Prompt writing guide
How to write AI video prompts that produce usable clips
A good AI video prompt is not a random list of cinematic words. It is a short creative brief that tells the model what to show, how the camera should move, what mood to create, and where the final clip will be used.
Use a simple six-part prompt formula
For most Omni videos, use this structure: subject, setting, action, camera movement, visual style, and final use case. For example, instead of writing “cool product video,” write “a matte black wireless speaker on a kitchen counter, soft morning light, slow dolly forward, subtle background parallax, premium ecommerce ad, vertical 9:16.” This gives Gemini Omni Flash concrete visual decisions to follow.
- Subject: the product, person, place, or scene that must stay recognizable.
- Camera: push-in, orbit, pan, tilt, handheld, drone-style, or locked shot.
- Use case: product ad, social hook, landing-page hero, tutorial opener, or concept test.
Text-to-video prompts need more scene detail
When you generate from text only, the model has to invent the whole visual world. Add details about the environment, lighting, composition, mood, and subject behavior. A prompt such as “a coffee cup” leaves too much open. A better version is “a ceramic coffee cup on a wooden desk, steam rising slowly, warm morning window light, shallow depth of field, slow cinematic push-in, cozy lifestyle ad.”
Image-to-video prompts should focus on motion
When you upload an image, the visual reference already defines the subject and composition. The prompt should not rewrite the whole scene. Focus on what should move and what must stay consistent: “keep the product centered, preserve the label, add a slow push-in, subtle reflections, soft studio lighting, no major shape changes.” This is especially important for ecommerce, portraits, real estate, and branded creative.
Avoid overloading the prompt
Many failed AI video generations come from asking for too many actions at once. If the prompt says the character should run, jump, turn around, change clothes, and move through several locations in five seconds, the output will often become unstable. Start with one shot, one subject, and one camera move. Generate a baseline, then add complexity gradually.
Prompt examples by use case
For product ads, describe the object, surface, lighting, and camera move. For social media clips, emphasize the hook, pacing, and vertical format. For educational videos, keep the scene calm and readable. For cinematic concepts, describe the atmosphere, lens feel, color palette, and motion style. The best prompt is specific enough to guide the model but short enough to avoid conflicting instructions.
How to iterate after the first result
Do not rewrite the entire prompt after one weak result. Change one variable at a time: camera motion, lighting, subject action, or aspect ratio. If the subject drifts, ask the model to keep it centered. If the motion is too strong, use words like subtle, gentle, slow, or locked composition. If details distort, simplify the scene or switch to image-to-video with a clearer reference.