UGC Prompt Studio
Give Claude a product photo. Get back eight different ad concepts, each with a shot list, a script, and a prompt the video model can actually execute.
Two files in. Five ads out.
This is one real job, unedited. I gave Claude a product photo and an avatar. Nothing else. It read the product's mechanics, wrote one handling sequence, and produced these as separate concepts with their own hooks, scripts and prompts.




What the flow actually looks like
8 stages. Each one makes the thing the next one needs, and every prompt comes labelled with the model it goes into.
- 01
Product sheet
YouOne product photo, phone quality is fine
ClaudeA GPT Image prompt that turns it into a clean reference
- 02
Creator
YouA face, or nothing at all
ClaudeAn avatar sheet prompt, or product-only concepts
- 03
Mechanics
YouNothing
ClaudeHow your product opens and works, read from the sheet
- 04
Handling
YouNothing
ClaudeOne handling sequence every shot has to follow
- 05
Strategy
YouBenefit, buyer, offer, claims to avoid
ClaudeHow the pack splits across different selling jobs
- 06
Concepts
YouHow many, and how long
ClaudeEight concepts: hook, setting, wardrobe, lighting, camera
- 07
Script
YouNothing
ClaudeSpoken lines, timed shots, SFX and a CTA per concept
- 08
Prompts
YouNothing
ClaudeA Seedance prompt and an anchor frame prompt for each
Made with it
More of the same pipeline. No camera, no crew, no location.
Routine order
Daylight, close on hands
Teaching order of use. A different selling job from the one above, not the same ad reworded.
Get ready with me
Mirror, morning light
The format people already watch, with the product carried inside it instead of announced.
Toner demo
Macro, product first
Pure mechanics. The lid comes off the way that lid actually comes off.
POV
First person, phone in hand
Same avatar, same products, a completely different camera relationship.
Where to run the prompts
The prompts are written for Seedance 2.5, which is what these were generated on. Any platform carrying it will run them.
What is in the download
The concept pack
Eight concepts by default, three to fifteen on request, each one a different selling job rather than the same ad reworded.
Product mechanics
Claude reads how your product actually opens and works from the reference, and refuses to invent a pump, hinge or dropper that is not there.
One handling sequence
A single deterministic way the product is picked up, opened and used. Every shot in every concept follows it, so the product behaves the same way twice.
Seedance prompts
One full, ready-to-run prompt per segment, written to the format the model actually respects.
Anchor frames
A GPT Image prompt for the first frame of each concept, so the video starts from something you chose.
Product and creator sheets
Prompts that turn a raw phone photo into a clean product reference, and an optional creator sheet for talking-head concepts.
Buy this if
- You sell a product and need short-form ads faster than you can film them
- You have tried AI video ads and the product came out wrong in the shot
- You want range to test against, not eight rewrites of one idea
- You have Claude and can install a skill
Do not buy this if
- You want cinematic story ads with characters. That is the other skill
- You have no product. This starts from a real thing you sell
- You want the ads made for you. That is the service, not this
The problem this solves
AI product ads fail in a specific, embarrassing way. The bottle grows a pump it does not have. The lid opens the wrong way. The hand closes on nothing. The ad looks fine until someone who has held the product watches it, and then it looks fake.
That happens because the model is guessing at mechanics nobody told it about.
This skill makes Claude read the mechanics off your reference first, write one handling sequence, and hold every shot in every concept to it. The product behaves like itself.
Eight concepts, not eight rewrites
The other failure is range. Ask any tool for eight ad ideas and you get one idea eight times: same hook shape, same framing, same energy.
A pack from this skill spreads deliberately across selling jobs and production modes. A problem-solution piece and an unboxing and a before-after and a talking-head are different ads that fail differently, which is the only reason to test more than one.
The rules it will not break
No captions inside the video prompt, because burnt-in text cannot be removed later. On-screen text stays a post-production plan.
No invented claims. It keeps to the proof you give it, which matters more in beauty and supplements than anywhere else.
One specific choice per prompt. A prompt that offers the model alternatives gets you an average of both.
What it is not
Not a script generator. You get shot lists and prompts, not paragraphs.
Not for cinematic story ads with characters and dialogue. That is Cinematic AI Ads, which is a different job and a different pipeline.
Questions
Do I need a product photo?
A clear description works, but a photo is much better. The whole mechanics step reads from what it can see, and a phone photo is enough to start.
Which models does it write for?
Seedance for the video and GPT Image for the sheets and anchor frames. The comparisons cover which platforms carry them and what they cost.
How long are the videos?
Fifteen seconds by default. Thirty and forty five are split into stitchable segments under fifteen seconds each, because that is what the model renders reliably.
Can I use the output commercially?
The skill is yours to use on your own and your clients’ products. You cannot resell or redistribute the skill itself. Whether a given clip is licensed for commercial use depends on the platform you generated it on.
UGC Prompt Studio
7 files, 606 lines. Instant download. Updates included.