Prompt From Video / Scene intelligence
Turn your idea or reference video into a detailed Sora prompt with clear scene direction, action, composition, and pacing.
A red rally car sweeps through a sunlit desert bend as the camera tracks alongside it. A layered dust plume builds depth behind the moving vehicle…
01The output stays grounded in visible evidence: shots, timing, movement, lighting, and sound.
World detail, composition, and narrative intent
Sora prompts can read like direction for a cinematographer. Begin with the world and its visual rules, place the subject inside it, then describe the action and camera. Concrete relationships matter: where the light comes from, what occupies foreground and background, and how the subject changes.
Use a temporal structure when the scene evolves. “At first,” “as,” and “by the end” establish causality without turning the prompt into a frame-by-frame script. For physically unusual ideas, explain the rule once and show its consequence. A city folding like paper is clearer when buildings crease along streets and cast changing shadows.
Composition carries emotional intent. Negative space can isolate a character, a low angle can make architecture oppressive, and a slow reveal can withhold scale. The analyzer preserves these relationships instead of reducing the clip to style keywords.
State what is ordinary or impossible and keep that logic consistent through the shot.
Place useful elements in foreground, midground, and background, then guide attention.
Describe the opening state, the change, and the final image in causal order.
Use a coherent lens, movement, palette, texture, and lighting source instead of stacked style names.
Editable prompt framework
Build the scene in layers rather than stacking visual adjectives. This template moves from the world rule to composition, action, and finish, so every sentence controls something visible. The labels help with drafting and can be removed when the completed prompt reads as one coherent piece of direction.
Template structure
[Location, time, and one consistent world rule]. [Subject] positioned among [foreground, midground, and background elements]. At first, [opening state]; then [causal change]; by the end, [final image]. [Framing, lens character, and one camera move]. [Motivated light, palette, atmosphere, texture, and continuity constraints].
State the location, time, material qualities, and any unusual physical rule. If the idea is impossible, explain one consistent mechanism and show its effect. Do not introduce a second transformation unless the shot has enough time and composition to make both changes readable.
Place the subject in relation to foreground, midground, and background elements. Name the framing, lens character, and source of light when they shape the reveal. Concrete spatial relationships give the scene more control than a list of genre or mood references.
Describe the opening state, the event that changes it, and the final image. Use causal phrases only where they clarify progression. Keep the main subject recognizable while the world reacts, and choose one camera move that supports rather than competes with the transformation.
Finish with motivated palette, contrast, atmosphere, surface detail, and continuity needs. Remove repeated adjectives and directions that cannot be seen. Put duration, aspect ratio, or interface-specific controls in their proper fields when the Sora version you use exposes them.
Make one pass for geography and another for time. On the geography pass, check that the camera could actually see every important element from the stated position. On the time pass, check that the opening, change, and ending fit the clip. Split the concept when either pass reveals two unrelated shots hiding inside one paragraph.
A reference video provides useful relationships even when you replace its content. Preserve the observed reveal, depth, lighting direction, or pacing, then rewrite the people and setting. The Sora prompt generator turns those choices into an editable brief; generation still varies with model version, seed, duration, and any source image you provide.
Treat style references as a final constraint, not the scene itself. First make the subject, space, action, and camera understandable without a named genre or artist. Then translate the desired finish into observable traits such as soft halation, hard side light, muted production design, or miniature depth cues. Concrete traits survive revision better and reduce conflicts between stacked style labels.
At blue hour, a quiet modern city begins folding along its streets like an enormous paper model. In the foreground, a cyclist stops beneath a streetlamp; distant towers crease inward one by one, changing the direction of long reflections. Slow aerial pullback reveals the geometric transformation, realistic material weight, cool cyan sky and warm windows.
A lone conservator crosses a dark natural-history museum carrying a small work light. Dinosaur skeletons form layered silhouettes from foreground to deep background. The camera glides parallel behind glass cases; dust enters the beam, and by the end a vast whale skeleton appears overhead. Quiet, patient, photorealistic.
Wide 35mm view of a glass greenhouse standing alone in windblown dunes. Inside, tropical leaves press against fogged panes while sand streams around the foundation. A slow push-in passes from harsh amber exterior light to saturated humid green, revealing one person watering plants in the distance.
Reference workflow
Upload a short reference or paste a public link in the workspace above. The analyzer follows the timeline rather than guessing from a thumbnail: it records shot order, subject motion, camera behavior, light, transitions, dialogue, ambience, and music.
Select the Sora result to see the same evidence written in this model’s preferred syntax. Keep the facts that define the reference, then edit creative choices such as subject, location, palette, or ending. The result remains a new production brief, not a promise of a frame-for-frame copy.
Quality control
Use enough detail to establish world, subject, action, camera, and finish. Remove adjectives that do not change a visible decision.
It is not mandatory, but a clear framing and one motivated move help control how the scene is revealed.
Explain one consistent physical or visual rule, then describe observable consequences such as movement, scale, light, and interaction.
Yes. You can use the on-page template to draft a Sora prompt manually or use the limited free reference analysis to extract one, then edit it before generation.
Yes. Fill in the world rule, composition, action through time, camera, and finish. Keep only the fields that change the scene you want to make.
