Veo 3.1 and Omni Flash: frames, references and video edits

Verified sources and integration boundaries

Google's current Cookbook describes Veo 3.1 starting and ending frames, multiple references and extension, alongside Gemini Omni Flash 1.1 multimodal generation and conversational editing. These living examples should be checked for current API and regional constraints.

HUUUIAI exposes separate Veo3.1 frame and multi-reference workflows, and Omni Flash frame, reference and edit workflows. Similar names do not mean the inputs serve the same role.

Choose by the reference's job

Large identity, perspective or pose differences between endpoints make transitions harder. Align the references before asking for the movement.

• First frame: the supplied composition opens the clip.
• First and last frames: the model builds motion between endpoints.
• Smart reference: subject or style can move into a new setting.
• Video edit: an existing clip is the baseline for changes.

Continuous-shot prompt

In one continuous shot, slowly push forward from the reference first frame. Preserve the person's clothing and the objects on the table, ending at the close framing of the final reference. No cuts and no added text.

Review continuity and edits

Check both the requested change and the actions or sound that should remain. At extension joins, inspect direction, speed and lighting. Official example limits must not be substituted for the actual input and pricing controls in HUUUIAI.