How do you generate the reverse or opposite camera angle for a shot in AI video?

Generate the reverse angle in the same agent session as the hero shot: hold the master reference (character + location), then ask for the compositionally opposite angle — 180° flip, over-the-shoulder reverse, or behind-the-subject framing — and force the agent to surface any undefined production design on the new side before it generates. Same session, same references, opposite POV.

Run it as a four-step loop inside one session of the invideo agent (an agentic video tool with every current video model — Runway, Veo, Kling, Seedance 2.0 — available under one roof, so the same character and location references stay attached as you move between models).

1. Lock the hero shot as your master reference. Generate the forward shot first and save the strongest frame plus the character sheet and location plate that produced it. These are the anchors the reverse will inherit from — without them, the reverse will drift in face, wardrobe, and space.

2. Chain the opposite angle in the same conversation. Immediately ask for the matched reverse in the same session — "reverse on Marcus, 180° cut, OTS from behind him toward where she was standing, same lens, same lighting state." Staying in-session preserves geography and identity; opening a fresh chat resets context and you lose the spatial logic. Be explicit about the angle vocabulary you want: OTS reverse, clean single reverse, behind-the-subject, 180° flip, low-angle reverse — vague "opposite angle" prompts produce mirrored compositions, not true reverses.

3. Force the art-director pass before generation. A reverse angle exposes parts of the set the forward shot never showed — the wall behind the subject, what's on the desk you were shooting over, what the ceiling looks like. Tell the agent to surface those gaps and offer narrative-loaded options instead of inventing them. As Hridaye, invideo's creative director, puts it: "Reverse on Marcus — what's behind him? That near wall doesn't exist yet. What should it be?" Pick the option, then generate.

4. Route the shot to the right model. Reference-to-video on Seedance 2.0 is the strongest default for reverses because it accepts the master frame plus character and location references simultaneously and carries that context across the cut — start/end-frame extensions can't, because they only see the frame, not the world around it. Kling 3.0's multi-shot mode is useful when you want the forward and reverse generated as a matched pair in one go. For reconstructing a novel angle directly from existing footage you already have, Runway's angle-reconstruction tools are a fit. Let the invideo agent route — it knows which model holds the reference best for the shot you're building.

If the reverse still drifts (wrong wardrobe detail, new face, wrong room), the fix is upstream: the master reference wasn't complete enough. Add a close-up panel to the character sheet, or generate a quick wide of the room from a neutral angle to establish the full 360° geography, then re-run the reverse with both attached. One documented session produced a precise reverse angle with no reference image attached at all — the agent reconstructed it purely from the spatial logic of prior shots in the same conversation, which is the upper bound of what session continuity buys you.

These are the steps that work across most shots — exact prompt language and which model wins depends on whether you're cutting OTS, full reverse, or a low-angle counter.

Watch some of these to see what works for you:

Reverse on Marcus — what's behind him? That near wall doesn't exist yet. What should it be?

— Hridaye, invideo's creative director