyo crosswalk at 2 a.m., neon signs reflecting in puddles, cool cyan and magenta lighting, shot on anamorphic lens with subtle lens flares, 35mm film grain, the camera glides alongside her at walking pace for the full 8 seconds."

That is the same idea with subject (woman, late twenties, trench coat), action (walking briskly across a crosswalk), camera (medium tracking shot, glides alongside), lighting (neon, cyan and magenta), style (anamorphic, 35mm grain), and duration (8 seconds) all named. The rendered clip will not be a miracle every time, but it will be closer to what you pictured, and the failure modes will be legible enough to iterate on.

The prompt library: 30+ Sora prompts by category

Each category below has 5 to 6 prompts. Copy the whole prompt, then edit only the parts you need for your specific brief. Do not strip the camera, lighting, or style cues; those are doing most of the work.

Product ads (6 prompts)

For direct-response ads, keep the shot count to one and the reveal early. If you would rather have a finished narrated ad than a raw clip, an AI ad maker builds one per run from your product photos and brief.

"Slow 5-second push-in on a matte black bottle of [PRODUCT] resting on a wet slate surface, single overhead soft key light with rim light from behind, shallow depth of field, water droplets catching the light, muted colour grade, shot on macro lens with a subtle rack focus onto the label at the two-second mark."

"Overhead top-down 8-second shot of two hands unboxing [PRODUCT] on a warm oak table, natural window light from the left, one hand lifts the lid slowly, revealing the product on a bed of cream tissue paper, cosy neutral palette, shot on 35mm with soft grain, camera holds perfectly still."

"10-second orbit shot circling a [PRODUCT] on a rotating pedestal against a seamless off-white infinity backdrop, three-point studio lighting with a soft key and warm rim, subtle dust particles catching the rim light, hyper-clean commercial style, shot on a long lens for compression, camera moves at a steady slow speed."

"6-second locked-off medium shot of a [PRODUCT] in the palm of a woman's hand, held out to a soft window, warm morning side light, hand rotates the product 30 degrees so the surface catches the light, out-of-focus kitchen counter in the background, warm domestic colour grade, shot on 50mm lens."

"8-second handheld POV shot of a person pouring [PRODUCT] into a clear glass on a kitchen island, natural top-down window light with soft warm bounce, liquid catches highlights as it falls, glass fills to two-thirds and the pour stops, everyday kitchen background softly out of focus, shot on 28mm lens with subtle handheld motion."

"5-second cinematic macro shot on the texture of [PRODUCT], camera slowly drifts across the surface at extreme close-up, single soft raking side light, black background, dust motes visible in the beam, muted premium colour grade, shot on macro lens with the shallowest possible depth of field."

Faceless YouTube B-roll (6 prompts)

Cut-away B-roll is the single biggest use case for Sora on a faceless channel: the voiceover carries the story, and Sora fills the visual slots between graphics. Each prompt below is a self-contained 5 to 10-second cutaway. If you want the full production loop, the faceless YouTube channel page covers script-to-final-MP4 for exactly this format.

"10-second wide aerial drone shot flying slowly over dense evergreen forest at dawn, low mist clinging to the treetops, cool blue morning light with warm sun breaking over the horizon on the right, subtle camera drift forward, cinematic nature documentary style, shot on high-resolution drone."

"8-second slow-motion shot of hands typing on a mechanical keyboard in a dim room, warm desk lamp side light, RGB glow from the keys, out-of-focus monitor in the background, tight over-the-shoulder framing, moody indoor tech style, shot on 50mm at f1.4."

"6-second locked-off shot of a chalkboard with a chalk hand writing the words '[CONCEPT]' in cursive, warm classroom lighting from above, tight framing on the writing, subtle chalk dust rising as the hand moves, documentary tutorial style, shot on 35mm."

"10-second overhead time-lapse of a busy Tokyo intersection at night with cars and pedestrians flowing across the crossing, neon signs and street lights, cool blue and pink colour palette, camera locked off, hyper-lapsed motion, shot in the style of a Baraka-inspired cityscape."

