Sora 2 Prompt Generator: What's New in Prompt Structure

A Sora 2 prompt generator turns a rough video idea into the prompt shape OpenAI's mid-2026 Videos API expects: prose for subject and motion, a separate dialogue or sound block when you need synced audio, and clear notes for settings that belong outside the text. You stop stuffing duration, resolution, and character IDs into adjectives. You leave with the new structure mapped field by field, a generator loop that fills those fields, and a worked example you can paste after a two-minute edit. Soft path: PromptMake /text at https://promptmake.net/text when you want a first draft of that scaffold without writing every label by hand.

What a Sora 2 prompt generator is

A Sora 2 prompt generator accepts plain language about a clip. It returns a structured draft aimed at OpenAI's Sora 2 line (sora-2 for faster iteration, sora-2-pro for higher-fidelity lanes on supported plans). The draft separates what belongs in prompt prose from what belongs in the request container: model, size, seconds, character IDs, and optional image references.

You stay the editor. The generator drafts the anatomy. You add product names, banned logos, speaker labels, and the live route you will paste into (API, ChatGPT video surfaces that still expose Sora-family models, or another host).

This page targets people who already know shot lists help video prompts, and who now need the Sora 2-specific layout: audio blocks, character reuse, extend and edit prompts, and Pro resolution ladders. If you still need the basics of shot IDs, one camera move per block, and continuity anchors, use the companion Sora prompt generator guide on shot lists and motion. Keep this article for structure that arrived with Sora 2.

A thin input looks like this: "Make a Sora 2 ad with music." The model then invents length, speakers, and camera path. A generator-ready input looks like this: "8-second kitchen hero for a matte black kettle. Slow push-in. Thin steam. Soft window light. Diegetic room tone only. No logos. Character ID for kettle mascot if we upload one." The generator maps that into labeled prose plus a checklist of container fields you set in the API or UI.

PromptMake /text sits in the free drafting lane. Guests get about three generations per day. Free accounts get about five. Ask for a Sora 2-ready prompt with cinematography, actions, dialogue or sound, and a short list of settings to set outside the prose. Soft link: https://promptmake.net/text.

What's new in Sora 2 prompt structure (mid-2026)

OpenAI's March 2026 Sora 2 prompting guidance treats the prompt as a creative brief and treats model, size, seconds, and character references as request parameters. Prose that says "make it 4K for 30 seconds" does not override those controls. That split is the biggest structural change you must teach a Sora 2 prompt generator, and it is the change most people skip when they reuse old still-image or early Sora habits.

As of mid-2026, the public cookbook also documents synced audio inside the prompt, reusable character assets from short reference clips, image input for a first-frame lock, video extension that uses the full prior clip as context, and edit flows for one-change tweaks. Consumer app buttons and API timelines still move. Confirm live model IDs and access in OpenAI docs before you ship a production pipeline. The craft below stays useful across those surface changes.

Container fields vs prompt prose

Set these in the API call or UI, not in wishful adjectives: model (sora-2 or sora-2-pro), size (for example 1280x720 on standard lanes; Pro adds ladders such as 1920x1080 and 1080x1920), seconds (common buckets: 4, 8, 12, 16, 20; default often 4), and optional characters with up to two uploaded character IDs. Optional image input must match the target size when you use a first-frame reference.

A Sora 2 prompt generator should output a short "set outside prose" checklist under the draft. Example: "Model: sora-2. Size: 1280x720. Seconds: 4. Characters: none." You copy the checklist into the request. You keep the prose on subject, action, camera, light, palette, and sound.

Shorter clips still follow instructions with fewer failures. Two clean 4-second shots often beat one overloaded 8-second wish. Pro resolution helps detail; it does not fix a prompt that stacks three camera moves.

Audio, dialogue, and the block layout

Sora 2 generates video with synced audio. Dialogue belongs in the prompt as its own block under the visual description so the model can separate spoken lines from scenery. Keep lines short. A 4-second shot fits one or two brief exchanges. An 8-second clip can hold a few more. Long speeches break timing.

Label speakers the same way every time: "Detective:" and "Suspect:" on alternating turns. For silent product shots, write one diegetic cue such as "thin steam hiss, quiet fridge hum" instead of asking for a full score. Generators that bury dialogue inside a single mood paragraph make lip motion and gesture harder to match.

OpenAI's descriptive template that works well in mid-2026 looks like this shape: prose scene, then Cinematography (shot, mood), then Actions as beats, then Dialogue if needed, then Background Sound. A Sora 2 prompt generator should fill those labels even when some stay empty.

Characters, image reference, extend, and edit

Character reuse: upload a short MP4 (about 2–4 seconds, 720p–1080p, 16:9 or 9:16) through the Characters API to get a character ID. Animals and objects work. Pass up to two IDs in the request and name each character in the prompt the same way every generation. The ID lives in the container; the name lives in the prose.

Image reference: attach a JPEG, PNG, or WebP that matches the target size. The image anchors the first frame. Your prompt describes what happens next. Generators should remind you to match resolution instead of writing "use my photo" with no upload plan.

