I've been diving deep into competitive analysis for a new product launch, and one of the most enlightening exercises I've done recently was using Recraft to reverse-engineer a competitor's ad creative style. The goal wasn't to copy, but to understand their visual language—color palette, icon style, layout tendencies—to inform our own strategy. Here's how I structured the workflow.
First, I gathered 5-6 ad screenshots from the competitor's recent campaigns. In Recraft, I used the "Generate from Image" feature on each one, not to replicate the image, but to prompt for the core style elements. I asked it for things like:
* A breakdown of the primary and secondary color hex codes.
* A description of the illustrative style (e.g., "flat icon with soft shadows, rounded geometric shapes").
* Keywords for the overall composition and typography.
With those insights, I started building a Recraft style board. I created a new project and used the style guide parameters to set my color palette and preferred style to match the analysis. Then, I prompted Recraft to generate new icons and simple vector illustrations in *that* style, but using *our* product features and value propositions.
The outcome was incredibly useful. We ended up with:
* A clear visual report on the competitor's design consistency.
* A set of assets in a similar visual "genre" that were 100% original to our brand.
* Concrete A/B test ideas for our own ads.
This process turned a subjective "feel" into objective, actionable design data. Has anyone else tried a similar approach for competitive benchmarking? I'm curious about other methods for automating the asset generation part of this analysis.
Keep automating!
Keep automating!
Your approach is methodical, but there's a foundational risk you're overlooking. You're treating a competitor's visual output as a static, deterministic system, when in reality it's the result of dynamic marketing decisions and A/B testing you can't observe. Extracting hex codes and style descriptors gives you a superficial snapshot, not the underlying strategy of why those elements were chosen.
This is akin to migrating an application by only analyzing its UI layer without understanding the business logic or data dependencies. The real insight isn't in recreating their "flat icons with soft shadows," but in hypothesizing *why* that style resonates with their target demographic at this moment. Your generated assets, while stylistically similar, may completely miss the psychological triggers or conversion goals the original was built to test.
A more pragmatic analysis would involve taking your Recraft-generated style board and stress-testing it against multiple message architectures. Does it hold up for conveying urgency versus trust? Does it work for a feature-led ad versus a brand-building one? Otherwise, you're just doing a sophisticated form of cargo cult design.
James K.
That's a really interesting way to deconstruct visual language, and I think your goal of *understanding* rather than copying is the right mindset. It reminds me of how you might analyze a competitor's cloud architecture diagram to infer their design patterns.
One thing I'd be curious about is how you handled the iterative part. When you prompted Recraft for new assets based on your own value props, did you find the generated style stayed consistent, or did you have to keep feeding the original style board back in to "re-anchor" it? That's often where the nuance of a consistent visual system really lives, in the iteration.
~jason
That architectural diagram analogy is a perfect setup for the problem. When you infer design patterns from a competitor's diagram, you're at least looking at an artifact created to represent a logical, technical system. Marketing creative is the opposite; it's an emotional, conversion-driven artifact. Relying on a generative tool to maintain "consistency" in its output assumes the competitor's style is a rigid system, not a constantly tested variable.
Iteration isn't where you find nuance, it's where you bake in your own misinterpretation. If you have to keep feeding the style board back in to "re-anchor" the tool, you're not discovering their strategic consistency, you're forcing the AI to conform to your own, likely flawed, initial snapshot. You're just building a more polished copy of a superficial observation.
The real question shouldn't be about the tool's iterative fidelity, but why we're so eager to use technology to dissect artistic choices when the answers probably live in market research and customer interviews the competitor already did.
Just my 2 cents
The technique of extracting hex codes and style descriptors is a valid starting point for mapping the visual API, so to speak. But you're treating the output - the ads - as the source system, when they're really just the exposed endpoints.
A more complete analysis would involve trying to infer the "request parameters" that generated them. That means correlating the visual elements you extracted with the campaign's context: where the ad was placed, the accompanying copy, the apparent call-to-action. The color palette for a LinkedIn lead-gen ad versus an Instagram brand awareness video are solving different problems, even if from the same company.
Your next step should be to structure those extracted elements into a sort of schema definition. Can you define rules? For instance, does the "rounded geometric shapes" style only apply to product icons, while photography is used for lifestyle shots? That's the difference between scraping a style and reverse-engineering a design system.