Sora Alternatives: What Happened and How to Move Your Prompts

What happened to Sora

• API: OpenAI announced the deprecation on 24 March 2026, and on 24 September 2026 the Videos API and the sora-2 and sora-2-pro models stopped working. Its official deprecations page lists no replacement model.
• App and website: several tech outlets reported that the Sora app and website shut down in late April 2026. We couldn't confirm this in an official OpenAI source, so treat it with some caution.

In practice, any workflow or prompt library built around Sora needs a new home. The good news is that the scene description (who, what they do, where, the light and the camera) carries over almost entirely. What changes is the format.

What carries over and what needs changing

• Carries over: the subject and their traits, the action, the location, lighting, style and film vocabulary. Every current model understands camera moves described in plain English.
• Language: Veo and Wan ask for English prompts. Dialogue in other languages is documented by Kling 4.0 and 3.0, Seedance 2.5 and MiniMax H3.
• Audio: each model has its own format. Veo uses quotes, "SFX:" and "Ambient noise:"; Seedance uses symbols such as { } and < >; H3 uses <d> tags. Runway Gen-4.5 audio isn't verified.
• Negatives: Veo and Wan have a separate field; Kling 4.0 advises against them apart from specific exclusions; Seedance uses constraint phrases; Runway doesn't support negative phrasing.
• Length: every model has its own limits (for example 4–8 s per clip in Veo 3.1 and up to 30 s in Kling 4.0). If your scene was long, split it into shots.

The starting prompt

We start from the kind of all-in-one prose prompt many people wrote for Sora: scene, camera, sound and a line of dialogue mixed together.

A lighthouse keeper climbs the spiral staircase of a stone lighthouse at night during a storm, holding a lantern. He's in his fifties, with a grey beard and a yellow oilskin coat. The camera follows him from below. You can hear the wind and rain against the windows. Halfway up he stops, looks up and mutters, "Not tonight." Cinematic, dark mood, no text on screen.

After: Veo 3.1

Camera first (cinematography, subject, action, context, style), dialogue in quotes with the speaker named, and sound in separate sentences. "No text on screen" moves to the negativePrompt field as a list of terms. See the Veo 3 prompts guide for more.

Low-angle tracking shot following a lighthouse keeper up a spiral iron staircase inside a stone lighthouse at night during a storm. He is in his fifties, with a grey beard and a yellow oilskin coat, holding a swinging lantern that is the only light source. Moody cinematic drama, deep shadows, cold blue light from the windows. Halfway up he stops, looks up and mutters, "Not tonight." SFX: heavy footsteps ringing on iron steps. Ambient noise: howling wind and rain lashing the windows.

subtitles, text, watermark

After: Kling 4.0

Kling 4.0 organises the prompt into general settings, a timeline and audio. State the dialogue language and describe the voice. Use specific exclusions instead of negatives. If you upload an image of the keeper, tag it @Image1 and say what to keep. The Kling prompts guide has the details.

Scene: the inside of a stone lighthouse during a night storm, rain lashing the small windows. Style: realistic, moody cinematic drama. Camera: low-angle tracking shot following the keeper up the stairs.
0–5 s: a lighthouse keeper in his fifties with a grey beard and a yellow oilskin coat climbs the spiral iron stairs, holding a swinging lantern.
5–10 s: he stops halfway, catches his breath and looks up toward the lamp room.
Audio: howling wind, rain on glass, footsteps on iron steps. Keeper (low, tired voice of a man in his fifties, slow pace, English): "Not tonight." No subtitles, no background music.

After: Seedance 2.5

Seedance 2.5 wants a one-sentence summary, then the plot with whole-second timings and no gaps, then consistency notes. Dialogue goes in curly braces. There's no negative field, so restrictions are written as sentences. More in the Seedance prompts guide.

A lighthouse keeper climbs the spiral stairs of a lighthouse during a night storm, moody drama, slow tracking shot from below.
0-3s: the keeper, a broad-shouldered man in his fifties with a grey beard and a yellow oilskin coat, starts up the iron stairs holding a lantern.
3-7s: the lantern swings as the wind rattles the windows; he stops halfway and looks up.
7-10s: close-up of his face lit by the lantern. The keeper's line (weary): {Not tonight.}
Consistency: the same grey beard and yellow oilskin coat throughout; the lantern is the only light source.
Generate only environmental and action sounds. Avoid generating any text or subtitles.

After: Wan

Wan takes English (or Chinese) prompts: an overall description, then timed shots. Give each character a fixed label instead of pronouns, put the action before the dialogue and describe the voice. Negatives go in negative_prompt. On Wan 2.2, which has no sound, drop the audio. See the Wan prompts guide.

A lighthouse keeper climbs the spiral iron stairs of a stone lighthouse at night during a storm, lit only by his lantern, moody cinematic realism. Shot 1 [0–4 s] low-angle tracking shot, the keeper, a man in his fifties with a grey beard and a yellow oilskin coat, climbs with the lantern swinging. Shot 2 [4–8 s] close-up, fixed camera, the keeper stops, looks up and says "Not tonight." in a low, weary, slow voice with a rough timbre. Sound: howling wind and rain against the windows.

subtitles, text, watermark, background music

After: MiniMax H3 (Hailuo)

H3 uses sections: the integrated description (picture, camera and dialogue), the soundscape in English, and non-diegetic music or N/A. Describe the camera in prose with type, amplitude and speed; give the speaker an ID (S1) and put the line in <d> tags. More in the Hailuo prompts guide.

integrated_multimodal_description: [Shot 1] A low-angle tracking shot follows a lighthouse keeper in his fifties, grey beard, yellow oilskin coat, as he climbs a spiral iron staircase inside a stone lighthouse during a night storm, his swinging lantern the only light; the camera moves upward with small amplitude at slow speed. At 00:05.000, the camera cuts to a close-up of his face as the keeper (S1) says: <d>[English] Not tonight.</d>
overall_soundscape: Wind howls around the tower and rain lashes the windows. Heavy boots ring on iron steps.
non_diegetic_music: N/A

After: Runway Gen-4.5

Runway asks you to describe only what should happen: its guide warns that negative phrasing can produce the opposite result, so "no text on screen" goes. Because Gen-4.5 audio isn't verified in official documentation, we also leave out sound and dialogue. More in the Runway prompts guide.

Low-angle tracking shot: the camera follows a lighthouse keeper in his fifties, grey beard and yellow oilskin coat, up a spiral iron staircase inside a stone lighthouse at night. His swinging lantern is the only light, and rain streaks the small windows. Halfway up he stops and looks toward the lamp room above. Moody cinematic drama, deep shadows, cold blue tones from outside.

Quick checklist for porting your prompts

• Choose the model for the feature you need (dialogue language, clip length, references, open weights), not out of habit. The AI video generator comparison helps.
• Split your prompt into parts: subject, action, location, light, camera, style, sound and dialogue.
• Rewrite audio and negatives in the target model's format.
• If the scene is long, break it into shots using that model's timestamp format.
• Test with one short clip, change one thing per attempt and note the model version.

If you'd rather not convert prompts by hand, PromptNow has a prompt generator for each of these models: describe the scene in the form and get the prompt back in that tool's syntax, with a separate negative field where it has one.