Sora prompt guide: How to write better video prompts in 2026

Summary

Learn how to write Sora prompts like a clear video shot brief: subject, action, setting, camera framing, lighting, and style. This guide explains a five-part prompt formula with weak-vs-strong examples, camera movement tips, lighting prompts, dialogue prompts, image-to-video prompts, and ready-to-use templates.

It also covers common Sora prompt mistakes and the latest 2026 Sora status, including web/app discontinuation and the API timeline.

‍

If you're searching for a Sora 2 prompt guide that skips the jargon and gets straight to what actually works, this is built for exactly that. This Sora AI prompt guide covers how to write Sora prompts and the mistakes that you should avoid. Also, get some instant prompt examples that you can simply copy and paste for instant outputs! 

What is a good Sora prompt?

A good Sora prompt tells the model exactly what to show in the end result. Describe the subject, the action, the camera movement, the lighting, and the setting in simple sentences, almost like you're briefing a camera operator on set. 

Skip vague mood words and skip long backstory. Sora video prompts work best when the instructions read like camera directions. If you want consistent results, structure your prompt the same way every time: subject and setting first, then action, camera, followed by lighting and style.

One more thing before you dive in. As of June 2026, Sora's web and app experiences are no longer running. OpenAI's Help Center confirms the Sora web and app experiences were discontinued on April 26, 2026, and the Sora API will be discontinued on September 24, 2026. We'll cover what that actually means for you a bit further down, because it changes how you should think about using anything in this guide right now.

That said, the prompting logic itself still matters. Whether you're working through the API while it lasts, testing similar tools, or just want to understand how to write Sora video prompts before the next model shows up, the structure below applies.

Sora prompt structure: The core formula

Here's the part most people skip, and it's the reason their videos come out generic. 

Sora doesn't read your prompt the way a person reads a story. It reads it the way a cinematographer reads a shot list. These structural habits mirror how OpenAI Sora prompts are demonstrated across the company's own documentation. 

Every detail you leave out gets filled in randomly by the model, and that's usually where things go wrong.

Think of your prompt like five short building blocks stacked in order. The best Sora prompts all share one trait: specificity, and that comes straight from following this structure instead of improvising.

(A quick context note before we break this down: you don't need all five blocks every single time, but the more of them you include, the less Sora has to guess). 

The five-part sora prompt formula

• Subject - who or what is in frame, described with specific, recognizable details (not "a man," but "a man in his 60s wearing a grey wool coat")
• Action - one clear, physically plausible movement or beat, not a whole sequence of events
• Setting - the location, time of day, and weather, if relevant
• Camera - the shot type and any movement (static, pan, dolly, handheld, overhead)
• Lighting and style - the mood, color tone, and visual reference (cinematic, documentary, anime, soft daylight, neon)

Put together, a basic prompt looks something like this:

"A woman in a red raincoat walks slowly across a wet cobblestone street at dusk. The camera tracks her from a low angle, moving alongside her at walking pace. Streetlights are just turning on, casting warm orange light against the blue evening sky. Cinematic, shallow depth of field."

Notice what's not in there. No backstory, no explanation of why she's walking, no list of emotions she's supposedly feeling. Just what the camera would actually see.

Weak vs. strong prompts

Why this order matters

Sora tends to weigh earlier parts of a prompt more heavily than the parts at the end. So if camera movement matters most to you for a particular shot, move it earlier in the sentence. If lighting is the whole point of the clip, lead with it. 

Once you understand how to write Sora video prompts using this five-part structure, everything else in this guide is just a variation on the same theme.

For a broader look at how the model itself was built and what made it different from other video tools, this breakdown of Sora and how text-to-video generation works is worth a look if you want the fuller picture before diving into prompting.

How to turn a basic idea into a strong prompt

One of the easiest mistakes beginners make is starting with an idea and stopping there. A simple concept can work, but the more specific you are, the less the model has to guess.

For example, imagine your starting idea is: "a dog running on a beach." It's a perfectly good concept, but it leaves a lot of unanswered questions. What kind of dog? What time of day is it? Is the camera close to the action or far away? What mood are you aiming for?

Instead of adding random details, build the prompt one layer at a time. Start with the subject, then add the action, setting, camera direction, and lighting.

Basic idea:

"A dog running on a beach."

Improved prompt:

"A golden retriever runs along a quiet beach at sunset as waves roll onto the shore. The camera tracks beside the dog at a low angle, capturing sand kicking up with each stride. Warm golden-hour light reflects off the water. Cinematic style, shallow depth of field."

