H3 Max Prompt Lab

Nano Banana Works is our sister publication. References to its still-image workflows below are owned-site recommendations.

Updated 21 September 2026 · camera-language remap · not a vendor listicle

Sora API Sunset → Seedance / Kling / Veo Prompt Remap

OpenAI is not shipping a Sora successor. The consumer app already went dark; the Videos API and every sora-2* alias hard-stop on 24 September 2026. This page translates Sora-style shot language into copy-paste briefs for Seedance 2.5, Kling, and Veo — unofficial, camera-first, not a “best alternatives” roundup.

Hard stop · 24 Sep 2026 · Videos API + sora-2*

Export now. There is no OpenAI replacement model.

The Sora web and app experiences were discontinued on 26 April 2026. The API window closes on 24 September 2026. OpenAI’s own deprecations table leaves the recommended replacement column blank for the Videos API and every Sora 2 alias — migrate vendors, then remap the camera language. Export leftover Sora app content at sora.chatgpt.com/sunset. Official dates and export steps: OpenAI Help — What to know about the Sora discontinuation.

Unofficial remap pack

This page is independent. It is not affiliated with OpenAI, ByteDance, Seedance, Kuaishou, Kling, Google, Veo, MiniMax, Hailuo, or fal. Remaps below are craft notes — not official templates, SDKs, or guaranteed output. Host UIs, duration caps, audio flags, and reference slots change; verify on the playground you actually pay before you spend credits.

App vs API: what dies when

Two different products, two dates. Do not treat “Sora is gone” as one event.

Dates from OpenAI Help Center and the API deprecations page (checked September 2026). Replacement column is blank because OpenAI published none.
Surface What stops When OpenAI replacement
Sora app + web Consumer Sora experiences (app / sora.com / ChatGPT Sora UI) 26 Apr 2026 — already discontinued None. Export via sora.chatgpt.com/sunset
Videos API /v1/videos create / list / retrieve / remix jobs 24 Sep 2026 — hard stop —
Model aliases sora-2, sora-2-pro 24 Sep 2026 —
Dated snapshots sora-2-2025-10-06, sora-2-2025-12-08, sora-2-pro-2025-10-06 24 Sep 2026 —

Deprecation notice to developers: 24 March 2026. After the app sunset and any final export window, OpenAI says it will permanently delete Sora-associated account data. Credits from ChatGPT/Sora purchases are not a video-model migration path.

If you call Sora through Azure / Microsoft Foundry, OpenAI’s 24 Sep date is not your clock. Foundry sora-2 version 2025-12-08 is listed to retire 15 October 2026. Dual-clock table: Azure Foundry Sora-2 retirement.

What actually changes in the prompt

Pasting a Sora paragraph into another vendor usually fails for language reasons, not “which model is best.” Sora-style briefs simulate a world and let the camera follow physics. Seedance, Kling, and Veo want directed camera events.

  1. Split the story. Sora often swallowed a 15–20s narrative in one block. Seedance wants timestamped beats; Kling wants a short shot list; Veo wants one audiovisual take (often ~8s).
  2. Name the camera move and its trigger. “Cinematic, photoreal” is not a move. “Slow dolly-in as the door opens” is.
  3. One move per beat. Dolly + orbit + whip in the same sentence is how Sora prompts used to read. Remap to one path, then cut.
  4. Lock identity with stills, not adjectives. Sora cameos / remix IDs do not travel. Attach references; say what each still controls. For stills, start at the Nano Banana Works posts hub.
  5. Declare audio. Sora assumed native sound. Say diegetic FX vs dialogue vs no music. Host-specific dialogue brackets vary — use the field your UI exposes.
  6. Cause → effect → aftermath for physics (rain, splash, crash). Do not jump from setup to result.

Ceremony beats (first look, vows, orbit dance): AI wedding video prompts. CCTV / freeze-orbit / product drop: Seedance 2.5 bullet-time prompts.

Copy-paste remap blocks (unofficial)

Each block is one Sora-style habit → three vendor remaps. Copy the column you will actually run. Jump: dolly-in · handheld · locked-off · orbit · aerial · whip pan · macro · multi-shot split · dialogue · identity lock · rack focus · rain physics.

