One shot. One action. One camera idea.
Sora prompts: engineer the shot before you generate the clip
Sora does not fail because you lacked adjectives. It fails when the prompt asks for a montage, a wardrobe change, three camera moves, and a plot twist in eight seconds. Video models want shot grammar: subject, action, camera, duration, style, light, avoid. Below is a free Shot Builder that assembles that grammar, followed by the system so you can write it by hand and debug the clips that almost work.
3. Camera move
4. Duration
5. Style
6. Lighting
7. Mood
Runs in your browser — copy into Sora when ready. Here is the cinematography logic behind the builder.
Think in shots, not screenplays
The biggest habit to unlearn from chat models is the multi-beat paragraph. “A hero walks into a bar, fights goons, then rockets to space while the camera does cool stuff” is a trailer edit, not a shot. Sora, given that soup, will smear time and identity trying to honor every clause. A shot is smaller and stricter: eight seconds, one subject, one continuous action, one camera move, consistent light. If you need a sequence, generate multiple shots and edit — or use a workflow designed for multi-shot storytelling. Do not smuggle an edit into a single generation and blame the model.
Cinematographers already solved this language. Push-in, pull-back, tracking, crane, static, orbit — these words compress pages of intention. Duration tells the model how ambitious the motion can be. Style and lighting set the render prior. Negatives fence the usual failure modes. The free builder is just those decisions in a form that cannot “forget” continuity language on the way out.
When a clip is almost right, change one lever. If the subject drifts, simplify action and strengthen continuity. If the clip feels flat, change camera or duration — not all adjectives at once. Iterating like a director on set beats rewriting a novel prompt from scratch every take.
Anatomy of a Sora shot prompt
Subject and setting
Lead with concrete nouns: who or what, where, distinctive materials and colors. “A red fox on a frost-covered bridge in a pine forest” gives the model anchors. “A beautiful animal in nature” gives it a lottery ticket.
Action beat
One continuous action the camera can witness without cutting. “Pauses, then trots across as snow begins” is a beat. “Lives its whole life” is not. If you catch yourself writing “then later,” you are writing a second shot.
Camera and duration
Pair them. A slow push-in in four seconds is a different animal than a slow push-in in twenty. Locked static makes performance and secondary motion the star. Orbital moves need time to read. The builder attaches a plain-language beat description so duration is not a naked number.
Style, light, mood
Style is the render contract: cinematic film, documentary, anime, commercial polish, noir, vintage, hyperreal, dreamlike. Light is physical: golden hour, neon, softbox, god rays. Mood is emotional pacing — and should be shown, not written as on-screen text. Keep all three pointing the same direction; “playful neon noir documentary anime” is how you get sludge.
Continuity and avoid
Continuity clauses are cheap insurance: single take, stable physics, consistent identity. Avoid lists catch known demons: text, watermarks, morphs, extra limbs, jump cuts. Short targeted avoids beat encyclopedias of hate.
How to use the Sora Shot Builder
Fill subject first — if you cannot describe the frame in one breath, the model cannot hold it. Add action, pick camera and duration, choose style, optionally light and mood, optionally avoid. Generate, copy to Sora, review. When Supercharge is useful, it is for polishing cinematography language in Studio, not for inventing a second plot.
Five Sora shot recipes you can steal
Nature push-in
A 8s continuous shot of a lone red fox on a frost-covered wooden bridge in a pine forest. Action: the fox pauses, ears twitch, then trots carefully across as snow begins to fall. Camera: slow cinematic push-in toward the subject. Establish, develop one action, hold the payoff. Visual style: cinematic film look, anamorphic lens character, subtle film grain, 24fps motion cadence. Lighting: golden hour sun, natural falloff, consistent direction across the shot. Mood: serene, conveyed through pacing, color, and body language — not text. Continuity: single uninterrupted take, stable physics, consistent subject identity and proportions from first frame to last. Avoid: text, logos, watermarks, jump cuts, morphing faces, extra limbs, cartoon look.
Why it works — One animal, one path, one camera move — continuity has nowhere to break.
Product hero orbit
A 8s continuous shot of a matte-black wireless headphone on a slate pedestal in a minimal studio. Action: the headphone rotates slowly as a soft reflection glides across the ear cup. Camera: gentle 180° orbital arc around the subject. Visual style: premium advertising finish, crisp product-safe motion, clean grade, high production value. Lighting: studio softboxes, natural falloff, consistent direction across the shot. Mood: intimate. Avoid: text overlays, logos, dusty sensors, warped geometry, jump cuts.
Why it works — Product ads fail when action competes with the object — rotation + orbit is enough.
Street handheld beat
A 12s continuous shot of a street violinist under a rain-slick awning at night. Action: she draws a long note, glances at a passing train of headlights, then smiles and continues. Camera: subtle handheld camera with natural micro-shake. Visual style: neo-noir aesthetic, high contrast, deep shadows, cool highlights, tense atmosphere. Lighting: neon night, natural falloff, consistent direction across the shot. Mood: melancholic. Avoid: text, watermarks, face morphing, sudden cuts, exaggerated shake.
