Ads that know your product is software.
Skip the stock-photo generators. Point AdForge at your app, and it learns your brand, writes the copy, and renders a finished ad — showing you what each one costs before you click.
- Runs with zero API keys
- Every render priced before you click
- Formats that reorder as you judge them
How it works
- 01
Point it at your site
The page is read and the brand kit fills itself in — audience, tone, pain points, the vocabulary you actually use, a palette. You edit rather than author.
Brand kit · modulate.proAudienceSolo founders shipping without a design teamToneDirect, technical, no hypePaletteWorks offlineNo account requiredExports to CSV - 02
Say what you make
Physical, mobile app, web app, or service. This one field decides whether the model is told to photograph an object or render an interface on a device.
What is it?Physical productMobile appWeb app or siteServiceRuns in a browser. Ads show the interface on a laptop or desktop screen.
- 03
Pick a format, see the price
Every model priced for the exact job, with the right one marked. The figure sits on the button, so cost is a decision rather than a surprise.
Model · 4 variants- Flux 2 Pro$0.020/img$0.080
- Nano Banana 2$0.040/img$0.160
- Ideogram v3$0.070/img$0.280
- Nano Banana Pro$0.240/img$0.960
Generate · $0.280 - 04
Read the brief, run it again
Each image ships with the prompt that made it, the model that ran it, and what it cost. Edit the brief and generate again.
View prompt- SUBJECT
- modulate.pro — a web application, not a physical object
- REFERENCE
- Reproduce the attached screen exactly. Invent no on-screen text.
- BRAND
- Direct, technical. Palette drawn from the site.
- LAYOUT
- Headline top third · product lower two thirds · CTA strip
- TEXT
- Headline rendered into the image
- AVOID
- invented UI, garbled lettering, a boxed edition of the app
- 05
Say better or worse
One tap under each render, against the last three — which are shown, so you are not recalling them. The formats reorder from your answers.
Against the last 3 for this productIs this closer or further?WorseBetterUGC Handheld · rank 100 → 104 from use
Most ad generators think you sell objects.
Ask one for an ad for your web app and it will invent something to photograph: a boxed edition, a branded device, a hardback book with your product name on a brass plaque. It looks competent. It is useless.
Without knowing the medium
The prompt says only “Product: modulate.pro”, so the model fills the gap with an object and puts your name on it. Nothing in the pipeline ever said the product was software.
Knowing the medium
The brief opens by stating what the product is, routes away from the object-compositing model, and formats built around packaging — unboxing, flat lay, macro detail — sort to the bottom with a badge.
The brief is right there.
Nobody else in this category will show you the prompt. It is where the quality actually lives, and it is the difference between a tool you can steer and a slot machine.
- SUBJECT
- modulate.pro — a web application, not a physical object
- REFERENCE
- Reproduce the attached screen exactly. Invent no on-screen text.
- BRAND
- Direct, technical. Palette drawn from the site.
- LAYOUT
- Headline top third · product lower two thirds · CTA strip
- TEXT
- Headline rendered into the image
- AVOID
- invented UI, garbled lettering, a boxed edition of the app
Copy the prompt
Take it anywhere. It is yours, not a trade secret.
Edit it and re-run
Send a prompt back verbatim, replacing the compiled brief rather than fighting it.
See which model ran
Named, with its per-image price — not hidden behind a “powered by” badge.
It gets closer, not just different.
A thumbs-up asks whether something is good, and you can run forty generations without one you would call good. That is forty renders and no signal.
UGC Handheld · rank 100 → 104 from use
One tap, no standard to meet
Better or worse than the last three — which are shown, so you are comparing rather than remembering.
Scoped so it means something
The comparison is against the same product, then the same format. “Better than the last three” is meaningless if those three were a different product entirely.
The catalog reorders
Formats you prefer rise and ones you skip fall, bounded so evidence adjusts the ranking rather than replacing it.
A runaway loop cannot produce a runaway bill.
Generation costs real money per image. Every other tool in this category handles that by not mentioning it. Plans are sold as credits, and a credit is a fixed slice of that cost — so the most a plan can ever cost to serve is one multiplication.
Reserved before the call
Spend is committed before the provider is contacted, so a burst of requests cannot all pass the same check.
Two ceilings, not one
A per-workspace cap you control, and a deployment ceiling set in the environment that nothing over HTTP can raise.
Prices are learned
Each model carries a configured, a vendor, and an observed price. The highest wins, so a vendor rise is absorbed rather than discovered on the invoice.
Built for people who ship software
Against what you would otherwise do
A designer at $60–120 a creative, a retainer, or a general-purpose image model that has never heard of your brand.
| AdForge | Freelancer | Agency | Generic AI | |
|---|---|---|---|---|
| Knows your product is software | ||||
| Cost visible before you commit | ||||
| Hard spend ceiling | ||||
| Same day | ||||
| Formats with a stated rationale | ||||
| You own the prompt | ||||
| Learns which formats you prefer |
No card. No key. The whole product.
Every provider falls back to a labelled placeholder, so you can walk the entire thing — brand kits, ideation, the Studio, spend tracking, the API — before deciding what to pay for. Preview generations cost nothing and never touch a cap.
Sign inQuestions, answered
Do I need design skills?
No. Pick a format and a product and generate. The formats carry their own art direction — lighting, lens, styling — so the decisions a designer would make are already encoded, and each one records where it came from and when it was last confirmed to still be running.
What if my product is a web app with no packaging?
That is the case this was built for. Tell it what the product is and it renders your actual interface on a device rather than inventing an object. Upload a screenshot and it is instructed to reproduce that screen exactly.
Which models does it use?
Flux, Ideogram, Bria, and both Gemini image models, each named with its per-image price. Routing is automatic and derives from the format, and you can override it per generation.
How do I know what it will cost?
The exact figure is on the Generate button before you press it. Spend is capped per workspace and per deployment, and every model carries a conservative price so an estimate is never lower than the bill.
How does the pricing work?
Plans are credits, at the same rate on every plan — no tier buys them cheaper. A render on the fast models is two credits; sharper models and 4K draw more, in exact proportion to what they cost to run. Every tier reaches every model, because the credit does the rationing instead of the plan.
What does “better or worse” actually do?
It records a comparison against the three renders shown beside it, and a preference score is fitted across all of them. Formats you prefer rise in the Studio and ones you skip fall. It is not a rating — you never have to decide whether something is good, only whether it is closer.
Can I use my own prompts?
Yes. A raw prompt replaces the compiled brief rather than being appended to it, so you can tell exactly which instruction produced the result.
Is my brand data used for training?
No. It is stored in your workspace and sent to the model providers only as part of the generations you run.