Extend: continue a completed clip with the full prior video as context, not only the last frame. Individual extensions can add up to about 20 seconds; OpenAI documents up to six extensions for longer totals. Write the next beat as a continuation prompt that repeats wardrobe, light, and palette anchors.

Edit: nudge one variable at a time ("same shot, switch palette to teal and sand"). Prefer edit for new workflows; treat legacy remix as compatibility where it still exists. A generator that outputs a full rewrite when you only need a color change wastes credits.

How a Sora 2 prompt generator builds the new structure

Treat the generator as a preflight desk for Sora 2. You bring the story beat. The tool adds missing anatomy: style line, cinematography, timed actions, dialogue or sound, constraints, and a container checklist. Then you tighten brand facts before you spend a video credit.

People who bounce after one messy Sora 2 clip often pasted a mood paragraph that begged the prose to set length and resolution. They blamed the model when the clip ignored "make it longer." The sections below fix that with the fields a Sora 2 prompt generator should fill, and a short review pass aimed at OpenAI's mid-2026 guidance.

Keep the first dump messy on purpose. Polish belongs after generation. A half-formed beat list yields a better scaffold than a stiff paragraph you rewrote three times for elegance the tool will rebuild anyway.

Fields the generator should always attempt

Style and look: format feel, lens family, grade, grain. Scene prose: who, where, weather, props. Cinematography: one framing, one camera move, depth of field if it matters. Actions: timed beats, not stacked verbs. Dialogue or Background Sound: one block, labeled. Constraints: no logos, no on-screen text, no extra hands. Container checklist: model, size, seconds, characters, image reference yes/no.

Ask the tool to leave unused slots blank instead of inventing speakers for a silent product shot. Empty Dialogue with a clear Background Sound line beats fake conversation that forces mouth motion you never wanted.

Fields you must add by hand

Brand-critical nouns: exact product finish, pantone-adjacent colors you care about, legal bans, must-show props. Character names that match uploaded IDs. Real speaker names for multi-person scenes. Route notes: which UI or API path you will use this week, because consumer surfaces have shifted through 2026.

Secrets stay out of public generators when policy forbids third-party paste. Strip customer faces, unreleased packaging, and internal codenames if you must test with real scenarios.

Step-by-step: rough idea to Sora 2-ready prompt

Run this loop when you want the mid-2026 structure instead of a lucky paragraph. The generator handles labels. You handle container settings, character uploads, and the final one-change edit plan. Expect one generate pass plus a two-minute edit before you spend a video credit.

Access note: confirm the live route you will paste into before you judge a prompt as weak. Model IDs, app availability, and API phases change. This article stays on structure. Shot-list craft without Sora 2 audio and character fields belongs in the companion Sora prompt generator guide.

Step 1: Dump the idea with audio and container intent

Write the job in plain words. Name subject, setting, one camera move, sound intent (silent / diegetic / short dialogue), and duration bucket you plan to set. Example: "Matte black kettle on oak counter. Morning window light. Slow push-in only. Thin steam in the last second. Quiet kitchen tone. No speech. Plan: sora-2, 1280x720, 4 seconds."

Skip persona theater. Do not invent "You are an award-winning DP." Outcome, subject, boundaries, and sound plan beat ego lines. If you already hold a character ID, write the name and ID on the same line so the generator can place each in the right layer.

Step 2: Generate the Sora 2 scaffold

Open a Sora 2 prompt generator or a general text enhancer and ask for the labeled template: scene prose, cinematography, actions, dialogue or sound, constraints, plus a container checklist. On PromptMake /text, paste the rough idea and state that you need a Sora 2-ready structure with synced-audio blocks and settings listed outside the prose. Soft link: https://promptmake.net/text.

Read the output once. Check that seconds and size are listed as checklist items, not buried as adjectives. Check that dialogue sits in its own block. Check that character names match any IDs you will pass. Guest use needs no signup for a small daily quota. Register if you hit the cap mid-project.

Step 3: Edit for two minutes, set the container, generate

Scan for four things: prose trying to set duration or resolution, missing sound block, stacked camera verbs, and character names that do not match uploads. Cut "cinematic masterpiece 8K" stacks. Add real prop nouns. Split any shot that tries two camera moves.

Set model, size, and seconds in the request. Attach characters or image reference if you planned them. Generate the shortest length that fits the beat. If the clip is close, use edit with one change. If you need more time, extend with a continuation prompt that repeats anchors instead of rewriting the whole world.

Save the final prose and the container checklist next to a shot ID. Label the note with model lane and seconds. A strong prompt on the wrong size still looks weak.

Worked example: kettle reveal with Sora 2 structure

Rough idea: "Product video for a matte black pour-over kettle. Morning kitchen. Calm. Soft sound. Maybe a tiny steam hiss."

Generator-style Sora 2 draft after one pass:

Scene: Sunlit kitchen counter. Matte black pour-over kettle centered on an oak board, brushed handle, no logo. Soft morning window light from camera left. Dust motes in the beam. Quiet room.