Remap 1 · slow dolly-in

Physics paragraph → dolly tied to an event

Sora liked an omniscient camera. Remap so the push-in starts when something happens.

Sora-style (do not paste as-is)

Photoreal cinematic shot of a woman in a wool coat standing in a narrow Tokyo alley at dusk. Steam lifts off a ramen cart. She looks down at a paper ticket, then up as a train rumbles overhead. The camera slowly discovers her face, neon reflecting in wet cobblestones, shallow depth of field, anamorphic flares, 12 seconds, immersive world simulation, natural physics, native audio of city rain and train.

Seedance 2.5 remap

FORMAT: 16:9, 12 seconds, photoreal, dusk neon, no text overlay.
SUBJECT + ACTION: A woman in a wool coat stands by a ramen cart in a narrow Tokyo alley; she reads a paper ticket, then lifts her face as a train passes overhead.
TIMELINE: 0–5s she studies the ticket, steam drifting across frame. 5–12s she looks up; only then does the camera begin a slow dolly-in to a medium close-up.
CAMERA: Locked medium-wide until the look-up; slow dolly-in triggered by the head raise. One move only.
CONTINUITY: Same coat, hair, ticket, cart. Do not change her face after the push-in.
AUDIO: Ramen steam hiss, distant train rumble, light rain on cobbles. No music.
CONSTRAINTS: No extra people in focus, no readable store logos, no whip or orbit.

Kling remap

Shot 1 (0–4s): Medium-wide, static. Woman in wool coat at a ramen cart, Tokyo alley, dusk neon, wet cobbles. She reads a paper ticket.
Shot 2 (4–10s): Same axis. Camera dollies forward 1.5m to a medium close-up as she looks up at an overhead train. Do not orbit. Do not pan off her.
Staging: steam in the foreground left; cart stays screen-right. Keep spatial layout identical between shots.
Motion: subject head-raise is the only body beat; camera path is a straight push-in.
Look: photoreal, shallow DOF, practical neon only. 16:9. No on-screen text.

Veo remap

8-second 16:9 photoreal clip. A woman in a wool coat at a ramen cart in a wet Tokyo alley at dusk. Medium-wide, then a slow dolly-in to her face as she looks up at a passing train. Steam, neon in puddles, shallow depth of field, practical lights only. Diegetic sound: steam hiss, train rumble, light rain. No music, no voiceover, no captions.

Remap 2 · handheld documentary

“Phone-camera energy” → bounded shake + follow

Sora treated handheld as a vibe. Other models need how much the frame drifts and what it is following.

Sora-style (do not paste as-is)

Found-footage handheld documentary of a night market cook flipping noodles, chaotic energy, immersive, the camera is right there in the smoke, 10 seconds, photoreal, native crowd audio.

Seedance 2.5 remap

FORMAT: 16:9, 8 seconds, photoreal night market, no text.
SUBJECT + ACTION: A cook at a wok station flips noodles once, then plates them.
TIMELINE: 0–3s settle on the wok. 3–8s one flip; camera follows the wok, not the crowd.
CAMERA: Handheld follow at chest height; small drift only (no whip, no crash zoom). Stay on the cook.
CONTINUITY: Same cook, same stall, same wok.
AUDIO: Wok sizzle, metal spatula, muffled crowd. No music.
CONSTRAINTS: Faces in the background unreadable. No extra camera operators in frame.

Kling remap

Single continuous take, 8s, 16:9. Camera: handheld follow, mild shake, tracking the cook’s hands and wok. Subject walks 1m left along the stall; camera trucks with them. Do not cut. Do not orbit. Night market practicals, smoke in the key light. Photoreal. Crowd as bokeh only.

Veo remap

Handheld close documentary, 8 seconds, 16:9. Night-market cook flips noodles in a wok; light camera shake, always on the pan and hands. Photoreal steam and oil pop. Sound: sizzle, spatula clang, distant crowd wash. No music.

Remap 3 · locked-off wide

Static master → explicitly forbid camera travel

Sora still invented micro-moves. Say “tripod, no push, no pan.” Wedding vows live in this family — see ceremony vows.

Sora-style (do not paste as-is)

Wide cinematic establishing shot of a small outdoor wedding ceremony, couple at the altar, guests listening, golden hour, emotional, 8 seconds, photoreal film look.

