Veo is available through more than one Google surface, and the active model determines which inputs, lengths and controls you can use. Check the model and settings shown in Flow or Gemini before building the shot around a feature.

Prepare one shot

Purpose: establish a quiet home-office scene
Subject: orange ceramic lamp on a walnut desk
Action: curtain moves gently; lamp remains still
Camera: slow push in
Frame: 16:9, subject right, clear space left
Audio: none required for this test
Reject if: lamp shape changes or false text appears

Choose the mode from the evidence you have

Starting material Useful route Main check
No reference Text to video Appearance and motion both need review
Approved first frame Frames to video Subject stability from the frame
First and last frames Supported frame route Transition plausibility
Existing clip Edit or extend only when current model supports it New section matches the original

Google’s Flow documentation lists features by model. Treat that table as current product state, not a permanent guarantee.

Prompt the motion

A warm ceramic lamp stands on the right side of a walnut desk in a quiet home office.
The sheer curtain moves gently in the background. The camera slowly pushes toward the lamp.
Soft morning light, restrained charcoal and orange palette, stable product shape.

Generate a baseline, then change one motion instruction. Save the model name and settings shown at generation time with the output.

Review before extending

Watch once for subject continuity, once for camera motion and once for artifacts at the frame edges. Only extend or assemble a clip that passes the original shot card.

Does every Veo model support the same controls?

No. Google publishes a current model-and-feature table for Flow. Confirm the active model, resolution, credit cost and supported input mode in the interface before generating.

What belongs in a Veo shot brief?

Name the subject, action, setting, framing, camera movement, duration and acceptance checks. Keep the first test narrow enough that one failure has a visible cause.

How do I compare two generations?

Use the same brief and score both against the same checks. Record the active model and controls because product versions can change the result.