Create video with the live Sora 2 model. Start from text or add a reference image, control the shot and duration, then review the generated picture and sound together.
Voor currently connects this page to a priced Sora 2 model. Before committing a long campaign or automated workflow, generate a short validation shot with the exact duration, format, reference type, and audio requirements you plan to use.
The reported Sora API retirement date is September 24, 2026. Preserve the prompt and accepted output, then use the Sora alternative migration workspaceto test Gemini Omni 1.1 Flash, H3 Max, and Wan 3.0 before the cutoff.
Each clip pairs the prompt with a concrete review task so you can inspect continuity, physical motion, or synchronized sound.
Prompt: An astronaut golden retriever named Sora levitates around an intergalactic pup-themed space station, flies through opening doors, then catches a tennis ball inside in zero gravity.
Watch whether the dog, suit, station design, and zero-gravity behavior remain coherent across the exterior-to-interior shot change. The review target is detailed instruction following and persistent world state.
Prompt: A figure skater performs a triple axel with a cat on her head.
The useful test is not whether the frame looks cinematic. Watch take-off, rotation, landing, balance, and whether the cat remains attached through fast motion.
Prompt: Two mountain explorers in bright technical shells, ice-crusted faces, eyes narrowed with urgency, shout in the snow one at a time.
Listen for speaker turn-taking, wind ambience, intelligibility, and whether mouth movement belongs to the active speaker. This example matters because Sora 2 is a video-and-audio model, not merely a silent image animator.
The next card stays visible at the edge—move across the strip to plan the shot in order.
Describe subject, ordered action, camera behavior, lighting, ambience, dialogue, and the final frame. The model requires a prompt; concise physical instructions are easier to diagnose than several unrelated scenes.
Optional. It becomes the opening visual contract and must match the chosen portrait or landscape ratio. State whether it controls identity, product geometry, wardrobe, or composition.
The live schema exposes 4, 8, and 12 seconds. Start with four seconds when testing motion or prompt interpretation; longer takes increase both cost and the opportunity for continuity drift.
Portrait produces 720×1280 and landscape produces 1280×720 in the selected model. Choose the delivery format before writing camera blocking.
An MP4 with synchronized audio. Review the audio track deliberately; a visually strong clip can still fail because dialogue, ambience, or effects do not match the visible event.
Name the subject, setting, light, and what must already be visible.
Choose one readable move—push, pan, orbit, handheld, or locked.
Quote short dialogue, identify the speaker, then add ambience or effects.
Ask for a stable final frame that can hold or cut into the next shot.
Contact, weight, sliding, and object permanence
Stable horizon, consistent lens, no surprise cut
Speaker timing, ambience, and audible events
Permission, disclosure, and reference ownership
Yes. The generator is locked to the live Sora 2 model in Voor rather than handing the prompt to an unrelated video model.
Yes. Describe the opening frame, action, camera movement, environment, sound, and ending. A single coherent shot is more reliable than several unrelated scenes in one prompt.
Yes. When the model form shows the reference-image field, upload an image you have permission to use and explain which identity, object, or composition must remain stable.
Sora 2 supports video with audio. Write dialogue exactly, separate ambience from speech, and review synchronization, pronunciation, and loudness before publishing.
The live form exposes the durations supported by the selected model. Longer shots cost more and give continuity more time to drift, so start with the shortest duration that can express the action.
Use a physical timeline: opening frame, subject action, camera path, lighting change, sound cue, and final frame. Avoid conflicting camera directions and vague adjective stacks.
Use only images and likenesses you have the right and consent to use. Do not create deceptive impersonation, non-consensual intimate material, or misleading footage presented as a real event.
Commercial use depends on your Voor plan, the model terms, and the rights in every prompt and reference. Clear music, brands, locations, and likenesses before release.
Check identity, hands, object permanence, contact with the ground, camera continuity, readable text, dialogue timing, audio artifacts, and whether the last frame is usable in an edit.
Choose a shot preset, edit the brief, set the supported duration and format, then generate and review picture and sound together.
Open Sora 2 generatorOptional cookies help us understand usage and measure ads. Essential cookies stay on so Voor works.