Surreal and Fantasy AI Video Scene Prompts

Surreal AI video prompts work best when you describe one impossible thing clearly and let everything else stay ordinary: a door standing alone in a desert, a staircase that folds into the sky, a teacup that pours upward. Fantasy prompts need the same discipline plus a creature or place with a defined look you repeat across shots. Write subject, action, setting, camera, light and style in that order, and most models hold the scene together.

Updated October 2026.

Surrealism and fantasy look different but behave the same on set: they both ask the viewer to accept one rule that is not true, and then to follow it calmly. The job of the prompt is to sell that rule with concrete detail so the model has something solid to build. This guide covers scale games, impossible architecture and objects that disobey physics under surrealism, then creatures, floating islands and atmospheric magic under fantasy — with a prompt structure you can reuse, original examples, and the notes on shot length and consistency that keep one clip on the same world.

Surrealism: scale, architecture, broken physics

Surreal beats land hardest when the impossible element is small enough to point at. Do not flood the frame with strangeness; introduce one oddity and let the camera treat it as normal. Here are original prompts across the three families.

Prompt 1 — Seedance 2 scale game

Seedance 2 (1080p, 5 seconds): “A lone fisherman sits on a wooden dock fishing, but the dock floats on a calm sea that is held inside a giant teacup; the rim of the cup curves up past the top of frame. Subject: fisherman on a dock. Action: casting a line, unhurried. Setting: still water inside an enormous teacup. Camera: slow push-in from behind. Light: soft overcast dusk. Style: hyperreal, muted colours, no text.”

Prompt 2 — Seedance 2 impossible architecture

Seedance 2 (720p, 6 seconds): “A modern city intersection where the road folds up into the sky like a sheet of paper and continues as a vertical street with cars driving upward. Subject: an empty intersection. Action: the road folds and traffic follows it upward. Setting: a bright, ordinary downtown. Camera: crane up that follows the fold. Light: clean midday sun. Style: photoreal, slight haze, no text.”

Prompt 3 — Wan 3.0 object against physics

Wan 3.0 (1080p, 12 seconds): “On a plain kitchen table, a cup of tea tips over but the liquid pours upward toward the ceiling in a slow, perfect stream, a bowl rising to meet it. Subject: a teacup and a bowl. Action: liquid pours upward and gathers in the floating bowl. Setting: a quiet kitchen. Camera: slow orbit around the table. Light: warm morning window. Style: hyperreal, shallow depth of field, atmospheric audio.” Wan 3.0 takes longer takes at every whole second, ideal for a slow surreal beat that needs to breathe.

Prompt 4 — Kling 3.0 scale contrast

Kling 3.0 pro (5 seconds): “A tiny elephant the size of a mouse walks across a saucer on a cafe table while a normal hand reaches in to place a sugar cube beside it. Subject: a miniature elephant. Action: walking across a saucer. Setting: a cafe table. Camera: macro static tripod. Light: soft warm afternoon. Style: photoreal miniature, crisp detail.” Hold the camera still so the scale reads; movement is what breaks a size trick.

Fantasy: creatures, floating islands, magic

Fantasy asks for a world with rules. Give your creature or place two or three fixed features and repeat them exactly, and the audience will believe the rest.

Prompt 5 — Seedance 2 floating island

Seedance 2 (4K, 5 seconds): “A small grassy island drifts through a golden sunset sky, a single tree and a waterfall spilling off its underside into mist. Subject: a floating island. Action: drifting slowly left to right. Setting: open sunset sky above cloud. Camera: wide, slow pull-back. Light: warm low sun, glowing rim. Style: painterly fantasy, rich colour, no text.” Seedance 2 is the only Seedance with 4K, so use it when the vista needs to hold on a big screen.

Prompt 6 — Seedance 2.5 magic atmosphere

Seedance 2.5 (1080p 10-bit, 6 seconds): “A robed figure stands in a dark forest clearing and raises one hand; motes of blue light lift from the moss and swirl upward around them. Subject: a lone mage. Action: raising a hand and casting. Setting: a misty night forest. Camera: slow orbit, eye level. Light: cool moonlight with the blue glow as key. Style: cinematic fantasy, 10-bit colour, volumetric haze, no text.”

