Disclosure: Shhots AI is our product. Credit costs below are the published rates as of August 2026.
This is a working prompt library for AI ad generation: copy-paste prompts organized by job, UGC ads, product photography, video ads, image ads, and real estate. Each tied to a real workflow, output format, and credit cost rather than a hypothetical. Every prompt here runs against Shhots AI through its MCP server from Claude, ChatGPT, Gemini, Cursor, or Perplexity (Pro and Scale plans), and the patterns transfer to any tool that takes a product photo plus instructions.
One framing note so the library makes sense: the Shhots web app itself is deliberately promptless. You pick named formats and presets, and add at most a one-line comment. Prompting happens in two places: the chat channel (where you drive generation from an AI assistant) and the optional comment fields (one specific instruction per generation). The library covers both.
How to use these prompts
Three rules carry every prompt on this page:
- Give the audience, not the aesthetic. “For skincare-curious men 25-34 who have never used a serum” beats “make it look premium.” The model translates audience into aesthetic better than you can specify it.
- List your allowed claims explicitly. Every claim your ad may make goes in the prompt; everything else stays out. This is what keeps generated ads compliant with your own legal review.
- Ask for the script before the render when the offer is nuanced. A script draft costs nothing; a re-render costs credits.
UGC ad prompts
A 10-second UGC ad costs 551 credits (~$4.50 on Pro); 15 seconds costs 826.
The workhorse UGC ad, the highest-performing default (UGC, 15s, 9:16):
Here's our product photo. Create a 15-second 9:16 UGC ad.
Audience: [who they are, one line].
Allowed claims: [claim 1], [claim 2]. Use no other claims.
Draft the script first and wait for my approval.
The three-format creative test, one script, three treatments, ready for a Meta test:
Take the approved script and generate it three ways: UGC, lifestyle,
and unboxing ASMR. Same product, same claims, 9:16, 15 seconds each.
The hook-variant batch, one-variable testing on the opener:
Keep everything about the approved UGC ad except the first line.
Write 5 alternative opening hooks: one problem call-out, one price
anchor, one before/after, one question, one bold claim. Show me the
five hooks before generating anything.
The objection ad, retargeting viewers who didn’t convert:
Create a 10-second 9:16 UGC ad that answers this objection head-on:
"[the objection, verbatim from reviews or comments]".
Audience has already seen the product once. Allowed claims: [list].
Product photography prompts
A product image costs 11 credits (~$0.09); a typical 4-image shoot is 45.
The studio hero shot:
From this product photo, create a 1:1 studio product shot on
[surface — travertine / marble / linen], soft directional light,
shallow depth of field, space above the product for headline text.
The lifestyle in-use scene:
Same product: a 4:5 lifestyle image of it being used by
[audience person] in [setting], golden hour, candid rather than posed.
Keep the label legible.
The seasonal refresh, same SKU, new campaign, no reshoot:
Rebuild the approved studio shot for a [holiday/season] campaign:
[palette/props, one line]. Product, angle, and framing stay identical.
The edit-in-place, fixing a generation instead of regenerating:
Warmer light, and remove the [unwanted object] in the background.
Everything else stays.
Video ad prompts (product, cinematic, faceless)
A 10-second product video costs 577 credits; 15 seconds 852.
The product-motion listing video:
From this product photo, create a 10-second 16:9 product video for a
marketplace listing: slow orbit, one feature call-out on screen
([feature]), end on the product with logo. No voiceover claims beyond:
[allowed claims].
The cinematic brand cut:
Create a 15-second cinematic ad from this photo. Mood: [two adjectives].
One location, dramatic lighting, no text on screen until the end card.
Script it first.
The faceless short, for channels and organic social:
Turn this outline into a 60-second faceless 9:16 video with AI
voiceover and on-screen captions: [3-bullet outline]. Conversational
read, no music bed under the voice.
The product-speaking pattern interrupt:
Make the product itself deliver the pitch: 10 seconds, 9:16, the
product "speaks" the script. Script: [2 lines, first line is the hook].
Image ad prompts
The static ad set, every placement from one generation session:
From the approved studio shot, create static image ads in 4:5, 1:1,
and 9:16 with this headline: "[headline ≤40 chars]" and CTA
"[CTA ≤15 chars]". Keep text inside platform safe zones.
The comparison frame:
Create a 1:1 before/after image ad: left side [the problem state],
right side the product result. Label the halves "[Before]" / "[After]".
No competitor names or products in frame.
Real estate prompts
Real Estate Studio is preset-driven, these are comment-field lines that steer a preset, not full prompts. Each generation costs 55 credits (~$0.45).
- Staging, keep one item:
Keep the existing rug; stage around it. - Staging, buyer profile:
Stage for a young family; nothing fragile at floor level. - Correction, hold the mood:
Keep the warm evening tone; just fix the verticals. - Removal, scope limit:
Remove the car from the driveway only; leave the basketball hoop. - Walkthrough script toggle: turn the price line off in the script editor when the listing price isn’t public yet. It’s a toggle, not a prompt.
Prompts that fail, and why
The anti-patterns we see most in generated-ad output:
- Aesthetic-only prompts (“make it pop”, “premium feel”). The model has nothing concrete to hold, so you get generic gloss. Give audience + claims instead.
- Unlisted claims, if the prompt doesn’t fence the claims, the script may invent benefits your legal review never approved. List them, every time.
- Multi-variable changes, “new hook, new format, new audience” produces an ad you can’t learn from. Change one thing per generation; the testing-matrix guide covers the discipline.
- Re-prompting instead of editing, for small fixes, an edit instruction (“warmer light, remove X”) modifies the previous result directly instead of rolling new dice.
FAQ
Do I need to write prompts to use Shhots AI?
Not in the web app. It runs on named formats and presets with an optional one-line comment. Prompts matter in the chat channel, where the MCP server lets Claude, ChatGPT, Gemini, Cursor, or Perplexity drive generation, and that’s where this library applies. MCP access ships on Pro ($49/mo) and Scale ($99/mo).
What makes a good AI ad prompt?
Audience, allowed claims, format, and one creative constraint, in that order of importance. The consistent losers are aesthetic adjectives without an audience, and prompts that leave claims unfenced.
What do these generations cost?
As of August 2026: product images 11 credits ($0.09), real estate edits 55 ($0.45), 10-second UGC ads 551 (~$4.50 on Pro), 15-second UGC 826, product videos 577/852 for 10/15 seconds. Credits come from one pool: Starter $19/mo (2,000), Pro $49 (6,000), Scale $99 (12,000).
Can I use these prompts with other AI ad tools?
The structure transfers (audience + claims + format + one constraint), though other tools won’t have the same formats, claim-safety guardrails, or credit quotes. The prompts as written assume a tool that works from a real product photo rather than generating a fictional product.
How do I run these from Claude or ChatGPT?
Connect the Shhots MCP server (a 2-minute setup covered in the Claude AI video generator guide), upload a product photo in chat, and paste any prompt from this page. You get a credit quote before anything renders.