The Screenshot-to-POV Ad Ideal for: turning a store listing screenshot into a faceless demo with hands and voiceover — no creator needed.
I attached a screenshot of my product from [MY STORE / Amazon / Etsy listing]. Create a 30-second UGC-style ad for it using the Higgsfield MCP — hands and voiceover only, no one on camera. The product is [PRODUCT NAME] — [WHAT IT DOES IN ONE LINE].
STEP 1 — RE-SHOOT FROM SCREENSHOT: upload the attached screenshot, then use gpt_image_2 (high quality, 3:4) with it as a reference: "Using the product in the attached screenshot as the exact reference — identical shape, colors, label, and logo; ignore any website interface, buttons, or surrounding text — recreate it as a clean photorealistic product photograph in [the setting where it is used], natural daylight, no watermark." Use this task ID everywhere; never pass the raw screenshot to the video.
STEP 2 — SCENE ANCHOR (gpt_image_2, high quality, 3:4): "First-person photo (POV) looking down at [the surface where the product is used — countertop, table, bathroom shelf], a person's hands resting on the edge, [one or two objects that belong there], bright window daylight, authentic iPhone photo, photorealistic."
STEP 3 — VIDEO (seedance_2_5, omni_reference mode, 9:16, 1080p, duration 30, generate_audio true, both task IDs as image_references):
"Continuous unedited 30-second vertical handheld iPhone video, one single take with no cuts, first-person POV over the scene from the reference. Only the hands and the [PRODUCT NAME] from the reference ever appear — no face, no body. A relaxed [male/female] voice narrates like they're talking to a friend, never like an announcer. 0-4s: hands lift the product into the light, turning it slowly. VO: '[HOOK — the moment you realized you needed this.]' 4-11s: hands show the key detail up close — [the label / the mechanism / the texture]. VO: '[WHAT IT IS, and the one spec or ingredient that actually matters.]' 11-18s: hands use it in real time — [THE CORE ACTION, described specifically]. VO: '[WHAT IT FEELS LIKE IN USE — the honest observation.]' 18-25s: hands present the outcome — [the result on camera]. VO: '[THE RESULT — what it replaced, saved, or fixed.]' 25-30s: hands set the product down centered in the light, still. VO: '[LOW-PRESSURE CLOSER — one sentence, no hype.]' Realism: real hand and skin detail, natural daylight, authentic amateur POV framing, no cinematic grade, no on-screen text. AUDIO: warm natural voiceover exactly on the quoted lines, real handling sounds synced to the hands' actions, dry room tone, field-recording quality; no music, no water droplets, no bubbles, no synthesized blips."
Customize the prompt: this is the fastest way to turn "I sell this" into a working ad — a screenshot goes in, a ready-to-use ad comes out, without showing anyone's face. Tell the model to ignore the interface around the product; otherwise, the ad may inherit stray buttons and price tags from your listing.
Seven rules for keeping it realistic
1 Product in frame within the first three seconds. Ads where viewers immediately understand what's being sold convert measurably better. Write the script with the product rising into frame during the hook line, never afterward.
2 Write the realism block or get generic AI content. "Pore-level natural skin texture, flat daylight, authentic amateur framing, no beauty filter, no cinematic grade" — leave this out and you'll get waxy skin, teal-orange bokeh, and an ad everyone ignores. The amateur look is the aesthetic; don't let the model go cinematic.
3 Quote every spoken word, with timestamps. Unquoted instructions like "she talks about the product" produce unintelligible speech. Quoted lines at key moments generate clean, lip-synced speech — and 70–80 words is the limit for 30 seconds.
4 Prohibit the sounds you don't want. End the audio specification with a short list of prohibitions — "no music, no water droplets, no bubbles, no synthesized blips." Native audio tends to add droplet tones; keep the list short in a concise prompt, because long prompts start to fail.
5 Expect the mirror lottery. Some generations come out mirrored like a real front-facing camera — authentic, but your label is reversed. It's consistent within the clip, so one ffmpeg hflip pass fixes every frame. Check the label before publishing, not afterward.
6 QA with still frames, not impressions. Extract a 3×3 contact sheet (ffmpeg tile filter) and check: exactly one product, no more than two hands, legible label, face matches the reference, no burned-in captions. Motion hides flaws that still images reveal instantly.
7 Volume is the strategy. Ad creatives lose effectiveness in about 72 hours. At approximately 390 credits per finished ad, generating a new variant twice a week costs less than an hour of a human creator’s work — same recipe, new hook, new creator, same product job id.
Want the complete system — not just the prompts? Playbooks for selling UGC ad packages to real brands. Start with the free community — the link is in the video description, right where you found this pack.