You start with a clear portrait and make a short AI video. For the first few seconds, the haircut, jawline and expression all match the reference.

Then the camera moves closer.

By the final frame, new freckles have appeared. The eyelashes look heavier, and a faint shadow beside the mouth has become a deep crease. Nothing looks obviously broken, but the face is no longer fully supported by the source image.

The problem is not always weak identity control. Sometimes the shot asks the reference to provide details it never captured.

A better close-up workflow begins before generation: check how much facial information the image contains, then choose a camera distance within that limit.

Preview the Final Crop First

Before writing a longer prompt, crop the reference image to match the intended final frame.

If the video will end with the eyes and mouth filling most of the screen, enlarge those areas in the original photo. Can you distinguish individual eyelashes? Is the skin texture clear, or is it made from a few soft pixels? Is the dark area near the chin facial hair, a shadow or compression noise?

This preview shows whether the final shot will preserve existing detail or ask the system to invent it. Repeating “keep the exact face” cannot add information that the source does not contain.

Give the Camera a Safe Destination

Complete this sentence before preparing the shot:

By the end of the clip, the camera has moved from ______ to ______.

Use visible framing rather than vague language such as “dramatic close-up.” For example:

The camera moves from a chest-up portrait to a head-and-shoulders frame.

Imagine a man waiting on an underground platform. He looks down while a train approaches in the distance. The camera moves closer as he raises his eyes.

If the reference is a medium portrait, a head-and-shoulders ending may preserve enough information. An extreme close-up of one eye may not. The wider ending can still create tension without demanding unsupported skin and eyelash detail.

Choose the Right Path

After previewing the crop, choose one of three approaches.

Stop the camera earlier

Use this when the close-up adds mood but is not essential. Reduce the push-in until the important facial features remain readable in the reference. A slightly wider shot may feel less dramatic, but it is more likely to keep the person recognizable.

Add a genuine close reference

If the tight shot is necessary, prepare another authorized photo of the same person at a closer distance and under compatible lighting.

Let the wider image establish the hairstyle, clothing and general identity. Let the closer image provide evidence for the eyes, mouth and skin. The references should not disagree about makeup, facial hair, expression or light direction.

Generate wider and crop afterward

Create the shot at a safer distance, approve the face, and test a modest crop during editing. A post-production zoom cannot recover missing detail, but it enlarges an accepted frame without rebuilding the face throughout a moving close-up.

Avoid combining a strong generated push-in, heavy sharpening and a tight crop. All three place pressure on the same uncertain details.

Describe the Move Simply

Write the transition in three parts.

First, state the action:

The man slowly raises his eyes as the camera moves from a chest-up view to a head-and-shoulders portrait.

Next, state what remains stable:

Keep his facial proportions, short dark hair, jacket and neutral expression consistent.

Then add the environment:

Soft station lighting, a quiet background and restrained movement.

This tells the shot what moves, what stays fixed and where the camera should finish.

When preparing a short image-to-video draft in Vidu Q3, an optional final frame can help define that stopping distance. Keep the test short and use only enough movement to judge whether the reference supports the planned crop.

Check the Whole Move

Do not review only the sharpest final frame. Watch the opening, middle and ending.

Look for details that appear suddenly or slide across the skin as the head moves. Check both sides of the face. A mark that appears only when the lighting darkens may be a shadow or noise reinterpreted as facial texture.

Also review the clip at its intended display size. A small change hidden in a social feed may be distracting in an interview or identity-sensitive scene.

Final Close-Up Check

Before keeping the shot, ask:

- Does the final crop require more detail than the reference shows?

- Do new freckles, creases, hairs or eyelashes appear?

- Do those details remain fixed from frame to frame?

- Could the camera stop earlier without weakening the scene?

- Is a suitable close reference available if the tighter shot is essential?

- Would a controlled crop in editing be safer?

A reference image is not an unlimited map of a face. It is evidence from one distance, one angle and one lighting condition.

Plan the final crop first. If the source cannot support it, change the framing, improve the reference or separate the crop from the generated motion.

The goal is not only to create a realistic face. It is to keep that face connected to the information the reference actually provided.