Google Flow Ingredients to Video: A Practical Reference Guide

Google Flow lets you add image or video references to a prompt and use them as ingredients. You can upload media, select an existing project asset, or type @ to find an asset by name.

Unlike text-only generation, this workflow gives the model a visual signal for the character, object, or location that should be reused.

Google recommends subject and product references on a plain or segmented background. Location and style references should not contain unwanted extra subjects.

Keep lighting, perspective, and visual style compatible across ingredients. Conflicting references are harder to blend and can increase drift.

State which ingredient is the character, prop, or environment and describe what it should do. Text instructions should complement rather than contradict visual references.

Give assets recognizable names and use the correct name after @. Across shots, lock the identity description and change only action, camera, or environment.

Prepare a shot-list prompt file, verify that every reference is attached to the correct shot, and test a small group first. Scale only after character, product, and style consistency is acceptable.

Track jobs individually, retry weak shots, and save prompts with filenames. One incorrectly used ingredient should not force a restart of the entire sequence.

For character drift, reduce reference count, use clearer images, and remove contradictory guidance. For product distortion, prefer a front-facing clean reference and state which details must remain unchanged.

Availability and supported duration depend on model, Google AI plan, and region. Check the active model before submitting a large batch.