References and output settings

Generation settings are creative constraints, not decorative preferences. Set them according to the final use of the asset.

Source media and references

You can add compatible media from your device or select existing brand media when the model accepts that input type. A reference can establish subject identity, product appearance, composition, style, motion, or dialogue—but one reference should not be expected to control all of those equally.

Use the fewest references that clearly communicate the requirement. Conflicting references can make the intended priority ambiguous. The model capability determines accepted media types and reference limits.

Aspect ratio

Choose ratio from the destination: vertical for stories and many short-form placements, square for balanced feed layouts, and horizontal for landscape surfaces. Recompose rather than relying on a destructive crop when the destination changes.

Resolution

Resolution affects output dimensions, processing, and sometimes estimate. Select it only after the format and concept are stable. A higher-resolution file does not repair a weak composition or inaccurate subject.

Duration

For video and audio, duration changes what can happen and how much the run may cost. Write a brief that fits the chosen time. A short clip should usually express one motion or moment rather than a complete multi-scene story.

Voice and audio mode

Voice selection appears only for compatible speech routes. Preview a voice when possible, then test it with representative brand language. Audio mode changes the type of instruction the system expects: a speech script is different from a music brief or a sound-scene description.

Preserve provenance

Keep the source asset and generated result distinct. Lip sync creates a new video rather than replacing the original. Workflow templates also retain reusable source inputs separately from generated examples.