How to Use Veo 3.1 for Photoreal Video With Sound

Veo 3.1 is the model you reach for when the clip must look filmed rather than generated: real-world physics, believable skin and water, and the best native dialogue in the field. On ManifoldGen it runs as image-to-video in two tiers — Fast for iteration, Standard for finals — both with synchronized sound.

Start from a still, always

Both ManifoldGen routes are image-to-video, and that is the correct workflow anyway. Veo follows composition religiously, so a still gives you casting, framing, and lighting direction for free — then the video prompt only has to describe motion and sound. Generate your frame with any image model, pick the take you would have shot, and animate it.

Write motion like a director, not a caption

The chef turns from the window, wipes her hands on her apron,
and looks back at the rain. Camera holds static; depth of field
stays on her hands. Rain streaks the glass; kitchen ambience,
distant radio, one low thunder roll.

• One subject action per clause — "turns", "wipes", "looks" are three beats Veo will sequence faithfully.
• State the camera explicitly — "static", "slow push-in". Silence invites drift.
• Name what should not move — "depth of field stays on her hands" pins focus better than hoping.
• Score it — ambience, one music cue, or a weather layer. Unspecified audio defaults to room tone.

Dialogue that passes

Veo renders speech with lip alignment and even accent control if you name it. Keep lines under twelve words, put them in quotes, and attribute them: He says, "Lock the door behind you." Faces read best at medium close-up or tighter; dialogue in a wide shot wastes the feature.

Choosing durations

Four seconds suits inserts and social loops; six handles a full action beat; eight fits a line of dialogue plus a reaction. If a scene needs more, generate two eight-second takes with matched stills (same character frame, different camera) and cut — Veo keeps identity across cuts far better than across a single long generation.

Troubleshooting

• Physics show off — water, cloth, and fire are Veo strengths, but they need space. Tight frames force shortcuts; widen the shot instead of slowing the motion.
• Subject ignores the source image — your prompt re-describes appearance. Strip adjectives about the person; keep only motion and audio.
• Dialogue sounds dubbed — shorten the line and mark delivery ("quietly", "over the shoulder").
• Eight seconds sag in the middle — add one sustained ambient clause (rain continues, crowd murmur persists) to carry the middle beat.

Run it at /tools/veo-3-1-fast for drafts and /tools/veo-3-1 for finals; automation mirrors live at /api/video-generators/veo-3-1-fast and /api/video-generators/veo-3-1.