Image-to-Video Workflow: Turn a Still Into a Controlled Shot

Image-to-video generation gives a still image time, camera movement, and subject motion. The source frame carries composition and appearance; the prompt should explain what changes after that frame. This makes source preparation more important than decorative motion language. A clipped hand, unreadable object, inconsistent reflection, or ambiguous depth in the still can become more visible once it moves. NamiFusion's Video tab exposes different methods and model-specific media slots. Some routes require a first frame, some allow an optional first or last frame, some use multiple references, and video-editing routes take a source video. Follow the selected model's visible form rather than transferring controls from another model.

Prepare the first frame and shot brief

Choose a source image with a readable subject, coherent perspective, and enough room for the intended movement. Keep important limbs and objects inside the frame. For a camera push-in, leave resolution and detail around the focal area; for a pan, include plausible space in the direction of travel; for a character movement, avoid a pose already at an anatomical extreme. Remove text that must remain perfectly readable unless the shot is designed to keep it static. Confirm that you have permission to animate the person, artwork, logo, or product shown.

Write a one-shot brief with four parts: subject action, camera action, environmental motion, and end condition. Keep each part physically compatible. “Person walks left” and “camera orbits right while zooming rapidly” can ask too much of a short clip. Decide which motion owns the shot. A useful brief also says what stays still, such as product geometry, facial identity, background architecture, or the camera horizon.

Run the image-to-video job

1. Open AI Generator and select the Video tab.
2. Choose a video method and model, then inspect its material slots and required markers.
3. Add the still to the first-frame or reference-image slot shown for that model. If a last-frame slot is available, use it only when you can supply a plausible endpoint with matching subject identity, composition, and scene geometry.
4. Enter the motion prompt.
5. Set duration, ratio, resolution, sound, or other controls only when they appear for the selected model. In some image-to-video routes, aspect ratio follows the source image.
6. Review the displayed cost estimate and generate.
7. Wait for completion in history; changing the page does not make a pending job complete sooner.
8. Download the result or use an approved still/result as an input for a deliberately planned next shot.

Prompt exercise: describe change over time

This prompt is an exercise, not a claim that a displayed clip was produced from it: “Keep the ceramic bottle and printed label unchanged. Over the shot, the camera makes one slow 10-centimeter push-in. A soft highlight travels gradually along the bottle’s right edge while the background curtain moves slightly from airflow. The bottle, cap, label text, tabletop, and horizon remain fixed. Natural easing, stable exposure, no cuts, no extra objects, end on a steady close product frame.” Notice that it names one camera move, one subtle environmental change, protected elements, and an end state. Exact units may guide intent but do not guarantee measured physical motion.

Keep the ceramic bottle and printed label unchanged.
One slow camera push-in. A soft highlight travels along the bottle's
right edge; the background curtain moves slightly from airflow.
Bottle, cap, label, tabletop, and horizon remain fixed.
Stable exposure, no cuts, no extra objects, end on a steady close frame.

Review motion frame by frame

Watch once at normal speed for story and rhythm, once muted for visual continuity, and once frame by frame around transitions. Check the opening frame against the source, then look for shape warping, changing logos or faces, extra limbs, disappearing objects, sliding contact points, inconsistent reflections, exposure pulses, camera jumps, and unstable backgrounds. Confirm that the requested action begins and ends cleanly and that the final frame can cut to the next shot. Listen separately when sound generation is enabled; plausible visuals do not guarantee useful audio. Never assume a technically completed task is editorially approved.

Revise failures and control cost

For excessive motion, remove secondary actions and ask for a locked camera or one slow move. For frozen output, state one visible action with a clear direction and endpoint. For identity or object drift, use a cleaner first frame, shorten the shot if that control is available, and protect the defining details in the first sentence. For a jump between first and last frames, make the endpoint closer or omit it. For unstable backgrounds, simplify depth layers and environmental motion. Duration, resolution, accepted media, queue time, sound support, and credit cost vary by model and settings. Validate the concept with the smallest useful run, check the current estimate before submitting, and only increase quality or duration after the motion itself works.