The reference image method

1. Step 01 · Claude

   Design the setting and pull references

   Paste into Claude, then fill the brackets:

   Step 01 promptCopy

   I'm making a UGC-style video ad for [product] aimed at [buyer: age, situation, what they care about].
   
   Design the setting for this ad in three parts:
   1. Scene: where this person would actually film a quick video about this (give 3 options, e.g. living room couch, kitchen, car).
   2. Character: who is talking (age, look, energy). Give 3 options and a one-line reason each, tied to the buyer.
   3. Wardrobe: what they'd really be wearing in that moment. Give 3 options and a one-line reason each.
   
   Then for the scene I pick, search Google Images and Pinterest and give me 3 to 5 real photos, with links. I'll use one of them as-is as the background.
   Only real, candid phone photos. No AI images, no studio shoots, no stock photos with perfect light.
   For the character, don't pull photos of real people. Write me a Higgsfield Soul prompt under 50 words for a new face:
   age, hair, eyes, skin, clothes, where they are and the light, ending with "Style: candid photograph."

   What I picked for the example: a lived-in kitchen on an overcast day, a woman about 48, a grey half-zip pullover with black leggings and a coffee mug.

2. Step 02 · Higgsfield Soul

   Make the face

   Step 02 promptCopy

   American woman, 48, warm and approachable, shoulder-length dark brown hair tucked behind her ears, a few grey strands, brown eyes, natural skin with fine lines, no makeup, grey half-zip pullover, standing in her home kitchen, mid-sentence with a small smile, overcast window daylight. Style: candid photograph.

   I ran it 4 times, changing only the hair and face details, and kept the one that looked most like a real person.

   Why Soul: when I let an image model invent the face from other people's photos, the skin came out waxy and too sharp. Soul makes faces that read like a real photo.

3. Step 03 · GPT Image

   Put her into the real photo

   Crop the scene photo to 9:16. Attach in this order: 1 the scene photo, 2 the Soul face. Then:

   Step 03 promptCopy

   Image 1 is the real photograph to keep exactly: the same kitchen, light, colours, grain and every object. Image 2 is the woman (keep her face, long dark hair with grey streaks and skin exactly). Add her standing on the tile floor between the island and the stove, a few metres back from the lens, visible from the top of her head to her knees with clear space above her head, slightly right of centre, holding a white coffee mug at chest height in her right hand, turned to the lens mid-sentence with a small easy smile. She wears a plain heather-grey half-zip pullover and black leggings. Style: an untouched real photograph with natural grain and true-to-life colour; she looks like an ordinary real person with natural skin, matching the light, perspective and sharpness of the kitchen.

   The room is a real photo, so it reads real. Keep her a few metres back from the lens: close-ups are where AI skin gives itself away.

4. Step 04 · Claude

   Turn the still into a video prompt

   Attach the still to Claude and paste:

   Step 04 promptCopy

   This is the first frame of a UGC ad clip. Write a video prompt for [Seedance 2.0 / Kling] that starts from this exact frame.
   The person says exactly: "[line]"
   
   Include:
   - A style line: raw phone footage, no colour grading, no cinematic lighting, no on-screen text or captions.
   - Camera setup: who holds the camera and how it moves (e.g. a friend walking slowly toward them), and that nothing of the camera is visible.
   - Environment: at least 3 real objects from the frame and where the light comes from.
   - A timeline in 2 to 3 second blocks. Start each block with the words being said, then what the face and hands do.
   - The exact dialogue, the voice (age, accent, tone) and the ambient sound.
   - An ending that cuts mid-action, no pose, no wave.
   Keep it under 300 words.

5. Step 05 · Seedance 2.0, 5 s, 9:16

   The video prompt I used

   Start from the still as the first frame:

   Step 05 promptCopy

   Image 1 is the opening frame and the woman in it: keep her face, long dark hair with grey streaks, the plain grey half-zip pullover, the black leggings, the white mug in her right hand and the kitchen exactly: white shaker cabinets, the speckled granite island, the stainless stove and microwave, the fruit bowl and the window with blinds on the left.
   
   Camera setup: filmed by a friend who starts about three metres away at her chest height, holding the camera in both hands. Nothing of the camera rig is visible anywhere in the shot. Camera physics: the friend walks slowly toward her for the whole shot, a steady handheld push in with a natural low tremor and a soft step bounce, ending about a metre closer on a waist-up frame. No zoom, no cut.
   
   Voice: an American woman in her late forties, warm, slightly husky, relaxed and a little amused, like talking to a friend, not announcer-like.
   
   0-2s: On "Okay, real talk" she lifts the mug a little, her eyebrows rise and her eyes crinkle, she glances down at the mug and straight back up to the lens.
   2-5s: On "Forty-eight" she gives a small head shake with a half smile and taps her chest once with her free hand; on "feel like mine again" her lips press into a closed smile and she breathes out through her nose.
   
   She says exactly: "Okay, real talk... Forty-eight, and my mornings finally feel like mine again." Do not paraphrase and do not add words. Only she speaks.
   
   Ambient: fridge hum, a spoon clinking somewhere, quiet room tone, no music.
   
   Style: real handheld footage, true-to-life colour, natural skin, an ordinary real person, plain and unmarked. No on-screen text, captions or subtitles.