Seedance 2.5 remap

FORMAT: 16:9, 8 seconds, golden hour, photoreal.
SUBJECT + ACTION: Two partners face each other at an outdoor altar; an officiant is soft in the background; a breeze moves a veil. Guests stay out of sharp focus.
CAMERA: Static locked-off wide on a tripod. No dolly, pan, tilt, or orbit.
CONTINUITY: Same couple, wardrobe, altar. No guest faces readable.
AUDIO: Quiet vows, leaves, one sniffle. No score.
CONSTRAINTS: Do not invent a camera move. Do not cut.

Kling remap

One shot, 8s, 16:9. Camera locked on tripod, wide. No camera motion controls: zoom 0, pan 0, tilt 0, truck 0. Outdoor altar, couple in sharp midground, guests as blur. Golden hour, shallow DOF. Motion only in veil and trees.

Veo remap

Locked-off wide, 8 seconds, 16:9. Outdoor wedding vows at golden hour. Camera does not move. Couple at the altar, guests unreadable, breeze in a veil. Soft natural light. Sound: quiet spoken vows, leaves. No music.

Remap 4 · orbit / 360

Sora “circle the subject” → degrees, height, stills

Orbits expose identity. Prefer stills. Longer CCTV / bullet-time / product packs: Seedance 2.5 bullet-time video prompts.

Sora-style (do not paste as-is)

Epic 360 orbit around a dancer in a warehouse as dust hangs in the air, cinematic, Matrix-like, 10 seconds, photoreal, the camera flies around her.

Seedance 2.5 remap

FORMAT: 16:9, 8 seconds.
SUBJECT + ACTION: A dancer in [WARDROBE] holds a mid-turn pose in a warehouse; dust motes hang.
REFERENCE ROLES: @Image1 face and outfit only. Do not copy the still’s background if it conflicts.
CAMERA: Chest-height 180° orbit, constant radius, level horizon. No crane, no speed ramp.
CONTINUITY: Same face and wardrobe for the full arc. Do not morph as the back of the head appears.
AUDIO: Cloth snap, one muted whoosh, then near silence. No licensed music.
CONSTRAINTS: No extra characters. No on-screen text.

Kling remap

Single take, 8s. Camera path: 180° orbit clockwise at chest height, ~2m radius, around a dancer frozen mid-turn in a warehouse. Keep horizon level. Subject almost still; camera does all travel. Attach a front still if the host allows. Photoreal, dusty practical shafts. 16:9.

Veo remap

8-second 16:9 clip. Smooth 180° orbit at chest height around a dancer mid-turn in a warehouse, dust hanging in light shafts. Photoreal, level horizon, no speed ramps. Sound: cloth and a short whoosh. No music. Keep the same person throughout the orbit.

Remap 5 · aerial / crane

Drone establish → altitude change + what is revealed

Sora “cinematic aerial” is vague. Name start height, end height, and the reveal. Venue version: golden-hour drone grounds.

Sora-style (do not paste as-is)

Breathtaking drone shot of a coastal villa at sunset, cinematic, sweeping, 12 seconds, photoreal, birds, ocean, luxury atmosphere.

Seedance 2.5 remap

FORMAT: 16:9, 10 seconds, golden hour, photoreal.
SUBJECT + ACTION: A coastal villa and empty garden paths; ocean in the background; no readable faces.
TIMELINE: 0–5s high wide over water. 5–10s descending orbit that reveals the stone facade and string lights.
CAMERA: Descending drone orbit, one continuous move. No snap zooms.
CONTINUITY: Same building, same time of day.
AUDIO: Wind, distant surf, faint birds. No music.
CONSTRAINTS: No brand names on the property. No invented helicopters in frame.

Kling remap

Shot 1: High aerial over ocean toward a coastal villa, sunset. Camera cranes down and trucks right.
Shot 2: Continue the same descending orbit to reveal garden paths and stone facade. Keep the villa as the spatial anchor. 10s total, 16:9, photoreal. No people in close-up.

Veo remap

8-second 16:9 aerial. Slow descending orbit of a coastal villa at golden hour, ocean behind, empty garden chairs. Photoreal, no readable faces. Wind and surf only. No music.

Remap 6 · whip pan into close-up

Two-beat camera → cut or timestamp, do not stack moves

Sora stacked whip + CU in one sentence. Remap as two beats or two Kling shots so the blur has a start and a land.

