UGC ads spy: one viral clip in, five Sora prompts out

Use this stack as a page: the UGC ad remaker.

The “UGC ads spy” template is a remake machine, not a spy agency. You give it a viral ad (TikTok, Instagram, YouTube), it names the hook and the format, then it spits five Sora prompts aimed at a brand-shaped UGC take.

The cards say 10× faster. That is a slogan. The useful claim is narrower: one reference, five prompts, then a generator that can actually shoot the clip. Speed is whatever your queue and your taste allow.

We track Sora as Sora, limited access, paid with no public monthly number on the matrix. Infographics say “Sora 2.” Do not invent a second tool page. If OpenAI ships a named generation inside the same product, it still lives on Sora until we split the row.

The stack:

1. Input — a URL or an upload you are allowed to analyze. We do not track TikTok as a tool.
2. Claude (or ChatGPT) — hook, script beats, what not to copy.
3. Sora — five video prompts, then the actual generations.
4. Swap lane — Kling AI, Runway, Pika, or Veo 3 when Sora is gated.
5. Still / ad layout lane — Midjourney or AdCreative if the “UGC” is actually a static you will later animate.

This is not us selling an ads-spy app. It is how you keep the analysis model from also being the camera.

Why pasting a viral ad into one chatbot fails

A language model can describe a hook. It cannot hold a face, a kitchen, and a product label across a five-second take unless you hand that job to a video model.

If you ask Claude to “just make the UGC,” there is no file. If you ask Sora to “figure out why this ad worked,” you burn generations on analysis it is bad at.

Copyright and platform rules still exist. Reverse-engineering structure (hook in 1s, talking-head, product insert, CTA) is not the same as lifting their footage, their voice, or their trademark. The spy metaphor is doing a lot of illegal-sounding work the pipeline does not need. Call it structure extract, then write your brand.

AdCreative is a different product: conversion-oriented ad stills, paid from $36/month, no free row on our matrix. Use it when the output is a Meta/TikTok static or a layout, not a Sora clip. Do not run AdCreative and Sora as if they were the same button.

The jobs (and what each must not do)

Input owns the reference, not the brand

Paste a public ad URL or upload a clip you have rights to study. Pull: first second, on-screen text, who is speaking, where the product appears, CTA, length.

Do not scrape private accounts. Do not feed a competitor’s cloned voice into ElevenLabs. Cloning without consent is still a scam vector — same warning as ElevenLabs, ChatGPT Voice, and Suno.

Claude owns the insights

Claude is paid from $20/month. Ask for a structured brief:

• Hook type (pattern interrupt, confession, demo).
• Script skeleton with timestamps.
• What is generic (you may reuse) vs theirs (you may not).
• Five shot lists that would survive a brand review.

ChatGPT is fine if that is your writing home ($20/month paid). Do not run both on the same brief and then average the prompts. Pick one.

Sora owns the five prompts — as video, not as poetry

Sora wants concrete visual language: camera, lighting, who is in frame, what moves. Five prompts should be five angles on the same product, not five synonyms of “cinematic UGC.”

Limited access means this lane dies for a lot of teams. That is not a moral failure. It is why the swap lane exists.

Kling / Runway / Pika / Veo own the fallback

Kling AI: free credit pool, paid from $10/month, native audio, clips up to 15 seconds on VIDEO 3.0. Good when the remake is a talking-head plus product insert.

Runway: free row, paid from $12/month. Camera controls if the viral ad is more “commercial” than “phone in a kitchen.”

Pika: free row, paid from $10/month. Fast iterations when you are still hunting the hook.

Veo 3: Flow has a free daily pool; Google AI Pro is $19.99/month on our matrix. Native dialogue when the UGC is a spoken testimonial.

See the longer handoff in the one-person creator stack when you also need stills and a brand voice. This post is the ads-spy slice: reference in, prompts out, one generator.

One remake, six steps

1. Save the reference. Note length and platform.
2. Claude (or ChatGPT) extracts hook / beats / constraints.
3. Human kills anything that is their trademark, face, or copy.
4. Model writes five Sora prompts (or five Kling prompts if that is the camera).
5. Generate. Pick one. Do not ship all five because the card said five.
6. Add your VO in ElevenLabs if the video model’s native audio is the wrong voice. ElevenLabs paid from $5/month.

What this stack is not

It is not a license to clone a creator’s face. HeyGen and Synthesia belong in the avatar post, not here, unless the brief is explicitly a talking avatar and you have a consented avatar.

It is not “Sora replaces your media buyer.” Media buying is still budget and placement.

The page with the three matrix slots is the UGC ad remaker — Gemini reads, Runway (or Sora) shoots, Descript cuts.