Veo 3.1 Reference Images Guide

Use one to three reference images in Veo3o1 to guide a character, product, object, or visual style while creating an 8-second AI video.

Reference Images mode is designed for shots where the identity or visual language matters more than an exact opening frame. It can help a character, product, vehicle, room, costume, or color system remain recognizable while Veo 3.1 composes a new shot.

Current reference-image requirements

The Veo3o1 generator validates these rules before submission:

ControlCurrent requirement
Images1–3
Model presetLite or Fast
Duration8 seconds
Aspect ratio16:9 or 9:16
Auto aspect ratioNot available in this mode

The available resolution and exact credit cost are displayed in the generator. If a combination is unavailable, change the model, duration, aspect ratio, or image count before submitting.

Choose references that agree

Each reference should contribute useful information. Three conflicting images create ambiguity; one clean image can be better.

For a character:

  • use a clear portrait or three-quarter view;
  • include a wider view when clothing and body proportions matter;
  • keep hairstyle, wardrobe, age, and color treatment consistent;
  • avoid heavy filters, motion blur, and large occlusions.

For a product:

  • show the complete product in at least one image;
  • include a second angle for shape and materials;
  • use a detail image only when that feature must remain visible;
  • avoid mixing package versions or colorways.

Only upload images you have the right to use. Do not use private, copyrighted, or identity-sensitive images without permission.

Create with reference images

  1. Open the Veo 3.1 video generator.
  2. Select Reference Images.
  3. Choose Lite or Fast.
  4. Upload one to three compatible images.
  5. Select 16:9 or 9:16.
  6. Keep the duration at 8 seconds.
  7. Choose the output resolution.
  8. Write a prompt describing the new action, environment, camera, and fixed identity details.
  9. Review the displayed credit cost and submit.

Prompt structure

Use this order:

Referenced subject + one action + environment + shot size and camera movement + lighting + identity details that must remain fixed

Example:

The product shown in the reference images rotates slowly on a dark stone
pedestal. Close product shot at eye level with a restrained camera orbit,
soft rim light, stable bottle shape and label colors, no added text.

Do not ask for several locations, costume changes, time jumps, and camera moves in one short clip. Generate separate shots when the concept contains multiple beats.

Troubleshooting

Identity changes during motion

Use a cleaner primary reference, simplify the action, and reduce motion blur or long occlusion. Repeat only the most important fixed details in the prompt.

References produce mixed details

Remove the image that conflicts with the others. Check clothing, color, lighting direction, product version, and camera perspective.

The result is too static

Keep the identity language but add one physical action and one camera action. For example: “turns toward the window while the camera makes a slow push-in.”

Auto aspect ratio cannot be selected

This is expected. Reference mode currently requires an explicit 16:9 or 9:16 choice.

Reference Images vs First & Last Frames

Reference Images guide identity without locking the endpoints. First & Last Frames uses one or two images to control the opening image and, optionally, the ending image. Read the First & Last Frames guide before choosing a mode.

Completed videos use temporary result links and remain visible in My Works for up to 14 days. Download any result you want to keep.

On this page