Prompt 7 — HappyHorse 1.1 creature

HappyHorse 1.1 Text-to-Video (720p, 5 seconds): “A gentle moss-covered stone golem, twice the height of a person, kneels by a mountain stream and cups water in its hands; small flowers grow along its shoulders. Subject: a stone golem. Action: cupping water. Setting: a misty mountain stream. Camera: static tripod, low angle. Light: soft dawn. Style: painterly fantasy, earthy greens and greys.” HappyHorse offers nine aspect ratios, so the same creature can be framed wide for a scene or tall for a poster-style clip.

Prompt 8 — Veo 3.1 Fast magic transition

Veo 3.1 Fast (Frame Transition): upload a first frame of an empty stone chamber and a last frame of the same chamber overgrown with luminous vines, then prompt “a slow dolly-in as the chamber fills with glowing vines, a gradual crane shot rising, cool magical light blooming from the walls, cinematic fantasy style.” Veo reads cinematography terms directly, so slow dolly-in and crane shot give you precise camera control.

Prompt structure: subject, action, setting, camera, light, style

Every prompt above follows the same six-part order. Write them in this sequence and the model parses the scene cleanly.

Subject — the one thing the frame is about.
Action — what the subject does, in plain words.
Setting — where it happens.
Camera — the move, from the built-in chips or a plain sentence.
Light — direction, colour and mood.
Style — the look, the palette, and any format note.

For camera, the built-in chips include push-in, pull-back, orbit 360 degrees, crane up, whip pan, FPV one-take, dolly zoom, crash zoom, top-down, handheld, rack focus and static tripod, plus lens and look chips such as anamorphic, 85mm portrait, wide 24mm, 16mm film, macro, tilt lens, dreamy glow and 70mm epic. Seedance read a plain “Camera: ...” sentence; Veo read cinematography terms such as slow dolly-in and crane shot; Wan, Kling and Runway read a “The camera ...” sentence.

Holding a scene together across shots

A surreal or fantasy series falls apart when the world changes between clips. Fix the world first: pick two or three features of your creature or place and describe them identically every time. Then vary only the action and the camera. If you plan several clips, sketch them as stills first and morph between them so the design carries over — the keyframe morph tool is built for exactly that. For a deeper routine, read the guide to AI video character consistency.

Short surreal beats vs longer takes

A single surreal beat wants four to fifteen seconds — one idea, held long enough to register and no longer. That range is the natural home of Seedance 2, which runs four to fifteen seconds in one-second steps and gives you a starting image plus up to two reference images tagged @Image1 and @Image2 for multi-subject scenes. When a beat needs to unfold slowly — a slow pour, a long drift — reach for Wan 3.0, which runs from two to thirty seconds at every whole second. Use Seedance 2 for the crisp punctuated moment and Wan 3.0 for the patient one.

Shot length and camera at a glance

Beat	Duration	Camera
Scale game	4-6 seconds	Slow push-in or macro static
Impossible architecture	5-6 seconds	Crane up, follow the fold
Broken physics	8-12 seconds	Slow orbit
Floating vista	5 seconds	Wide, slow pull-back
Magic atmosphere	5-6 seconds	Slow orbit, eye level

Cost is shown before every generation and failed generations are refunded automatically, so you can test a wild idea without risk — see pricing for the credit maths across models.

Frequently asked questions

What is the best prompt order for a surreal scene?

Write subject, action, setting, camera, light and style in that order. Naming one impossible element inside an otherwise ordinary scene gives the model something solid to build around.

Which model suits a single surreal beat?

Seedance 2, which runs four to fifteen seconds in one-second steps and supports two reference images tagged @Image1 and @Image2. It is the only Seedance with 4K, useful for a large vista.

When should I use Wan 3.0 instead?

When the beat needs to unfold slowly. Wan 3.0 handles two to thirty seconds at every whole second, so it suits a long drift, a slow pour or a patient atmospheric shot.

How do I keep a fantasy world consistent across clips?

Fix two or three features of the creature or place and describe them identically every time, change only the action and camera, and reuse the same reference images or morph between stills.

How long should each clip be?

Aim for five to six seconds for a punctuated surreal moment and eight to twelve seconds for a slow impossible action. Longer does not help a single idea; use the extra time only for a slow unfolding.

Ready to build your first surreal or fantasy scene? Open Seedance 2, write the six parts in order, and start with one impossible thing.