Sora 2 Prompting Guide: How to Use Sora 2 (2026)

Write the prompt: the three-part formula

1. 1

   Open with a scene intent, not a subject

   Every good Sora 2 prompt starts with one to three sentences of scene intent — you are briefing a cinematographer, not filing a search query. The walkthrough's example names the style first, then the subject and the setting: "High-quality 2D anime film style, cinematic realism. A samurai woman with long white hair walks slowly along a coastal path beside the ocean... Wooden fences line the path as waves crash softly below, creating a calm yet powerful atmosphere." Style, character, location and mood, in that order, give Sora everything it needs before you direct the shots.

   

2. 2

   Direct the opening beat with a timestamp

   Shot composition is where you direct. Sora 2 accepts timestamp control, so a segment like "[00:00-00:03] The woman continues walking slowly forward along the path. The camera follows smoothly behind her at waist height" tells the model exactly what the first three seconds contain — down to wind lifting her red cloak and petals drifting past her feet. Sora generations run 10 seconds on the free plan and 12 on Pro, so divide your runtime into three or four beats.

   

3. 3

   Give each segment its own camera move

   Keep the timestamps coming and change the camera in each one. The next beat reads "[00:03-00:06] The camera transitions to a low side angle as she slows her pace", with her hand brushing the wooden fence and leaves tumbling across the path. One timestamp equals one shot: framing, camera angle and blocking all live inside the brackets, which is what stops Sora from drifting into a single static take.

   

4. 4

   Put dialogue inside the moment it is spoken

   Spoken lines belong in the timestamp where they happen. In the walkthrough's third beat, the card reads "[00:06-00:09] The camera moves closer into a medium shot... She whispers quietly, 'I'll return someday'" — the whisper sits highlighted in blue exactly where the medium shot lands. Place the line where you want to hear it and Sora syncs the delivery to that moment instead of guessing.

   

5. 5

   Try cuts instead, and place sound effects in their cut

   Sora 2 also accepts cut-based control — Cut 1, Cut 2, Cut 3 — and times the cuts itself. Sound design follows the same placement rule: the walkthrough drops "a distant ship horn sounds once, deep and resonant" straight into Cut 2 next to the wave crash, because that is the moment it should sound. Save ambient audio for the lighting-and-tone paragraph; specific effects go where they happen.

   

6. 6

   Close with a lighting and tone paragraph

   End every prompt with one paragraph that defines the look: lighting, palette, atmosphere and emotional tone. The walkthrough's closer — "The lighting is warm and cinematic, with late-afternoon sunlight glinting off the ocean surface. The sound of distant waves and wind completes the tranquil, reflective feeling of the scene" — ties the shots together and carries the background audio. Phrases like golden hour or soft shadows earn their keep here.

   

Let ChatGPT draft it, trained on OpenAI's own guide

1. 7

   Start from OpenAI's official Sora 2 Prompting Guide

   You do not have to memorize any of this. OpenAI publishes an official Sora 2 Prompting Guide on its cookbook site covering the same ground: precise subjects, camera setup, lighting and palette, action beats, tone and mood. The Artlist walkthrough leans on it for every prompt in the video — and it works even better as a ChatGPT training document, which is the next step.

   

2. 8

   Teach ChatGPT the guide with one instruction

   Paste the cookbook URL into ChatGPT with a single line: "analyze this page ... and use it to build prompts for sora 2". The model reads the official guide, pulls out its key patterns, and from then on writes prompts in Sora's own vocabulary — storyboard-style beats, one clear subject per shot, API parameters kept outside the prose. It is a one-time setup you can reuse for every clip you plan.

   

3. 9

   Brief it like a director, not a typist

   Now describe the video you want in plain words: "Create a 12 second Sora 2 prompt that will result in a super cinematic scene of a man flying over a field, similar to these photos. It should start with the man walking in the field and then start hovering and finally fly." Reference images add context for free, and ChatGPT offers extras such as a 3-shot storyboard version split into separate clips for editing.

   

4. 10

   Get back a timecoded prompt you can edit

   What comes back is a production document: "duration: 12s, aspect_ratio: 2.39:1, style: ultra-cinematic photoreal, dusk palette" at the top, a character reference locked to your photos, then timecoded beats — 0.0-2.5s establish walk, 2.5-3.5s transition hover, 3.5-11.5s flight with camera notes down to the lens. Edit any beat by hand; you approved the story, ChatGPT just typed the jargon.

   

Generate: from prompt box to finished clip

1. 11

   Paste it in and pick Sora 2 Pro, 12 seconds, 1080p

   In the generator — the walkthrough uses Artlist, which hosts Sora 2 and Sora 2 Pro next to models like Veo and Kling — paste the prompt, select the Sora 2 Pro model, and set 16:9, 12 seconds and 1080p before hitting Generate. Those settings mirror the prompt header, so the timecodes land where you planned. The same paste-and-configure flow applies in OpenAI's own Sora app.

   

2. 12

   Prepare a reference image for image-to-video

   Image-to-video gives you more control than text alone, because the model inherits the character, wardrobe and style from a still. Generate one in the image tab of the same panel, then refine it with an instruction like "place this character on a futuristic motorcycle, riding through a dystopian barren land, motion blur to emphasize speed, full body, wide shot". That edited still becomes the first frame your prompt directs.

   

3. 13

   Write image-to-video beats that continue the still

   Prompting from an image changes the verbs. Instead of describing what exists — the picture already shows that — you describe what happens next: "[00:00-00:03] The woman gently shifts her weight, then begins walking a few slow steps toward the fence. The camera holds roughly the same composition, tracking slightly forward." The opening beat moves the subject from stillness into motion while the framing holds.

   

Read the result, then remix

1. 14

   Read the result against the beats you wrote

   The finished clip is the brief, beat by beat: the man walks through the dusk field, hovers, and finally flies low over the wheat while the camera tracks him. Judge your own results the same way — did each timestamp deliver its shot? Even this production-grade prompt needed several regeneration rounds in the source video, so treat a first render as a draft, not a miss.

   

2. 15

   Expect real dialogue scenes, not just b-roll

   The same formula produces acting. The dialogue demo in the source video — two people, a lamp-lit apartment, lines like "You think an apology fixes everything?" — shows Sora 2 handling conversation, emotion and staging inside a single take. Name who speaks and how the line lands, exactly as the timestamp and cut examples showed, and the scene plays it back.

   

3. 16

   Iterate with a shot-by-shot remix request

   When a result is close but not right, refine in the same chat: "I need a 12 second sora 2 sequence of this character, riding a futuristic grungy motorcycle... natural motion blur, tracking shot with natural camera shake". ChatGPT offers a shot-by-shot remix version, and targeted notes — more speed, an intentional stop, a camera orbit — fold into the next prompt. You will never get the same clip twice, so regenerate before you rewrite.

   

Frequently asked questions