Cinematography: Medium shot, eye level. Slow dolly in only. Shallow depth of field on the kettle body. Mood: calm, premium, still.

Actions: 0–3s kettle holds in frame as the camera creeps closer. 3–4s thin steam rises from the spout tip.

Dialogue: (none)

Background Sound: Quiet kitchen tone, faint steam hiss in the final second. No score.

Constraints: No people, no on-screen text, no brand logos, no hand entering frame.

Set outside prose: Model sora-2. Size 1280x720. Seconds 4. Characters none. Image reference none.

Your two-minute edit: add the real wood tone from brand photos, ban any logo reflection, and keep Dialogue empty. Generate. If steam looks wild, edit with "same shot and camera; steam only in the last half-second, thinner." If you later need a mascot object, upload a 2–4s reference clip, pass the character ID, and rewrite the scene line to name that character while the checklist gains the ID.

That loop took one generator run plus one edit. You filled brand facts the tool could not know, and you kept container fields out of the prose.

Mistakes that waste Sora 2 generator runs

Mistake 1: Asking prose to set model, size, or seconds. Put those in the request. Keep the prompt on subject, motion, light, and sound.

Mistake 2: Mixing dialogue into scenery sentences. Give Dialogue its own block with labeled speakers.

Mistake 3: Naming a character in prose without passing the character ID, or passing an ID without using the same name in the prompt.

Mistake 4: Stacking camera moves and subject verbs in one 4-second wish. One camera path, timed beats.

Mistake 5: Attaching an image reference that does not match the target size. Match resolution first.

Mistake 6: Regenerating from scratch when an edit with one change would keep the good parts.

Mistake 7: Treating this page as a rewrite of the legacy Sora prompt guide. Legacy craft covers hand-written shot types and examples. The companion Wave 2 guide covers shot lists, motion, and continuity generators. This page covers Sora 2 structure: container split, audio blocks, characters, extend, and edit.

When to use a Sora 2 prompt generator (and PromptMake /text)

Use a Sora 2 prompt generator when you need the mid-2026 anatomy filled before you open the Videos API or a ChatGPT video surface. Use it for product heroes with diegetic sound, short dialogue spots, character-reuse campaigns, and extend chains where each next beat must inherit anchors.

Pick sora-2 while you iterate on structure. Move a locked prompt to sora-2-pro when you need higher resolution ladders and you already trust the prose. Confirm live sizes and seconds in OpenAI docs; ladders and quotas can shift by plan.

PromptMake /text drafts the labeled structure from a rough idea so you spend attention on brand nouns and container settings. Soft link: https://promptmake.net/text. Guests get a small daily quota without signup. Free accounts raise that cap. Measure success by a usable clip after one edit pass, not by how long the prompt looks.

If you outgrow free drafting quotas, register for a higher daily cap. Until then, three careful Sora 2 scaffolds beat thirty unedited mood paragraphs.

FAQ

What is a Sora 2 prompt generator?

It is a tool that turns a rough video idea into a structured prompt shaped for OpenAI Sora 2, including prose blocks, audio or dialogue layout, and a checklist for container settings. You type intent in plain language. You get a draft you edit, then paste into your live Sora 2 route. Free tiers limit daily runs; paid tiers raise those caps.

How is Sora 2 prompt structure different from older Sora habits?

Sora 2 expects you to set model, size, seconds, and character IDs outside the prose, and to place dialogue or diegetic sound in clear blocks under the visual description. Older habits buried length and resolution inside adjectives and mixed speech into scenery. Mid-2026 OpenAI guidance treats the request as a container and the prompt as creative direction.

Should a Sora 2 prompt generator write dialogue in a separate block?

Yes when the clip has spoken lines. Put Dialogue under the scene description with labeled speakers and short turns that fit the seconds you set. For silent product work, leave Dialogue empty and write one Background Sound cue. Generators that hide speech inside mood prose make timing harder to match.

How do character references change the prompt?

You upload a short reference clip to create a character ID, pass up to two IDs in the request, and use the same character names in the prompt text. The ID belongs in the container. The name belongs in the prose. Animals and objects can become reusable characters under OpenAI's mid-2026 Characters API rules.

Can I start on PromptMake without signing up?

PromptMake /text allows a small guest quota each day with no account. Create a free account if you need more daily runs. Soft link: https://promptmake.net/text. Ask for a Sora 2-ready scaffold with cinematography, actions, sound, and a container checklist, then paste the edited result into your live route.

Is this the same as the Sora prompt generator guide on shot lists?

No. The shot-list guide covers generator craft for shot IDs, single-move cameras, and continuity anchors. This Sora 2 prompt generator article covers what is new in prompt structure: container versus prose, audio blocks, characters, image reference, extend, and edit. Bookmark both if you want craft plus Sora 2 layout.

Does this replace the legacy Sora prompt guide?

No. The legacy guide teaches hand-written video prompts with shot types and copy-paste examples. This page teaches a generator workflow for Sora 2 structure as of mid-2026. Keep the legacy bookmark for craft examples; use this page when you need the new field layout and container checklist.