Sora 2 Prompt Generator

What is Sora 2?

Sora 2 is OpenAI’s second-generation video model. You describe a scene in words and it generates a short video clip, and unlike many earlier text-to-video systems it can also produce matching audio such as dialogue, ambient sound and effects. It is available through OpenAI’s Sora app and, for developers, through the OpenAI API. Exact clip lengths, resolutions and availability depend on your plan and region and change over time, so check OpenAI’s current documentation for the specifics.

Because Sora 2 understands fairly long, natural descriptions, the quality of your prompt has a big effect on the result. Vague prompts such as “a cool city video” leave too many decisions to the model. Specific prompts that read like a director’s shot notes tend to produce clips that are more consistent, better framed and closer to what you imagined.

A prompt structure that works well for Sora 2

You do not need special syntax. What helps most is giving information in a predictable order, so nothing important gets buried. The generator above follows this structure for you:

1. Subject and action: who or what is on screen and exactly what they do. One clear action per shot beats three competing ones.
2. Setting and time: location, era, weather and time of day. These details drive the overall look.
3. Framing and lens: wide, medium or close-up, camera height, and a lens feel such as “shallow depth of field” or “wide-angle”.
4. Camera movement: static, slow push-in, tracking, orbit, handheld. Name one main move.
5. Lighting and color: golden hour, overcast, neon, high contrast, a warm or cool palette.
6. Style: photorealistic, 35mm film, anime, stop-motion, commercial. Pick one and stay consistent.
7. Audio: dialogue in quotes with the speaker named, plus ambience and one or two sound effects.

Sora 2 prompt examples

Each example below follows the same structure. Paste them as they are, or use them as a starting point and change the subject.

Night market in Bangkok, light drizzle. A middle-aged street food vendor tosses noodles in a flaming wok; the flames briefly light his focused face. Medium shot at chest height, 35mm lens feel, shallow depth of field with colorful bokeh from string lights behind him. Slow push-in toward the wok. Warm orange firelight against cool blue rain. Photorealistic, cinematic color grade. Audio: sizzling wok, rain on tarp roofs, distant scooter horns, a customer laughs off-screen.

A lone astronaut sits on the edge of a crater on the moon, Earth glowing low on the horizon. Wide shot, then the camera slowly dollies in to a medium shot of the visor reflecting Earth. Harsh, single-source sunlight with deep black shadows. Realistic, quiet, slightly humorous tone. Audio: soft suit breathing, faint radio static; the astronaut says in a deadpan voice, "I can see my house from here."

Vertical 9:16 product shot. A glass perfume bottle slowly rotates on a slab of wet black stone; water droplets bead and roll down the glass. Macro close-up, static camera, the bottle fills the center third of the frame. Soft overhead softbox with a thin rim light tracing the bottle’s edges, deep black background. Luxury commercial style, crisp detail. Audio: gentle water drips and a low, airy synth tone.

Tips for better Sora 2 videos

• check_circleMatch the action to the clip length. A few seconds is enough for one gesture or one camera move, not an entire story.
• check_circleKeep characters consistent by describing them the same way every time: age range, clothing, hair, one distinctive detail.
• check_circleUse cinematic vocabulary the model has seen in film descriptions: “rack focus”, “over-the-shoulder”, “low-angle”, “golden hour”.
• check_circleFor dialogue, keep lines short and natural. One or two sentences per clip are easier to deliver convincingly than a monologue.
• check_circleGenerate a few variations. Small wording changes can give noticeably different results, so treat your first prompt as a draft.
• check_circleChoose vertical 9:16 for Reels, Shorts and TikTok, and 16:9 for YouTube or presentations.

Mistakes to avoid

• check_circleStacking styles. “Anime, photorealistic, watercolor, 3D” confuses the look. Choose one.
• check_circleToo many camera moves in one shot. “Pan left, then crane up, then zoom” in a short clip usually produces jittery motion.
• check_circleAsking for readable on-screen text. Short words sometimes work, but long text, logos and small print are unreliable in most video models.
• check_circleUsing real people’s names or copyrighted characters. OpenAI’s policies restrict these, and such prompts are often refused.
• check_circleWriting instructions instead of descriptions. “Make it go viral” tells the model nothing; “a quick, energetic handheld shot with a surprise reveal” does.

Text or JSON prompts?

Sora 2 reads plain English well, so text prompts are the default here. JSON output is useful when you want to keep a library of prompts, swap one field at a time (for example only the lighting), or feed prompts into a script or API workflow. If you want more control over structured fields, try our JSON Prompt Generator. If you already have a clip you like, the Video to Prompt tool can describe it as a prompt you can adapt for Sora 2.

Frequently asked questions