Notice that the core idea never changed. The stronger version simply gives the model a clearer picture of what you want to see. When in doubt, think like a camera operator and describe exactly what would appear on screen.

Sora prompt examples by use case

Different goals call for different prompt habits. A product shot needs a different language than a moody short film clip. Below are working examples broken out by the situations people ask about most.

A short note before the breakdown: these are written as full prompts you can adapt, not fragments. Swap out the specific details for your own subject and keep the structure intact.

Sora camera movement prompts

Camera direction is where most beginner prompts fall apart, because people either skip it entirely or use vague terms like "cool camera angle." Be specific instead. Use real cinematography terms; Sora has clearly been trained on footage tagged with this vocabulary, so it responds well to it.

Examples that tend to work:

• "Slow dolly-in on a chess board as a hand moves a piece, shallow depth of field, the background blurring into soft bokeh."
• "Handheld camera following a runner through a forest trail, slight shake, natural motion blur on the trees passing by."
• "Overhead drone shot slowly rising above a wheat field at sunrise, revealing a lone farmhouse in the distance."
• "Static tripod shot, locked frame, a cat walks into frame from the left and settles on a windowsill."

Stick to one camera movement per prompt unless you're working with a longer clip duration. Asking for a pan, a zoom, and a tilt all at once usually confuses the output.

Sora lighting prompt examples

Lighting carries a huge amount of the visual mood, and it's one of the easiest things to get specific about without writing a novel.

Try combinations like these:

• "Golden hour lighting, long shadows stretching across the pavement, warm amber tones"
• "Overcast daylight, flat and even, muted colors, slightly desaturated"
• "Volumetric fog with a single spotlight cutting through, high contrast, near-darkness around the edges"
• "Neon-lit alley at night, pink and blue reflections on wet ground, slight haze"

If you're not sure what term to use, describe the time of day and weather instead. "Late afternoon, slightly cloudy" works almost as well as technical lighting jargon for a lot of shots.

Sora dialogue prompts

If your clip includes a character speaking, the line needs to be written out clearly and kept separate from the visual description. According to OpenAI's official guidance on the model, a clear prompt describes a shot as if you were sketching it onto a storyboard, stating camera framing, depth of field, action in beats, and lighting and palette, with dialogue described directly so the model can tell visual instructions apart from spoken lines. 

A working format looks like this:

"A barista behind a coffee counter looks up and smiles at the camera. Warm cafe lighting, soft focus on the espresso machine behind her.

She says: 'Your order's almost ready, just give it a minute.'"

Separating the spoken line from the scene description on its own line, with quotation marks, gives Sora a cleaner signal about what's dialogue and what's an audio cue versus a visual direction.

Sora image-to-video prompts

Starting from a reference image changes how you should write the prompt. The image sets the first frame. Your text only needs to describe what happens after that, not what's already visible.

If you don't have a reference image ready, OpenAI's own Sora 2 prompting guide points out that its image generation model works well for creating one, since the image anchors the first frame while the text prompt defines the motion that follows.

Example prompt paired with a still image of a woman standing on a rooftop overlooking a skyline:

"She turns around slowly and smiles, then walks toward the edge of the rooftop, the wind moving through her hair."

Notice this doesn't redescribe her outfit, the skyline, or the lighting. That's already locked in by the image. The text just adds the movement.

Sora prompts for product videos

For commercial or marketing use, clarity beats creativity almost every time. Buyers don't care about artistic camera work; they want to see the product clearly.

• "Close-up shot of a glass perfume bottle rotating slowly on a white turntable, soft studio lighting, clean reflections on the surface."
• "A pair of running shoes on a treadmill, camera at floor level, dynamic side lighting highlighting the texture of the fabric."
• "A smartphone lying on a wooden desk, the screen lighting up to display a notification, soft window light from the side."

Keep backgrounds simple and lighting even unless the brand specifically wants a moody, stylized look.

Sora prompts for cinematic videos

Cinematic prompts can afford more atmosphere, but they still need a single, clear action. Trying to cram an entire scene's plot into one prompt almost always backfires.

"A detective stands in a rain-soaked alley at night, lit only by a flickering neon sign overhead. He lights a cigarette, the flame briefly illuminating his face. Wide shot, shallow focus, noir color grading."

That's one beat, one camera angle, one lighting setup. If you need more story than that, you're better off generating it as separate clips and editing them together rather than asking for it all in a single prompt.

