Google: Veo 3.1

Veo 3.1 is Google’s video-generation model for creating a short audiovisual scene from text or an image-led brief. Its production profile supports text-to-video, image-to-video and guided first/last-frame creation, with generated sound, asset references and a separate continuation workflow. Use it to develop a cinematic concept, a short product narrative or a visual segment with a defined scene, action and listening experience.

Describe the subject and the visible event first. Establish the surroundings, time of day and light, then define the camera’s viewpoint or movement and the intended visual style. An original scene brief might show a glass vessel on a dark studio surface, with a slow lateral move revealing a reflection while a soft environmental sound accompanies the reveal. Each instruction gives a visible or audible feature that can be assessed in the completed result.

Google’s prompting framework treats the subject, action, scene, camera and aesthetics as separate parts of the brief. Use the parts that serve the shot rather than adding every possible descriptor. Write direction in English for the documented prompt profile. If the scene has several actions, state their relationship and keep them proportionate to a short clip, so a reveal or reaction remains readable.

A starting image establishes the opening appearance and composition. Use it when the character, product or setting already exists as a visual asset, and describe the motion that should develop from it. A last frame adds a target composition for the ending. Prepare boundary frames that communicate a plausible transition and preserve the details that identify the main subject, then inspect the movement between those frames after generation.

Asset references serve another purpose: they guide a character or object’s appearance in a newly generated scene, while boundary frames define specific moments in that scene. Choose references that show the important visual traits and make each asset’s role clear in the prompt. The production model’s reference-image-to-video specification uses eight-second output for that reference workflow, separate from ordinary text-led duration selection.

The maker’s ordinary production profile generates four-, six- or eight-second clips at 24 frames per second. It lists landscape 16:9 and portrait 9:16 framing, up to four video outputs per prompt and output resolutions of 720p, 1080p and 4K. The output video format is MP4. Its image-led specification accepts JPEG or PNG, lists an input image maximum of 20 MB and supports 720p/1080p input configurations. Choose the output profile for the intended placement and keep the frame composition readable at that size.

Native sound can accompany the visible scene. Specify spoken lines, atmosphere and sound effects as parts of the same brief, naming the subject or event that produces each one. A dialogue instruction and a camera instruction answer different questions: one establishes the performance, the other establishes how the viewer sees it. Keep the words within the clip’s useful speaking time and review the sound for intelligibility and its relationship to the action.

For a continuation, the maker provides a separate video-extension workflow. Begin from an appropriate generated source clip and describe the next visible event or change in the scene. Preserve the original and continued segment so their joining point can be reviewed for appearance, motion and sound. The normal new-generation duration and resolution choices should not be treated as a universal contract for every extension operation.

A negative description can focus exclusions on the task’s visual requirements. For a series of product variations, keep the subject and composition stable while changing a single creative direction, such as the lighting or camera distance. Record the prompt, references and selected settings with the result. This makes it possible to compare a useful creative difference rather than a collection of unrelated generations.

The regular Veo 3.1 production profile and Veo 3.1 Fast have different output-resolution profiles. Fast lists 720p and 1080p; regular 3.1 also lists 4K. Keep the chosen edition attached to the result, without silently replacing a production model identifier with an older Preview profile or another Veo generation. Both the appearance of a subject and the supported operation depend on the selected workflow.

Generation is an operation whose completed result provides video artifacts for playback and later editing. Keep the returned files with their source frames, reference assets and brief, and preserve supplied content-provenance information. Inspect each actual output for the subject’s appearance, action, framing, sound and final composition. Use the chosen clip as a production building block, with titles, transitions or other finishing work added in the surrounding edit.