"8-second slow tracking shot along a row of vintage books on a wooden library shelf, warm tungsten library lighting, camera moves horizontally at reading pace, shallow depth of field on the spines, muted warm colour grade, shot on a 50mm lens."

"5-second close-up on a stopwatch on a wooden desk with the second hand ticking, warm afternoon side light, subtle rack focus onto the digits as they change, out-of-focus notebook in the background, editorial documentary style, shot on macro lens."

Cinematic shots (6 prompts)

These are the prompts for opening shots, transitions, and mood-setters. Keep camera moves slow; Sora handles subtle motion better than fast whip-pans.

"12-second slow dolly-in on a lone figure standing at the edge of a cliff overlooking a misty valley at sunrise, warm orange sun rising behind them, cool blue morning haze in the valley, silhouetted figure in a long coat with hair moving in the wind, cinematic wide shot, shot on anamorphic lens, muted teal-and-orange colour grade."

"10-second slow crane shot rising up above a rain-soaked city street at night, neon-drenched storefronts, pedestrians with umbrellas moving below, cool cyan and magenta reflections on wet pavement, cinematic Blade Runner-inspired mood, shot on wide lens with subtle lens flares."

"8-second locked-off wide shot of a solitary house on an empty prairie under a vast twilight sky, deep purple and orange gradient sky, single warm window light glowing in the house, no camera movement, painterly cinematic style, shot on medium-format wide lens."

"10-second slow tracking shot moving parallel to a horse galloping across a golden wheat field at sunset, warm side light casting long shadows, camera holds the horse in the centre of frame, shallow depth of field on the background, cinematic Terrence Malick-inspired style, shot on 35mm film."

"8-second slow push-in on an astronaut standing on a red desert planet looking at a giant ringed planet on the horizon, harsh key light from an unseen sun on the right, deep magenta and orange sky, subtle dust drifting past the visor, cinematic sci-fi style, shot on wide lens with subtle lens flares."

"12-second slow overhead crane shot descending from black into a candlelit medieval hall with a long banquet table, warm firelight from wall sconces, guests seated on either side, camera lowers slowly onto the table's centrepiece, cinematic period-film style, shot on 32mm anamorphic."

Tutorial and explainer B-roll (5 prompts)

For explainer videos, the B-roll should visually reinforce a concept without stealing focus from the voiceover. Keep motion subtle and framing clean.

"8-second locked-off top-down shot of two hands sketching a rough mind map in a black notebook with a fine-liner pen, warm desk lamp side light, camera holds still, hand adds a new branch to the diagram, minimal editorial style, shot on 50mm macro lens."

"6-second slow zoom into a laptop screen showing a clean line graph rising over time on a soft grey dashboard interface, cool balanced office light, camera pushes in from a medium shot to a tight framing on the graph line, minimal corporate documentary style, shot on 35mm."

"10-second tabletop shot of a hand arranging three wooden blocks in a row on a white surface, soft even overhead lighting, hand slides the middle block into position at the four-second mark, minimal clean background, shot on 50mm lens with locked-off framing."

"7-second slow tilt-up shot from a set of physical dominoes falling in sequence on a wooden desk, warm side light, camera rises from tabletop level to eye level as the last domino falls, muted warm colour grade, tabletop tutorial style, shot on 35mm macro."

"8-second overhead shot of a barista's hands preparing a pour-over coffee in a chemex, natural window light from the side, water pours in a steady stream from a gooseneck kettle, steam rising softly, minimal editorial food style, camera locked off, shot on 50mm macro."

Character-driven scenes (4 prompts)

Sora renders humans best when you specify age band, wardrobe, and one clear emotion. Do not stack three emotions in one prompt. If you need the same character across multiple clips, an AI talking avatar is a better fit than trying to reroll a stranger every time.

