Reshoot the Same Scene from Multiple Angles · Let the Model Be Its Own DP

Gemini Omni FlashWorkflowCamera MovementRealisticStructured PromptImage to Video40s
Instead of describing a shot, this hands the video model a director-of-photography job: upload a ten-second source clip and ask it to reshoot the same action with N camera setups, with the people, location, lighting, and action timing all held fixed — the model picks the angles itself. The angle count is left as an [NUMBER] placeholder, so swapping in a different number makes the prompt reusable.
PROMPT · Video Prompt
Create with AI
Upload your video to google gemini or flow Adjust / Paste this prompt Use the uploaded video as the only source of truth. TASK Reshoot this exact same 10-second scene as a new [NUMBER]-angle cinematic recut. You are the director of photography. Analyze the action, blocking, emotion, and objects that matter. Then choose the [NUMBER] strongest camera setups a skilled DP would use for this specific scene. Do not ask me which angles to use. Decide them. HARD LOCKS - Same people, faces, bodies, clothing, hair, ages - Same location, set dressing, props, weather, time of day - Same lighting direction and color - Same action, timing, and outcome from second 0 to second 10 - Same physics and continuity - Do not add or remove characters - Do not change what happens, only how it is filmed - No text, subtitles, logos, or watermarks YOUR JOB AS DP 1. Read the scene: what is the subject doing, what should the viewer notice, where is the emotion or product payoff. 2. Choose [NUMBER] distinct, useful camera angles. Prefer a mix of coverage, not [NUMBER] slight variations of the same shot. 3. Typical useful mix when it fits the scene: establishing/wide, medium working shot, insert/detail on hands or object, character or POV, and a motivated move (orbit, push-in, over-shoulder, low hero, overhead). Drop any of these if they do not serve this scene and pick a better one. 4. Each angle must feel like a real camera with a real lens, not a random teleport. CAMERA RULES - Shot size, height, lens feel, and movement must change between angles - Moves must be motivated by the action (push in on a reveal, close-up on contact, overhead on layout, OTS on interaction) - Keep framing clean and stable enough to watch - If you cut between angles, make the cuts land on action - If a continuous move is stronger than a cut, use one continuous move that changes perspective - Do not invent a new story just to justify a flashy camera OUTPUT A single 10-second video containing exactly [NUMBER] camera angles of the uploaded scene. Cover the full original action. Do not freeze, loop, or skip beats. Photoreal cinematic, matching the source grade. NUMBER OF ANGLES: [NUMBER]
✍️ Editor’s Notes

The trick here is writing the prompt like a DP's work order instead of a shot description: TASK sets the baseline (keep everything the same, change only how it's shot), HARD LOCKS uses a ban list to freeze what must not change — people, location, lighting, timeline — and YOUR JOB AS DP flips that around, handing judgment to the model and letting it pick its own mix of angles, while offering a menu of establishing/medium/insert/POV/motivated-move options as reference rather than a script to copy. CAMERA RULES adds another layer of execution rules: moves need motivation, cuts need to land on action. The angle count is written uniformly as a [NUMBER] placeholder, so swapping the number reuses the whole template — pulling the variable out and exposing it separately is the single most useful trick to borrow from this kind of workflow prompt.

Create with AI
Related Works
×