Canonical: https://aivideomodel.org/guides/image-to-video-workflow

Reviewed: 2026-10-25

Production

# Turn a still image into a controlled video shot

For image-to-video, use the source image to establish the visible composition and focus the prompt on what moves. Start with a small motion, inspect identity and geometry, then expand the shot.

By [AI Video Model editorial](https://aivideomodel.org/about#editorial) · Reviewed October 25, 2026 · Source-based guide

## Choose a reference that supports the intended move

Use an image you own or are authorized to use. Check focus, visible edges, product shape and the space around the subject. A tight crop may suit a small push-in but give a wider camera move very little context to reveal.

Decide the final frame shape before generating. If the input will be cropped into a portrait frame, inspect that crop first. Keep critical objects and faces away from edges that may be lost. A low-detail source is also a poor basis for judging whether a model preserves fine detail.

## Separate subject motion from camera motion

Write two short sentences in your shot plan: what the subject does, and what the camera does. For example, “The perfume bottle stays still. The camera slowly tracks sideways.” This makes the source of any change easier to diagnose.

If everything moves at once, you may have trouble identifying whether a warped label came from object motion, changing perspective or an invented surface. Begin with one moving element and add complexity after you have a usable baseline. This is a production method, not a guarantee of better output.

## A repeatable tabletop product test

Prepare an unbranded bottle on a plain table under soft side lighting. Choose a short output and a gentle camera move. An original test prompt is: “The camera slowly tracks from left to right across the bottle. The bottle remains fixed on the table. Lighting stays constant; one continuous shot.”

Inspect the cap, silhouette, reflection, contact shadow and background edges. Play the clip normally, then pause before, during and after the largest viewpoint change. If the bottle’s geometry changes, reduce the camera move or provide a more informative reference when the tool supports it.

## Diagnose failures by what changed

| Visible failure | Next controlled experiment |
| --- | --- |
| The subject changes identity | Reduce motion; use an authorized reference with clearer defining features. |
| The whole image feels frozen | Request one observable movement with a clear endpoint. |
| The background bends | Try a smaller viewpoint change and inspect straight lines. |
| Text becomes unreadable | Generate a clean visual plate and add accurate typography in an editor. |
| The ending does not connect to the next shot | Plan an ending state or use supported end-frame controls. |

These are suggested experiments, not verified repairs for every model. Keep the original attempt and note what you changed. If several trials fail the same essential requirement, change the shot design or tool rather than endlessly extending the prompt.

## Approve the moving image, not its thumbnail

A polished first frame can hide failures later in the clip. Review the entire shot at the intended aspect ratio and playback size. Check hands, small accessories, edges, reflections and the final frame that will join the next shot.

Save the input image, prompt and settings alongside the approved export. If you upscale or retime it, retain that intermediate step in your production record. The delivered file may have different dimensions from the base generation, so label them separately when comparing tools.

## Sources and verification

Reviewed October 25, 2026. Provider documentation supports specifications; it is not evidence of an independent performance test.

- [Image to Video Prompting Guide — Runway ↗](https://help.runwayml.com/hc/en-us/articles/48324313115155-Image-to-Video-Prompting-Guide)
- [Gemini Omni generation and editing guide — Google ↗](https://ai.google.dev/gemini-api/docs/omni)
- [Model and API specifications — Luma ↗](https://lumalabs.ai/llm-info)

## Keep exploring

- [Write text-to-video prompts that describe a filmable shot](https://aivideomodel.org/guides/text-to-video-prompts)
- [Plan AI video product ads around the product facts](https://aivideomodel.org/guides/ai-video-for-product-ads)
- [AI video model comparison](https://aivideomodel.org/compare/ai-video-models)