Useful Sora prompt keywords you can try 

If you're not sure how to describe a scene, certain keywords can help guide the model toward a specific look or feel. You don't need to use all of them in every prompt. Pick the ones that match the result you're trying to create.

Camera & framing

• Close-up
• Extreme close-up
• Medium shot
• Wide shot
• Overhead shot
• Low-angle shot
• High-angle shot
• Tracking shot
• Handheld camera
• Static tripod shot
• Drone shot

Lighting

• Golden hour
• Soft daylight
• Overcast lighting
• Warm indoor lighting
• Candlelit
• Neon-lit
• Backlit
• Volumetric lighting
• High contrast lighting
• Diffused light

Visual style

• Cinematic
• Documentary
• Photorealistic
• Anime-inspired
• Noir
• Vintage film
• Minimalist
• Futuristic
• Realistic
• Stylized

Weather & atmosphere

• Light rain
• Heavy fog
• Snowfall
• Misty morning
• Clear skies
• Dusty air
• Storm clouds
• Windy conditions

Quality & detail

• Shallow depth of field
• Soft bokeh
• Sharp focus
• Highly detailed
• Natural motion blur
• Realistic textures
• Clean background
• Crisp details

These keywords work best when combined with a clear subject and action. Instead of writing "a cinematic city," try describing exactly what the camera sees: "A futuristic city street at night, neon-lit signs reflecting on wet pavement, wide shot, cinematic style."

Copy-paste Sora prompt templates

Here are a few fill-in-the-blank Sora prompt template options based on the structure above. Swap the bracketed parts for your own details.

General template:
"[Subject description] [does one specific action] in [setting, time of day, weather]. [Camera type and movement]. [Lighting description], [style reference]."

Product template:
"[Product] [positioned how, on what surface]. [Camera angle, static or slow movement]. [Lighting setup], clean background, [mood, e.g., minimal/luxury/playful]."

Talking character template:
"[Character description] [does a small action] in [setting]. [Camera framing].

[Character] says: '[exact line]'"

Image-to-video template:
"[Subject from the reference image] [does one new action]. [Camera movement if needed]."

Run a few versions of each through small tweaks before settling on a final clip. Even a single adjective shift in lighting or action can change the output more than you'd expect.

‍

10 ready-to-use example prompts

If you're not sure where to start, use these examples as inspiration. You can copy them as-is or swap out the subject, setting, lighting, and camera directions to create your own variations.

1. Rainy city street

A woman in a yellow raincoat walks down a narrow city street at night as rain falls steadily. The camera tracks alongside her at a walking pace. Neon signs reflect on the wet pavement. Cinematic style, shallow depth of field.

2. Mountain sunrise

A lone hiker stands on a mountain ridge overlooking a valley at sunrise. Wide shot. The camera slowly pushes forward as golden light spreads across the landscape. Realistic, highly detailed.

3. Cozy coffee shop

A barista prepares a cup of coffee behind a wooden counter in a cozy café. Medium shot. Warm indoor lighting. Soft focus in the background. Documentary style.

4. Beach at sunset

A golden retriever runs along a quiet beach as waves roll onto the shore. Low-angle tracking shot. Warm golden-hour lighting. Cinematic style.

5. Product showcase

A luxury wristwatch rotates slowly on a black pedestal. Close-up shot. Studio lighting creates clean reflections on the metal surface. Minimalist background.

6. Forest trail

A cyclist rides through a dense forest trail surrounded by tall pine trees. Handheld camera follows from behind. Soft morning light filters through the branches. Natural, realistic look.

7. Futuristic city

A futuristic city skyline glows with neon lights on a rainy night. Aerial drone shot moving slowly between skyscrapers. Reflections shimmer on the streets below. High-detail sci-fi aesthetic.

8. Wildlife scene

A deer stands beside a calm forest lake at dawn. Static wide shot. Light mist drifts across the water as sunlight begins to break through the trees. Photorealistic style.

9. Restaurant kitchen

A chef plates a gourmet dish in a busy restaurant kitchen. Close-up shot focusing on the final touches. Warm lighting highlights the colors and textures of the food. Documentary style.

10. Fantasy landscape

A wizard stands on a cliff overlooking floating islands suspended above the clouds. The camera slowly circles around him. Soft sunrise lighting. Epic fantasy atmosphere, highly detailed.

These examples follow the same formula used throughout this guide: subject, action, setting, camera, lighting, and style. Once you're comfortable with the structure, you can mix and match elements to create an almost endless variety of scenes.