"10-second medium close-up on a woman in her mid-thirties wearing a soft grey knit sweater, seated at a warm-lit kitchen table, hands wrapped around a ceramic mug of tea, she looks out of the window with a quiet thoughtful smile, natural morning side light, muted warm colour grade, shot on 50mm at f2 with shallow depth of field."

"8-second medium shot of a bearded man in his forties wearing a navy fisherman's sweater, standing at the wheel of a small fishing boat at dawn, salt spray on his face, he squints into the wind with a steady grounded expression, cool overcast morning light with a warm sunrise glow behind, cinematic documentary style, shot on 35mm."

"10-second tight over-the-shoulder shot of a young chef in a white apron plating a dish in a bustling restaurant kitchen, focused concentration on the plating, warm overhead kitchen lights with steam rising in the background, out-of-focus movement of other cooks behind, editorial food documentary style, shot on 50mm."

"8-second medium tracking shot of a child in a yellow raincoat running through puddles on a grey sidewalk, camera moves at their eye level, laughter and pure joy on their face, cool overcast light with the yellow coat popping against the muted background, playful cinematic style, shot on 35mm handheld."

Music video and creative (5 prompts)

For music-video work, the prompt should imply rhythm even in a silent clip. Motion beat cues like "slow-fast-slow" or "on the two-second mark" translate surprisingly well. For a fully rendered music video with matching visuals, the AI music video generator is the pipeline; Sora is only the visual generator.

"10-second slow-motion shot of a dancer in a flowing red dress spinning against a jet-black background, single strong side light, the dress catches the air and forms wide sweeping curves, dramatic contrast, shot on 85mm at f1.4, editorial fashion film style."

"8-second symmetrical wide shot of a lone figure walking toward the camera down a long white corridor with flickering fluorescent lights, deep shadows on the ceiling and floor, figure walks in slow deliberate steps, cool clinical colour palette, cinematic music-video style inspired by Chris Cunningham, shot on 25mm."

"12-second overhead spinning shot looking straight down on a group of five dancers lying on a black floor arranged in a five-pointed star, arms and legs moving in slow synchronised waves, single overhead spotlight, high-contrast lighting, editorial music-video style, camera rotates slowly clockwise, shot on wide overhead rig."

"10-second slow dolly shot moving through a smoky nightclub at 3 a.m., silhouetted dancers around the frame, warm red and deep blue lighting alternating, subtle strobe effect, atmospheric cinematic mood, shot on 35mm with lens flares, camera drifts through the crowd at walking pace."

"8-second locked-off wide shot of a solitary figure standing in a field of white flowers under a soft grey sky, wind moves through the flowers in slow waves, figure stands perfectly still in a long dark coat, muted painterly colour grade, cinematic music-video style, shot on 32mm."

Common Sora prompt mistakes

Five failure patterns show up over and over on Sora clips that get deleted. Every one is fixable in the prompt itself.

1. Stacking multiple shots in one prompt. "Wide establishing shot, then a close-up on her face, then a tracking shot as she walks away" is three prompts, not one. Sora will pick one and ignore the rest, or blend them into something incoherent. One prompt, one shot, one motion. If you need coverage, render the shots separately and cut them together in a proper edit, the way the video script generator breaks a script into scene beats.

2. Vague style words. "Cinematic", "epic", "beautiful", and "professional" are dead words in a video prompt. They describe an outcome, not a style. Replace them with a specific reference: "shot on 35mm film", "anamorphic lens with lens flares", "muted teal-and-orange colour grade", "Wes Anderson symmetrical framing". The more specific the style reference, the closer Sora lands to what you pictured. The same rule applies to prompts on other models; if you are cross-checking against Runway, our Runway alternatives comparison walks through the equivalent prompt language on that model.

3. Skipping the camera cue. A prompt without camera language will default to a locked-off medium shot roughly 70 percent of the time. That is fine if you meant that; it is a wasted render if you wanted a dolly-in. Name shot type (wide, medium, close-up, macro), framing (over-the-shoulder, POV, overhead), and movement (locked-off, slow push-in, tracking, orbit) every time.