Sora-style (do not paste as-is)

A detective in a rainy diner. The camera whip-pans from the neon window to an extreme close-up of his eyes, cinematic, 8 seconds, photoreal, tense music.

Seedance 2.5 remap

FORMAT: 16:9, 8 seconds, rainy night diner, photoreal.
SUBJECT + ACTION: A detective in a dark coat sits at a booth.
TIMELINE: 0–3s hold on neon-wet window glass. 3–5s whip pan right. 5–8s land on a close-up of his eyes; hold.
CAMERA: One whip pan only, triggered at 3s. No second move after landing.
CONTINUITY: Same man before and after the whip.
AUDIO: Rain on glass, a cup set down. No music, no score.
CONSTRAINTS: No extra whip. No text.

Kling remap

Shot 1: Medium, rainy diner window with neon smear. Detective in booth, profile.
Shot 2: Fast pan right that blurs, then snaps to a close-up of the same detective’s eyes. Same location, same wardrobe. 8s, 16:9. Camera motion is the pan only.

Veo remap

8 seconds, 16:9. Rainy diner at night. Start on neon in the wet window, then a single whip pan lands on a close-up of a detective’s eyes. Photoreal. Rain on glass, cup clink. No music.

Remap 7 · macro product

Hero pack → label lock + one table move

Sora invented typography. Attach a label still. Product orbit/drop variants: product orbit and product drop.

Sora-style (do not paste as-is)

Luxury commercial of a soda can sweating on marble, camera pushes into the logo, cinematic product film, 8 seconds, photoreal, fizzy sound, studio lights.

Seedance 2.5 remap

FORMAT: 16:9, 8 seconds, clean product studio.
SUBJECT + ACTION: [PRODUCT] in [PACK COLOR] sits on marble; condensation beads slide.
REFERENCE ROLES: @Image1 locks label typography, logo placement, and material. Do not rewrite the label.
CAMERA: Macro push into the label only. No 360 unless the still set includes a back panel.
CONTINUITY: Identical packaging throughout.
AUDIO: Soft fizz, marble tick. No voiceover, no music.
CONSTRAINTS: No invented slogans. No extra products.

Kling remap

Macro table shot, 8s, 16:9. [PRODUCT] on marble. Camera: slow dolly-in 30cm toward the label. Zoom mild. No orbit. Keep packaging identical to the attached still. Condensation motion only. Studio white + thin rim. No readable competitor brands.

Veo remap

8-second 16:9 studio macro. Slow push-in on a [PACK COLOR] [PRODUCT] on marble; condensation beads slide; keep the label unreadable or match a still you own. Soft fizz, no music, no voiceover.

Remap 8 · multi-shot story split

One Sora “short film” prompt → separate beats

This is the most common Sora hangover. Do not ask one clip for a whole wedding or heist. Split. Ceremony sequence: AI wedding video prompts.

Sora-style (do not paste as-is)

A mini wedding film: first look in a hallway, then vows, then a first dance under string lights, then a rainy exit to a car, emotional photoreal cinematic 20 seconds with music.

Seedance 2.5 remap

Do not generate the whole film in one prompt. Run four jobs (or timestamp a 20–30s Seedance clip as four beats):
0–7s FIRST LOOK: hallway windows, slow dolly, two partners turn. Same wardrobe as stills.
7–14s VOWS: cut to locked-off altar wide. No camera travel.
14–22s FIRST DANCE: slow orbit, string lights, guests as bokeh.
22–30s RAINY EXIT: doorway, umbrella, petals; dolly with them to the car.
CONTINUITY: Restate wardrobe + hair at each cut. AUDIO: diegetic only, no commercial song.
Better: generate each beat as its own clip and stitch in an editor — copy ready-made beats from the wedding prompt pack.

Kling remap

If using Kling multi-shot (2–6 shots), cap at three connected shots, not a feature:
Shot 1: hallway first look, slow dolly.
Shot 2: altar vows, locked-off wide, same couple / wardrobe.
Shot 3: first-dance orbit, string lights.
Stop there. Rainy exit is a separate generation. Keep spatial continuity (same venue architecture language). 15s max class — do not ask for 20s of four locations.