Why it works — Handheld + neon needs a simple human beat so motion noise does not become the story.
Anime action hold
A 4s continuous shot of a young mage on a cliff edge above a sea of clouds at dawn. Action: she raises a staff; wind lifts her cloak; a small spiral of light gathers at the tip — no cut. Camera: locked-off static tripod shot. Visual style: high-quality anime key animation, clean linework, expressive motion, cel-shaded color. Mood: epic. Avoid: live-action look, text, multi-scene montage, deformed hands.
Why it works — Short duration + static camera lets anime motion read cleanly without camera chaos.
Pull-back reveal
A 12s continuous shot of a child’s hands assembling a wooden model rocket on a kitchen table. Action: final fin snaps into place; hands pull away; the frame reveals a parent watching from the doorway. Camera: smooth pull-back reveal that widens the frame. Short arc: setup → motion → resolve. Visual style: 1970s film stock palette, soft halation, gentle gate weave, nostalgic texture. Lighting: soft overcast through a window. Mood: intimate. Avoid: text, modern UI overlays, jump cuts, identity morph.
Why it works — Pull-back only works if the opening frame is a detail worth leaving — start tight on hands.
Clip looks wrong: symptom → fix
“Subject face or body morphs mid-clip”
One identity only; add continuity clause; simplify action; avoid “transforms into.”
“Feels like random cuts”
Remove multi-beat “then” chains; demand single uninterrupted take; shorten duration.
“Camera motion is nauseating or aimless”
Pick one camera move; match duration; try static if performance matters more.
“Looks like the wrong genre”
Align style tokens; remove conflicting medium words (photo + anime + 3D).
“Text or logos appear”
Add explicit avoid for text/logos/watermarks; do not ask for captions in-frame.
“Action never finishes”
Lengthen duration or simplify the action so it can complete inside the window.
“Lighting flickers or flips direction”
Specify consistent light direction; avoid “day turns to night” inside one shot.
“Great still frames, dead motion”
Strengthen the action beat and secondary motion (cloth, snow, steam, dust).
Sora vs Veo vs image prompts
| Model / job | Prompt emphasis | PromptFork page |
|---|---|---|
| Sora video shot | Tight shot lines, camera tags, duration, avoid list | This page |
| Veo video scene | Fuller scene prose, temporal beats, physics language | Veo prompts |
| Still image | Subject, style, light, lens, negative prompt | AI image prompts |
| Chat drafting | Role, task, audience, format | ChatGPT prompt generator |
If you are generating stills for storyboards first, use image tooling, then promote the winning frame description into a Sora shot with camera and duration added. If you are on Google’s stack, switch to the Veo Shot Builder rather than pasting Sora tags unchanged.
A library of shots is a directorial style
Directors are not more talented at typing — they reuse shot languages that work. Keep a PromptFork library of push-ins, orbits, handheld streets, product turns, and reveal pull-backs. Fork, swap subject, keep the camera physics. That is how your clips start looking like a body of work instead of a slot machine.
Find
Search for shot recipes by camera move or genre instead of starting from blank Sora.
Copy
Paste a continuity-safe prompt and only rewrite the subject line.
Fork
Save your tuned style and avoid lists so the next clip inherits your taste.
For adjacent work — app demos, research on visual trends, ChatGPT scripts for voiceover — jump through the network: Lovable, Perplexity, ChatGPT generator, ChatGPT optimizer, and the model-agnostic FORGE grader.
On-set discipline for AI video
Work in takes. Keep a note of what changed between generations. Prefer shorter clips that work over longer clips that almost work. Build sequences in the editor where human taste still wins. Write voiceover and music after picture lock when you can — prompting Sora to solve sound, story, and camera simultaneously is how good ideas die in committee inside one text box.
When stakeholders ask for “more epic,” translate the note into a lever: longer duration, crane rise, volumetric light, slower push. When they ask for “more real,” shift style to documentary, reduce camera gymnastics, strengthen physics and secondary motion. Taste becomes operable when it maps to the anatomy above.
Finally, save the prompt that produced the hero take, not just the MP4. Files get lost; libraries compound. That is the entire PromptFork thesis applied to video.
A practical take log looks boring and saves hours: filename, duration, camera move, one-line subject, and the single change you made from the previous take. After ten generations you will see patterns — maybe your push-ins always need eight seconds, maybe handheld only works when the action is tiny. Those patterns belong in your forked recipes. Without a log, every session relearns the same expensive lessons and your “style” is just whichever clip you happened to like before lunch.
Sequence planning deserves the same honesty. Storyboard three shots on paper before you open Sora: wide establish, medium action, tight detail. Generate each as its own prompt with matching light and style language so the edit can cut. Trying to force all three into one generation is how continuity dies. Editors have always been part of cinema; AI video did not repeal that. What it did change is the cost of a take — low enough that discipline, not scarcity, is now the scarce resource.