4. Cramming dialogue into the prompt. Sora is a visual generator. Any dialogue you add to the prompt will render as lip movement that does not match audio you plan to add later, or as garbled text on-screen. Save the script for the voiceover layer; keep the prompt purely visual. Our how to write an AI video script breakdown covers the script-and-visuals split in more detail.

5. Over-long prompts. A prompt over 100 words starts to confuse the model. Sora attends to the strongest signals and drops the rest. If you find yourself writing "and", "also", and "additionally" repeatedly, split the idea into two clips. One clear prompt beats one crowded prompt every time.

6. Ignoring duration. Sora will default to a length of its choosing if you do not specify. Every prompt in the library above ends with an explicit duration cue ("for the full 8 seconds", "10-second orbit", "12-second slow crane"). That single phrase changes both the pacing of the motion and the density of what the model tries to fit into the clip.

How to iterate a weak Sora prompt into a strong one

Iteration is the whole game. Here is the same weak prompt from earlier, walked through three iteration passes.

Pass 0 (weak):

"A woman walks through a city at night."

The rendered clip will be generic because the prompt is generic. Sora invents subject, camera, lighting, and style. There is no anchor point.

Pass 1 (add subject and setting):

"A woman in her late twenties wearing a trench coat walks across a wet Tokyo crosswalk at 2 a.m. under neon signs."

Better. Sora now has age band, wardrobe, location, and time of day. But the camera and lighting are still up to the model, so consecutive renders will look wildly different.

Pass 2 (add camera and lighting):

"Medium tracking shot of a woman in her late twenties wearing a rain-slick trench coat, walking briskly across a wet Tokyo crosswalk at 2 a.m., neon signs reflecting in puddles, cool cyan and magenta lighting."

Now the shot type, movement, and lighting palette are locked. The rendered clip will be much more consistent across takes because the model has fewer choices to make.

Pass 3 (add style and duration):

"Medium tracking shot of a woman in her late twenties wearing a rain-slick trench coat, walking briskly across a wet Tokyo crosswalk at 2 a.m., neon signs reflecting in puddles, cool cyan and magenta lighting, shot on anamorphic lens with subtle lens flares, 35mm film grain, camera glides alongside her at walking pace for the full 8 seconds."

Full six-part anatomy. If the first render is not right, the next iteration only touches the variable you want to change. Want a wider shot? Swap "medium" for "wide" and leave everything else. Want warmer lighting? Swap the palette. Iteration becomes surgical instead of a full rewrite.

Iterate one variable at a time. If you rewrite three parts of the prompt between renders, you lose the ability to attribute what changed to what caused it. Change one variable, render, evaluate, then change the next.

Where MakeAIVideo fits in the Sora workflow

Sora produces individual clips of up to about 20 seconds. A finished narrated video is a different job: script, voiceover, scene assembly, captions, music, and export in the aspect ratio the platform actually wants. MakeAIVideo is built for that finished-video job, with its own generated scenes.

Two workflows people run in practice:

Script first, prompts second. Done by hand, you write or generate a script, split it into scene beats, write a Sora prompt for each beat, render, then assemble. A pipeline does the script split, the voiceover, the scenes, and the assembly automatically; MakeAIVideo generates its own scenes (real stock footage, or AI scenes on MiniMax Hailuo or Kling 2.6) rather than using Sora. MakeAIVideo's text to video mode is built for this workflow: describe the video and it returns a narrated MP4.

Sora clips first, script wrapped around them. You already have a small library of Sora clips you like, and you want to string them into a coherent video with a voiceover. That is an editing job: cut the clips in your editor of choice and lay the voiceover underneath. If you are starting from stills rather than clips, MakeAIVideo's image to video mode turns each one into a short clip.