Veo remap

Veo wants one audiovisual take. Pick a single beat, ~8 seconds. Example: first look only — two partners turn in a windowed hallway, slow dolly, golden hour, fabric rustle, no music. Generate vows, dance, and exit as separate clips and edit them together.

Remap 9 · dialogue + native audio

Assumed Sora sound → spoken line + diegetic bed

Hosts disagree on how to mark dialogue (UI field vs brackets). Put the spoken line in the slot your vendor documents; keep language and delivery outside the quote.

Sora-style (do not paste as-is)

Two friends on a fire escape at night, one says something funny about the city, cinematic, native audio, 8 seconds, photoreal, New York vibe.

Seedance 2.5 remap

FORMAT: 16:9, 8 seconds, night fire escape, photoreal. Enable native audio if the host has a flag.
SUBJECT + ACTION: Two friends sit on a metal fire escape. One turns and speaks; the other laughs once.
CAMERA: Slow dolly-in, medium two-shot. No orbit.
DIALOGUE LANGUAGE: American English, dry, quiet — not shouted.
SPOKEN LINE (put in your host’s dialogue field / brackets): "We should have left at nine."
AUDIO BED: Far traffic, a window AC, one laugh. No music, no subtitles unless you add them in an editor.
CONTINUITY: Same two people. CONSTRAINTS: No extra speakers off-camera.

Kling remap

8s two-shot, 16:9, night fire escape. Camera: slow push-in. Character A (left) says, quietly in American English: "We should have left at nine." Character B laughs once. If Kling audio is a paid add-on, enable it; otherwise picture-only and add sound later. Traffic + AC bed. No song.

Veo remap

8-second 16:9 night scene. Two friends on a fire escape. Slow dolly-in. One says, quietly, "We should have left at nine." The other laughs. Photoreal. Diegetic traffic and a window AC. No music, no captions.

Remap 10 · identity / cameo lock

Sora cameo or remix ID → still roles

OpenAI remix video IDs and cameos do not port. Generate stills first (pose lock, product lock) via the Nano Banana Works posts hub, then assign each image a job.

Sora-style (do not paste as-is)

Use my cameo / remix this Sora video: same person walking through a sunflower field, camera follows, 8 seconds, photoreal, wind audio.

Seedance 2.5 remap

FORMAT: 16:9, 8 seconds.
SUBJECT + ACTION: The person walks a dirt path through sunflowers, glancing once at camera.
REFERENCE ROLES: @Image1 = face / hair / outfit only. @Image2 = three-quarter, same wardrobe. @Image3 = outfit detail. Do not take the field from the stills if they were shot indoors.
CAMERA: Side-tracking follow at waist height. One path.
CONTINUITY: Same face after the glance. Restate hair color and jacket when they turn.
AUDIO: Wind in stalks, footsteps. No music.
CONSTRAINTS: No new identity. Consent required for real people.

Kling remap

Image-to-video if available: lock the attached still as the person. 8s, 16:9. Camera trucks sideways with them through sunflowers. Walk cycle only — no wardrobe change. If using Elements / character library, pick that asset instead of prose identity. Wind motion in the field. Photoreal.

Veo remap

8 seconds, 16:9. Attach up to the host’s reference-image limit. The same person as in the stills walks through sunflowers; side-tracking camera. Keep face, hair, and jacket locked. Wind in the stalks, footsteps. No music. Do not invent a different person.

Remap 11 · rack focus

Unstated Sora rack → focus plane A, then B

Name both planes and when the rack happens. Do not also dolly unless that is the only move.

Sora-style (do not paste as-is)

Cinematic shot through a rainy cafe window, focus pulls from raindrops to a writer inside, photoreal, 8 seconds, moody, jazz implied.

Seedance 2.5 remap

FORMAT: 16:9, 8 seconds, rainy day, photoreal.
SUBJECT + ACTION: Raindrops on cafe glass in the foreground; a writer at a table inside, notebook open.
TIMELINE: 0–3s focus on raindrops. 3–8s rack focus to the writer; they turn a page.
CAMERA: Locked-off through the window. Rack only — no dolly.
CONTINUITY: Same writer, same notebook.
AUDIO: Rain on glass, muffled interior. No jazz, no licensed music.
CONSTRAINTS: No camera travel.