Lighting continuity across a sequence is worth an extra sentence in every prompt in the set. If shot one is golden hour and shot two forgets to say so, the grade will fight you. Copy the lighting and style lines from the hero prompt into siblings and only vary subject framing and camera. That small habit is how AI-assisted spots start to look directed instead of sampled from unrelated lucky rolls.
Negatives evolve with your failures. The first week you ban text and watermarks. The second week you ban morphing hands because a product turn went wrong. The third week you ban “extra fingers on the fox” because nature clips have their own demons. Keep the avoid list short by retiring fixes you no longer need and promoting the ones that recur. A living avoid list is more valuable than a static mega-prompt you never reread.
If you collaborate, share prompts the way you share project files. A Slack dump of “try this vibe” is not a handoff. A forked PromptFork entry with subject brackets and locked camera language is a handoff. Your producer should be able to swap the product SKU without inventing cinematography. That is the bar: prompts as production assets, not chat folklore.
Continuity across shots: the hard problem video makes obvious
A single beautiful clip is a postcard. A sequence is a story — and stories break when jackets change color, rooms forget their windows, or the hero ages between cuts. Write continuity into the prompt as explicitly as camera move: same charcoal wool coat, same warm practical lamp stage left, same scuffed red sneakers. Put those constants in every shot of a sequence even when the action changes. The free shot builder is a per-clip tool; your library is the sequence brain.
Duration discipline matters more than people admit. Asking for a thirty-second epic in one generation invites mush. Prefer short beats you can cut: three to eight seconds of clear action, then a new shot with a new move. Edit in a real timeline. AI video is a camera department, not a substitute for structure.
When identity drifts, simplify. Fewer background characters, clearer hero framing, fewer simultaneous motions. When physics looks fake, name the material and the force: heavy canvas flag in strong side wind, not “epic flag.” Concrete physics language beats cinematic adjectives every time.
Questions people ask about Sora prompts
What makes a good Sora prompt?+
A good Sora prompt describes one continuous shot: who or what is on screen, what single action happens, how the camera moves, how long the beat lasts, the visual style, lighting, and what to avoid. Sora is a video model, not a chat essayist — montage language, multiple scenes, and vague adjectives produce mush. Concrete nouns, one action, and explicit camera grammar outperform poetic paragraphs.
How long should a Sora video prompt be?+
Long enough to lock subject, action, camera, duration, style, and negatives — usually a tight half-page of structured lines, not a screenplay. Overlong prompts with conflicting directions dilute each instruction. If you need a multi-shot sequence, write multiple prompts or clearly separate beats only when your workflow supports stitching; the builder on this page optimizes for a single coherent take.
How do I control camera movement in Sora?+
Name the move in cinematography language: slow push-in, pull-back reveal, orbital arc, tracking shot, crane rise, locked static, handheld micro-shake, whip pan. Pair the move with duration so the model knows how much time the motion has to complete. Avoid stacking three camera ideas in one shot unless you want chaos.
Why does my Sora clip morph or break continuity?+
Usually the prompt invited change: multiple identities, wardrobe swaps mid-shot, conflicting styles, or action that implies cuts. Add an explicit continuity clause — same subject identity and proportions first frame to last, single uninterrupted take, stable physics — and simplify to one action. Negatives help (“no morphing faces, no jump cuts”), but clarity of the positive action matters more.
Should I include dialogue in Sora prompts?+
Only if lip-sync and speech are core to the shot and your Sora workflow handles it well. Many strong clips are silent with ambient sound cues only. If you need dialogue, keep lines short and visual action primary; do not let speech replace a clear physical beat. For pure product or nature shots, ambient-only is cleaner.
How is a Sora prompt different from a Veo prompt?+
Both want subject, motion, and camera — but dialect differs. Sora often responds well to tight film-tag style lines (shot length, lens character, avoid list). Veo often prefers fuller scene prose with explicit temporal beats and physics language. PromptFork keeps separate builders so wording stays native: use this page for Sora and the Veo Shot Builder for Google Veo.
Do Sora prompts work for other video models?+
The shot grammar transfers: one subject, one action, camera, duration, style, negatives. Exact phrasing and strengths differ by model. Start from a clean shot brief, then tune dialect. For Veo-native phrasing, switch tools rather than forcing Sora tags into Veo.
Does the free Sora Shot Builder call an API?+
No. It assembles a deterministic prompt in your browser from your choices. You copy into Sora yourself. Supercharge with AI opens PromptFork Studio with the shot seeded for optional refinement under Studio’s free daily allowance.
What should go in the avoid list for Sora?+
Common quality sinks: on-screen text, watermarks, logos, jump cuts, morphing faces, extra limbs, unwanted zoom jitter, cartoon look on a photoreal request, low resolution. Keep the list short and targeted to failures you have actually seen — a novel of negatives fights itself.
What does forking a Sora prompt mean?+
Forking saves a proven shot recipe into your PromptFork library so you can change the subject and keep camera, style, and continuity language that already worked. That is how you build a personal shot library instead of reinventing cinematography prompts every time.
Direct the shot. Stop hoping the model invents cinema.
Build a continuity-safe Sora prompt in under a minute — or fork a shot recipe that already holds together.