Neither workflow makes MakeAIVideo a competitor to Sora. Sora generates raw clips; MakeAIVideo makes whole narrated videos with its own scenes, and does not take uploaded Sora clips. If you are shipping vertical-first content for TikTok and Reels, the AI Shorts generator handles the aspect-ratio and caption formatting automatically. That matters because Sprout Social's social media video specs guide puts Reels, TikTok, and Shorts all at 9:16 while standard YouTube stays at 16:9.

The wider case for pairing raw generation with an assembly pipeline is well-covered in HubSpot's video marketing benchmarks, which repeatedly find that short-form video is where organic reach is compounding on both TikTok and Reels. For a wider view of Sora rivals to source additional clips from, the Sora alternatives roundup covers Runway, Kling, Veo, and Luma.

The prompt is the input to Sora. A finished narrated video is a different job. Two separate craft skills. This guide covers the prompt; MakeAIVideo covers the finished video.

Sora prompts FAQs

What is a Sora prompt?

A Sora prompt is the natural-language description you give OpenAI's Sora model to generate a video clip. Strong prompts specify subject, action, camera, lighting, style, and duration in one sentence or short paragraph.

How long can a Sora prompt be?

Prompts under 100 words work best. Sora attends to the strongest signals in a prompt and drops the rest, so a tight 40-to-80-word prompt with clear camera and lighting cues usually beats a rambling 200-word prompt.

How long are Sora clips?

Sora currently renders clips up to around 20 seconds in the general consumer flow, with the exact ceiling depending on your access tier and the model variant that is live. Check OpenAI's Sora product page for the current limits.

Do I need to describe the camera in a Sora prompt?

Yes, always. A prompt without camera language defaults to a locked-off medium shot most of the time. Name the shot type, framing, and any camera movement explicitly to control what the render looks like.

Can Sora do image-to-video from a still frame?

Yes. Sora supports image-to-video generation as well as text-to-video, letting you provide a still frame plus a prompt describing the motion. MakeAIVideo's image to video mode covers the same job inside a finished-video tool.

How do I stop Sora from generating garbled text on-screen?

Do not put dialogue or long text inside the prompt. Sora treats text you write in the prompt as visual content to render, which often produces unreadable lettering. Save the script for the voiceover layer instead.

Can I use Sora clips commercially?

OpenAI's terms allow commercial use of Sora outputs on paid plans in most cases, but always check the current terms on the Sora product page and follow any regional restrictions. Also check platform disclosure rules like the FTC's AI endorsement guidance for paid promotions.

Why do consecutive Sora renders of the same prompt look different?

Because Sora, like most diffusion-based video models, samples from a probability distribution on each render. Locking the camera, lighting, and style cues in the prompt reduces the variance between takes, but two renders of the same prompt will never be pixel-identical. The same variance shows up on Kling, Runway, and Veo, which is why our Kling AI alternatives roundup covers the same seed-and-lock approach on that model.

What is the best way to iterate on a Sora prompt?

Change one variable at a time. If you rewrite three parts of the prompt between renders, you cannot tell which change caused which visual difference. Adjust the camera, render, evaluate, then adjust the lighting on the next pass.

Do I need a separate tool to turn Sora clips into a finished video?

Usually yes. Sora produces individual clips, not a finished narrated MP4. A finished-video tool like MakeAIVideo covers the script, voiceover, scene assembly, captions, and export in one job; MakeAIVideo pricing starts at $9 a month.

Where to start

Pick one category from the library above and render three prompts back-to-back using the templates as-is. Do not edit them on the first pass; the point is to see how the six-part anatomy actually renders in your specific style and use case. Once you have a feel for how Sora responds, iterate one variable at a time on the prompt that came closest to what you wanted.

If you already have a script and want the whole thing rendered as a finished narrated video, the video script generator hands you a scene-split script ready to feed into script to video, the speech time calculator checks the runtime, and image to video covers any stills you want animated. Plans start at $9 a month for 450 credits, topped back up each billing month.

Save one prompt template you like, edit the bracketed parts, render, evaluate. That is the whole loop.