Product ad video

Produce a short ad or promo clip starting from a product still. The still anchors the product's exact look (shape, color, label, material), and the prompt directs the motion and camera around it. This is image-to-video for a flagship video model, run as an async task. For dropping a new product into footage you already shot, swap instead of generate (see Technique).

Inputs to collect

• The product still. A clean hero shot of the product. This is the anchor and is required. (Ask only if none provided.)
• The vibe and the motion. What the spot should feel like (sleek tech, warm lifestyle, energetic) and what should move (slow push-in, orbit, product rotating, hands interacting, scene reveal).
• Duration and aspect ratio. Most flagships run 5 to 10 seconds. Confirm the target (vertical for social, wide for landing pages).
• Optional: voiceover or music (route to voiceover), and whether there is existing footage to swap the product into rather than generate fresh.

Models

• Default flagships (image-to-video): Seedance 2.0 (bytedance:seedance@2.0), Veo 3.1 (google:3@2), Kling 3.0 Turbo (klingai:kling-video@3.0-turbo), Wan2.7 (alibaba:wan@2.7). All are live and accept a starting image. Pick by feel: Veo 3.1 for cinematic polish with native audio, Seedance 2.0 for physics-aware motion and multi-reference, Kling 3.0 Turbo for a faster cheaper draft pass, Wan2.7 for reference consistency with native audio.
• Swap a product into existing footage: Pruna P-Video-Replace (prunaai:p-video@replace) replaces one on-camera object in a source video using a reference image, keeping the actor, motion, timing, camera, and audio. Use this when you have a shot creator pitch and want to fan out one product variant per call.
• Confirm the live model, its capability tag (io:image-to-video for the flagships, io:video-to-video for the swap), and its exact input fields via runware-models + runware-run before calling. Do not hardcode a stale pick.

Workflow