Why does Sora ignore your prompt? Common fixes

This is one of the most searched frustrations with the model, and it usually comes down to a handful of repeat issues rather than the model just being unreliable. If you've typed something specific and still ended up asking yourself why does Sora ignore my prompt, the answer almost always comes down to one of the points below.

A quick context note before the list: most of these fixes take less than a minute to apply, so it's worth running through them before assuming the prompt itself is the problem.

• Too many actions in one prompt. If you ask for a character to walk, turn, wave, and speak all in a four-second clip, the model will pick and choose what to render. Stick to one main action per short clip.
• Vague descriptors instead of specifics. Words like "beautiful," "epic," or "amazing" don't give Sora anything concrete to work with. Replace them with details you could actually point a camera at.
• Conflicting instructions. Asking for "slow motion" and "fast walking" in the same prompt creates a contradiction the model has to resolve on its own, usually badly.
• Missing camera direction. When you don't specify a camera movement, Sora picks one for you, and it's often not what you pictured.
• Clip length mismatch. Longer clips ask the model to maintain consistency for more time, which increases the odds of drift. Shorter, well-defined clips tend to follow instructions more reliably than long, ambitious ones. Developers working through the cookbook note that the model generally sticks closer to instructions in shorter shots, and that stitching two short clips together in editing can sometimes beat asking for one long uninterrupted one.
• Forgetting that unstated details get invented. If you don't specify the time of day, the outfit, or the camera angle, the model fills in something plausible but not necessarily what you wanted. Be exhaustive with the details you actually care about.

The examples below show some of the most common prompting mistakes and how to fix them.

When a prompt isn't working, simplify it first, then add details back one layer at a time.

If you've tried all of the above and you're still getting inconsistent results, it might not be a prompting problem at all. It could be a model limitation, which we'll get to next.

Limitations and prompts Sora won't generate

No guide to writing prompts is complete without being honest about what doesn't work, because writing a "perfect" prompt won't fix a limitation baked into the model itself.

A quick note before the list: these aren't bugs; you can prompt your way around. They're structural limits of how the model handles certain kinds of footage.

• Physics in fast or chaotic motion. Crowds, collisions, water splashes, and rapid object interactions tend to look slightly off, no matter how the prompt is worded.
• Long, continuous sequences. The longer the clip, the more likely details drift partway through, especially with background elements or secondary characters.
• Fine detail consistency. Hands, small text, and intricate patterns are common trouble spots across most text-to-video models, not just this one.
• Content that violates usage policies. Prompts depicting real public figures without consent, graphic violence, sexual content, or anything infringing on copyrighted characters will get blocked or filtered, regardless of how the prompt is structured.
• Exact recreation of existing footage or branded content. Asking for a scene "exactly like" a specific movie or show typically won't produce a usable result and may run into content restrictions anyway.

None of this means the tool isn't capable of strong results. It just means some ideas need a phased approach, shorter clips stitched together, or a different tool entirely for certain kinds of motion.

‍

Is Sora still available in 2026? Current status update

This is the part that's easy to miss if you're reading older guides, including the official one, and it matters a lot if you're planning to rely on this tool going forward. According to OpenAI's own Help Center, the Sora web and app experiences were discontinued on April 26, 2026. That means sora.com and the standalone mobile apps are no longer accessible for generating new content. 

If you have videos saved in your account, OpenAI's guidance recommends exporting them as soon as possible, since data will be permanently deleted after the discontinuation period and any final export window passes. 

The API has a longer runway, but it's still on a fixed clock. The Sora API is scheduled to be discontinued on September 24, 2026. Until that date, developers building on the API can keep generating videos through it, but it's worth planning a migration path now rather than waiting until the deadline. 

So, where does that leave the prompting techniques in this guide? Still useful, for a few reasons. The underlying logic behind structured, specific prompting carries over to nearly every text-to-video model on the market, not just Sora. 

If you're using the API while it's still running, none of the above changes. And if you're researching this topic to understand how prompt-based video generation works in general, the formula holds regardless of which specific tool ends up replacing it.

Final thoughts

Writing better Sora prompts comes down to treating the model less like a creative collaborator and more like a camera operator who needs clear, literal instructions. Subject, action, setting, camera, lighting. That order, kept simple, fixes most of the common frustrations people run into.

The bigger thing to keep in mind going into the back half of 2026 is timing. Whether you're working through the API while it's still live or applying these same prompting habits to whatever tool comes next, specific beats clever every time. That part isn't going away anytime soon.