Plan one two-person shot in Google Flow

Write the shot's purpose before opening the controls

Suppose you have two separate portraits, a doorway photograph and a drawing of the final embrace. These assets do different jobs. The portraits describe who appears; the doorway describes a setting; the drawing proposes where bodies finish. Before opening Google Flow, decide whether the shot must preserve an exact composition or create a new composition around recognizable references.

For an original planning exercise, imagine a survivor waiting inside the doorway while the other partner approaches from outside. If you only know the people and location, a reference-led shot leaves the arrangement to be generated. If you already composed a starting image and a compatible finishing image, a frame-led shot expresses the transition between those specific states.

Write which decision is fixed and which is open. An input record might say: identities supplied separately, setting supplied separately, camera placement open. Another might say: starting composition fixed, ending pose fixed, movement between them open. This distinction is more useful than placing every available picture into whichever upload slot appears first.

Match your assets to the current creation help

Google's current Flow creation help documents frame-based and reference-based ways to guide video. Its editing help separately covers operations such as extending footage, saving a frame and building scenes. Read the current help for the option available in your account, rather than assuming that a menu shown in an older walkthrough still applies.

Reference assets describe the people or setting you want a new shot to use. A starting frame fixes a particular composition from which motion begins. An ending frame expresses a desired destination. These inputs can overlap in subject matter while serving different purposes. Two portraits entered as start and end images should not automatically be described as two independent cast identities.

For existing footage, separate editing from extension. Google Flow currently limits Extend to Veo-generated clips; an uploaded reference film is not automatically eligible. Extended clips also cannot use the insert, remove or camera edit modes described in the same help page. If you already have the present-day and memory clips, arranging them in sequence is an assembly task rather than a reason to generate continuation.

Give every reference a clear role

Prepare a short input record before selecting assets. For a reference-led shot, label the two people separately and say which image describes the environment. For a frame-led shot, name the start and intended end state. The photo-selection guide covers portrait quality; this record concerns what each asset is supposed to control in this particular shot.

Avoid packing incompatible visual decisions into the same input set. A character wearing a scarf in one reference and no scarf in another creates an unresolved wardrobe choice. A front-facing portrait and an extreme side-view endpoint may also demand a large change in viewpoint. Decide whether those differences are intentional before treating an unexpected result as a prompt problem.

Choose Ingredients or Frames, then create the shot

In a Flow project, open the model selector, choose Video, then choose Ingredients for visual references. Add the relevant assets and describe each asset’s role in the shot. Select the available video preferences and generate. For the doorway case, keep the two people and setting separately identifiable in your notes; do not assume that the order of attachment creates an undocumented identity assignment.

For a composition-led version, choose Video and Frames instead. Place the composed opening image in the start slot and the compatible finishing image in the end slot, then describe the movement between them. Review the selected model and available settings before generating. Region and image-type restrictions can affect these options. The following case is a proposed input choice, not a tested result.

Review the middle as carefully as the first frame

Review the output against the input decision. In an Ingredients attempt, inspect whether both references remain distinct and whether the chosen location was interpreted coherently. In a Frames attempt, compare the opening and ending compositions with the supplied images, then inspect the movement between them. Matching the two endpoints does not prove that the intervening bodies moved plausibly.

For a later embrace shot, inspect approach, contact and release as separate stages. Hands can become hidden behind shoulders; that occlusion is where a still frame stops telling you enough. Review both bodies through the overlap, including the partner farther from the camera. A result that preserves one face while changing the other has not met a two-person acceptance condition.

Also inspect the proposed end frame as a cut point. Does the shot settle long enough to connect to another clip? If it ends mid-turn, you may need a different trim or a simpler action. More footage will not automatically make a difficult cut easier.

Use the failure to choose one revision

When a shot fails, classify the problem before changing the prompt. An unclear role assignment calls for a better input record. An impossible endpoint calls for a compatible pose. Too much simultaneous movement calls for a shorter action brief. A camera move that hides Partner B calls for a different composition or a fixed camera.

For example, if a frame-led attempt swings the camera because the end image shows the doorway from another angle, replace that endpoint with a compatible view while keeping the start image and action brief fixed. If the join improves, the comparison identifies an input conflict. Use the separate prompt guide for general motion wording; this experiment tests the role of the endpoint.

Record the selected workflow and relevant model or mode shown in Google Flow at the time of the attempt. Documentation describes supported operations, but it cannot establish the quality of your particular two-person output. The review log needs the result you actually inspected.

Build the sequence from accepted shots

Once the shot passes, decide what the next image must inherit. It may need the same screen position, a similar embrace silhouette or a specific wardrobe detail. Google's editing help documents saving a frame and building scenes; use those operations where they fit your continuity plan. An accepted still can describe a connection, while the next full clip still needs its own review.

Choose extension only after checking the clip’s eligibility and the later edits you still need. Choose assembly when the accepted shots already exist and your task is their order or trim. A memory reveal may need only two compatible cut points. Extra generated footage can introduce new continuity problems without resolving the relationship between the existing shots.

The Zombie Couple site itself currently provides a local two-photo preview and reference-film playback; generation and checkout are disabled. This article describes an independent Google Flow planning workflow. Your next step is to make one shot brief, choose its documented input mode and inspect that complete result before spending effort on a longer sequence.

• Give the shot one action with a visible stopping point.
• Identify what each reference or frame is intended to control.
• Compare both people through the entire motion, including overlap.
• Save the accepted shot and its input record before planning the next scene.