Here is the workflow that turns one product page into a wall of ad creative in about half an hour. No design software, no freelancer queue, no blank canvas. The whole thing runs from a single product URL. We will use Paintgym for the walkthrough because it is the tool built for this exact flow, but the thinking applies wherever you generate creative.
Block thirty uninterrupted minutes. Have your product page open and a rough sense of who you are selling to. That is all the prep you need.
Step 1: Scrape your product (3 minutes)
Everything starts with good product data, because the briefs are only as sharp as what the model knows about you.
Paste your product URL and let the scraper pull the structured data: product name, price, description, key features, product images, and brand colors. You can see exactly what gets extracted by running the URL scraper tool on any product page before you commit.
Two tips here. First, point it at a clean product page, not a homepage. A product detail page has the price, the benefits, and the hero image in one place. Second, glance at what came back and fix anything thin. If the description is one vague line, the ads will be vague too. Paste a sharper benefit or two into the product details. Thirty seconds of cleanup here pays off across every ad.
Step 2: Pick your concepts (5 minutes)
You do not want all 49 frameworks for every product. You want the handful that match your offer and your buyer.
If you are not sure which to run, the concept picker answers five quick questions and recommends the five frameworks most likely to fit. Otherwise, choose a deliberate spread:
- One or two hero or benefit concepts for cold prospecting, like One Core Idea or Three Main Benefits.
- One or two proof concepts for the middle of the funnel, like Social Proof or a Reddit thread screenshot.
- One direct-response concept for retargeting, like the money-back guarantee or an objection handler.
The full menu and the logic behind each one is in our guide to all 49 concepts. For a first run, eight concepts is a good number. It is enough spread to learn something and few enough to review properly.
Step 3: Add a hook (2 minutes)
A hook is the angle that makes someone stop. The same product can be sold as a discovery, a guarantee, or a contrarian take, and which one wins is not obvious until you test.
Paintgym ships a bank of 20 proven hooks. You can browse them and see one filled in for your category with the free hook generator. Pick a hook that fits the concept. A curiosity hook like "nobody talks about this" pairs well with a bold claim. A social-proof hook like "I tried every option and this is the only one that works" pairs well with a reviews concept.
You do not have to attach a hook to every concept, but doing it on your prospecting ads sharpens the opening line the brief is built around.
Step 4: Write the briefs (5 minutes, mostly waiting)
This is the step that used to take a creative team a day. A Claude model reads your product data, the concept framework, and the hook, then writes a complete image-generation brief for each ad: the composition, the on-image copy in exact words, the layout, the color palette, and the framing.
You can read a real brief for your own product, before generating anything, with the brief preview tool. It writes one brief for the Bold Claim concept so you can see the quality and the structure.
While the briefs generate, skim them. A good brief names your real price and real benefits and never invents a claim or a testimonial. If one drifts off your product, that is your signal to tighten the product data and rerun. Briefs are free to write, so iterate here before you spend anything on images.
Step 5: Choose your image model (1 minute)
Paintgym renders with two image models: Gemini, also called Nano Banana Pro, and a GPT image model. They have different strengths. Gemini tends to handle product realism and packaging well. The GPT model can be stronger on certain typographic and layout-heavy concepts.
If you cannot decide, run both. Paintgym can generate the same concept with each model side by side, which is the fastest way to learn which one wins for your product. After a project or two you will have a feel for your default.
Note that the eleven screenshot concepts, like iMessage and the Notes app, do not use an image model at all. They render as real HTML and get screenshotted, so they cost nothing in credits and the text and price are always exactly right.
Step 6: Generate (5 minutes, mostly waiting)
Hit generate and the system renders your selected concepts. As each image finishes, a separate review model checks it for the common failure modes: garbled or misspelled text, the wrong product, a broken or unbalanced layout. Weak renders get flagged or regenerated instead of quietly landing in your folder.
This review step is the difference between an AI tool that produces a pile you have to sort and one that hands you usable ads. You still make the final call, but you are reviewing finished work, not triaging junk.
This is a good moment to start a second project if you have another product. Generation runs in the background, so you are not stuck watching a progress bar.
Step 7: Review, rate, and refine (5 minutes)
Now you have your wall. Go through it with a cold eye and ask three questions of each ad.
First, would this stop me in the feed. If the answer is no, it does not matter how on-brand it is. Second, is every word legible and correct. Third, does it actually feature my product, not a generic stand-in. Rate the strong ones and refine the close-but-not-quite ones. Refining feeds your note back into the model and produces a new version, so a small fix like "make the headline bigger" or "use the navy from my brand" is one click.
A useful habit: pick your two favorites per concept and export those. You are not trying to ship all 49. You are trying to find the three or four that beat your control.
Step 8: Export and launch (2 minutes)
Export the winners as high-resolution files sized for Meta. Bulk export pulls the whole set at once.
Then the real work begins, which is testing. Upload your spread, let them run, and read the results after a few days of spend. Kill the losers, scale the one or two that move, and come back to generate fresh angles around what worked. The thirty-minute creation cycle means you can do this every week instead of every quarter.
Tips that make the output better
A few habits separate a good run from a great one.
- Feed it real proof. If you have a specific stat, a real review, or genuine press, put it in the product details. The model will not invent these, by design, so the ones you provide are the ones it can use.
- Vary the hook, not just the concept. Two bold-claim ads with different hooks test two different ideas. Two with the same hook test the same idea twice.
- Trust the screenshot concepts for cold traffic. The native formats like iMessage and Reddit threads tend to punch above their weight on prospecting because they do not look like ads.
- Rerun seasonally. The occasion concept and a fresh hook turn an evergreen product into a timely ad without a new shoot.
The point of doing this fast
The reason speed matters is not the time you save on any single ad. It is that cheap, fast creative changes how you operate. When an ad costs a few dollars and a few minutes instead of a few hundred dollars and a few days, you stop rationing tests. You run more angles, find winners faster, and refresh before fatigue sets in. We broke down the cost and performance math in AI ads versus human-made ads.
Ready to run your own thirty-minute cycle. Walk through the full flow or start free and paste your first product URL.