Street interview style: one question, 5 answers, 40 seconds, real or generated
The vox-pop format: one short question, five contrasting answers, the last one lands the point. The prompt writes the question, the answer spread and, if generated, the per-person shot prompts.
Plan a street-interview video about {topic}. The format works because the question is one the viewer answers in their head before the first stranger does, and the five answers disagree. Two modes: REAL (I go out and ask) or GENERATED (each answer is a 5-second generated clip of a different person). Say which mode and I will plan for it. The question: max 10 words, answerable in one sentence, with no right answer that everyone would give. Write 3 candidates and pick one. Structure (40 s, 9:16): - 0-3 s: the question, spoken by the interviewer off camera over the first person's face, and on screen as text for the whole video (top third, small). - 3-33 s: 5 answers, 6 s each. Spread: one confident wrong, one unsure, one funny, one unexpectedly expert, one that is the point of the video. The last one is the point. On-screen: a 2-3 word label per person, no names. - 33-37 s: the interviewer's one-line reaction or a cut back to the best answer. - 37-40 s: 'What would you say?' No download ask; if the topic is a product, it is what the expert answer mentions, once. Word budget: 90-100 spoken words across the five answers (each answer 15-20 words) plus the question. REAL mode: return the question, the five answer types to look for, a 3-line note on framing (eye level, 50 cm, busy background out of focus) and the release line to ask. GENERATED mode: return the question and the five answers as lines, then five generation prompts, one per person, each with: a distinct person block (age range, look, one accessory), a street setting, medium close-up handheld, overcast daylight, the person speaking one sentence, 5 seconds, 'no on-screen text, no logos'. Mark the video as AI-generated in the caption. Topic: {topic} Mode: {real / generated} The point I want the last answer to make: {the point}
In Claude Code, paste the prompt as a message.
Example
Topic: how long it takes to learn enough of a language to order food. Mode: generated. Point: ten minutes a day for a month is enough to order.
Question: ‘How long until you can order dinner in a new language?’
Answers: 1 (confident wrong): ‘Years. I did five years at school and I still point at the menu.’ 2 (unsure): ‘A month? Six? Depends how much you… I don’t know.’ 3 (funny): ‘I just say the name of the dish louder. Works in every country.’ 4 (expert): ‘About two hundred words. Food, numbers, please, thank you. A few weeks if you do a bit every day.’ 5 (the point): ‘I did ten minutes a day for a month on an app and ordered in Lisbon without switching to English. Ten minutes.’
Generation prompt 4: A woman in her forties, short grey hair, a green scarf, standing on a busy European street, shops out of focus behind her. Medium close-up, handheld, slight natural shake, overcast daylight. She speaks one sentence to someone just off camera, a small shrug at the end. Street ambience, no music. 5 seconds, 9:16. No on-screen text, no logos.
(Four more prompts in the same shape, each with a different person block and a different street.) Caption ends: AI-generated video.
Tips
- In REAL mode, ask the question ten times and use five. The spread comes from volume, not from casting.
- In GENERATED mode, keep faces out of the first frame of each clip until the person is mid-sentence; a generated face held still for a second is where the viewer notices.
- Run the disclosure checklist before posting a generated version: label it in the upload form, not only the caption.
| works with | Claude Code: Runs as a skill or a prompt in the terminal; can read your files and write the result. · Claude: Paste into claude.ai or the Claude app. · ChatGPT: Paste into ChatGPT; works with any GPT model. · Any LLM: Model-agnostic: Gemini, Mistral, a local model. · Veo: Google text-to-video / image-to-video prompts. · Kling: Kling text/image-to-video prompts. |
| cost | Free. The video or image model you paste it into has its own pricing. |
| status | Unverified: written from the technique, not yet run end to end by us. |
| license | CC BY 4.0 for the text here; skills are MIT. |
| updated | 2026-11-01 |
| for agents | street-interview-format.md |
Related
Before you post a generated video: disclosure and rights checklist
Six checks for AI-generated clips before they go on TikTok, Reels, Shorts or a client's site: labels, likeness, music, logos, model terms and provenance.
Keep the same character across clips: the reference sheet prompt
A character description block you reuse verbatim in every prompt, plus a generated reference sheet, so a face, outfit and build stay the same from shot to shot.
Countdown listicle: 5 items in 45 seconds, best one last, numbers on screen
The ranked list format, with the rule that makes it hold: each item gets shorter as the count drops and number one gets the longest shot. The prompt returns script, text plan and the caption list.
POV: a 15-second first-person scene with the text doing the storytelling
The POV format in which the camera is the viewer's eyes and one line of on-screen text sets the situation; the prompt writes the text, the single action and the beat where it turns.
Reaction / duet style: respond to a clip in 30 seconds without stealing it
The split-screen or picture-in-picture response format, written so the reaction adds a fact or a demonstration instead of a face. The prompt plans the pause points and what you say at each.
'Stitch this' opinion: 5 seconds of their claim, 25 seconds of your answer
The stitch format in which a short borrowed clip sets up a claim and you answer with one argument and one demonstration. The prompt picks the 5 seconds to keep and bans the straw man.