Text defines intent
Use language for action, story beat, camera instruction, exclusions, and the final delivery format.
Video capability guide
Gemini Omni video input workflows
Plan text, image, audio, and source-video inputs around the creative decision each reference should control.
- scene and action
- Text
- appearance and framing
- Image
- voice and rhythm
- Audio
- motion and edit source
- Video
Explore workflows by input type
Build the whole shot from language
Use text when you want the model to invent the subject, setting, motion, and camera from a blank starting point.
Useful inputs
Subject and action
Environment and lighting
Camera movement
Aspect ratio and pacing
Give every input one responsibility
More references do not automatically improve the result. Clear responsibility prevents contradictory direction.
Images define appearance
Use high-signal frames for identity, product detail, wardrobe, composition, or visual style.
Audio defines rhythm
Use audio references for voice, music, timing, or emotional cadence when the selected workflow supports them.
Video defines motion
Use source clips for movement, timing, editing, extension, and actions that are difficult to specify with text alone.
Plan a multimodal request
Resolve conflicts before they reach the generation queue.
State the outcome
Write one sentence describing what the finished clip should communicate.
Assign reference roles
Name what each file controls and what it must not change.
Add preservation rules
Call out identity, product geometry, logos, camera axis, or timing that must remain stable.
Review the smallest draft
Validate the direction with a short clip before requesting a larger batch.
Input quality checklist
A clean input set is often more valuable than a longer prompt.
Use sharp, well-lit reference images
Trim source video to the relevant action
Avoid references with conflicting styles
Name the role of each input
Keep one dominant camera instruction
Review rights for every uploaded asset
Gemini Omni video questions
Which input type should I use first?
Use text for open exploration, an image for visual consistency, audio for voice or rhythm, and source video when motion or timing already exists.
Should I combine every available input?
No. Add an input only when it controls a specific part of the result. Extra references can create conflicting instructions.
How do I improve character consistency?
Use clear identity references, avoid conflicting faces or wardrobes, state what must remain unchanged, and revise one visual variable at a time.
Can I remove AI provenance markers?
This site does not offer a SynthID, C2PA, or watermark-removal workflow. Keep applicable disclosure and provenance information with generated media.
Put the workflow into practice
Move from input planning to generation and reusable prompt structure.