Kling remap

Static camera, 8s, 16:9, shot through cafe glass. Start sharp on raindrops on the pane; rack focus to a writer inside who turns a page. Camera motion controls all zero. Photoreal, cool daylight. No music.

Veo remap

8-second 16:9 locked shot through a rainy cafe window. Focus starts on raindrops, then racks to a writer inside turning a notebook page. Photoreal. Rain on glass, muffled room tone. No music.

Remap 12 · rain / physics chain

Sora world-sim → cause, effect, aftermath

Sora inferred splashes. Other models skip the middle unless you write it. Frozen-rain orbit variant: bullet-time rain.

Sora-style (do not paste as-is)

A cyclist hits a puddle on a neon street and water explodes everywhere, beautiful physics, cinematic slow motion, 8 seconds, photoreal, night, native audio.

Seedance 2.5 remap

FORMAT: 16:9, 8 seconds, wet neon street, photoreal.
SUBJECT + ACTION: A cyclist rides toward a puddle.
TIMELINE / PHYSICS: The front wheel enters the puddle. Water sheets up the fork. A crown splash peaks. Droplets hang, then fall as the bike continues through.
CAMERA: Low three-quarter tracking on the bike’s left; slow motion only during the splash peak, then resume. No orbit.
CONTINUITY: Same rider, same bike, same jacket.
AUDIO: Tire hiss, splash, distant neon buzz. No music.
CONSTRAINTS: No second crash. No readable shop names.

Kling remap

8s, 16:9, night wet street. Camera trucks with a cyclist on the left, low angle. Beat: wheel hits puddle → splash crown → droplets fall → bike exits puddle. One tracking path, no whip. Photoreal neon in asphalt. Motion emphasis on water, not camera acrobatics.

Veo remap

8-second 16:9 photoreal night shot. A cyclist’s front wheel hits a puddle; water sheets up, a splash crown peaks, droplets fall, the bike rolls through. Low tracking camera on the left. Tire hiss and splash. Neon in wet asphalt. No music, no readable store names.

Where to draft cheap, then spend

Short H3 Max / Hailuo / fal clips are still useful for framing tests before a Seedance-class or Veo pass. Use the H3 Max text to video prompt generator or image-to-video motion tips. Sora-exit remap for MiniMax: H3 as a Sora alternative. Free-quota checklist: MiniMax H3 Max free notes. Picker vs copy-paste pack: Seedance vs Kling vs Veo cheat sheet. Alibaba path: Wan 3.0 prompt guide.

FAQ

When did the Sora app die vs the API?

The Sora web and app experiences were discontinued on 26 April 2026. The Videos API and all listed sora-2* aliases shut down on 24 September 2026. Those are OpenAI’s dates, not ours.

What did OpenAI recommend as a replacement?

Nothing. The deprecations table’s “recommended replacement” column is blank for the Videos API, sora-2, sora-2-pro, and the dated snapshots. You migrate to another vendor.

I call Sora through Azure / Microsoft Foundry — is 24 Sep still my date?

Not for the Foundry deployment. Microsoft’s English retirement schedule lists Foundry sora-2 2025-12-08 for 15 October 2026. See the Foundry dual-clock page. You still need to export and remap; the extra weeks are not a replacement model.

How do I export Sora app videos?

OpenAI’s Help Center says to go to sora.chatgpt.com/sunset and click Export; you get an email when the archive is ready. Do that before any final deletion window.

Are these Seedance / Kling / Veo prompts official?

No. They are unofficial camera-language remaps. Syntax, duration caps, and audio flags differ by host and change without notice.

Why not paste my Sora prompt unchanged?

Sora briefs often simulate a world (physics, cameo, remix, stacked camera moves). Seedance wants timestamped direction and reference roles; Kling wants a camera path / shot list; Veo wants a short audiovisual take. Same story, different grammar.

Seedance vs Kling vs Veo — which remap column?

Use Seedance 2.5 when you have stills/refs and a longer directed beat. Use Kling when the shot is a choreographed camera path or a short multi-shot. Use Veo when you want one photoreal take with diegetic sound in the same paragraph. This is craft advice, not a benchmark ranking.

Does this site generate Sora or Seedance video?

No. It is a static prompt lab. Paste the text into the vendor you already use.

Sources (OpenAI)