UGC ad video

Produce a short, creator-style ad: a presenter talking straight to camera (testimonial, demo, unboxing) or a faceless voiceover over product footage. The output reads like organic social content, not a polished broadcast spot. The levers are a talking-avatar presenter, the script or audio that drives it, and the background/framing/captions tuned for the target feed.

Inputs to collect

• The format: direct-to-camera testimonial, product demo, unboxing, or faceless voiceover. This decides whether you need a presenter on camera at all.
• The script (for the TTS path) or your own audio file (for the lip-sync path). One or the other, never both.
• The product and the single claim the ad should land. (Ask for real specifics. Do not invent features.)
• Target platform and orientation: 9:16 for TikTok / Reels / Shorts, 16:9 for YouTube / web.
• Optional: a chosen avatar/voice, a brand background, whether captions should be burned in.
• Consent for a real person. If the ad uses a real presenter, voice, or likeness, confirm the user owns it or has permission. Do not impersonate a real or public figure without authorization, and disclose synthetic media where it could be mistaken for real.

Models

• Presenter on camera. Default: HeyGen Avatar V (heygen:avatar@5). A registered-look talking avatar with two drive modes (script-to-speech or your own audio) plus background, fit, and caption controls. Best general pick for a person reading the ad.
• Presenter alternative: OmniHuman-1.5 (bytedance:5@2) when you want to animate a specific supplied photo into a speaking presenter rather than pick from a registered catalog.
• Faceless voiceover: generate the spoken track with the voiceover skill (a TTS model), then lay it over product footage produced by product-ad-video. No avatar needed.
• Confirm each model is live and inspect its schema via runware-models + runware-run before calling. Never hardcode a stale choice, since the catalog changes weekly.