The "Inside Job" Relationship

Let’s get the elephant out of the room: the entire GTM team at Pruna (all two of us) are former Strapiers. Quentin left a few years back, and I literally handed in my laptop a month ago.

The Pruna Team first noticed Strapi was getting serious about AI when features started appearing directly where content teams actually work, not as a side experiment, but as core workflow tools. The AI Media Library, announced at StrapiConf 2025, was the early signal. Seeing a CMS natively handle things like auto-generating alt text and captions on upload showed them that something much more ambitious was taking shape.

When Quentin saw those signals, he didn’t send a cold LinkedIn message. He just called the family, AKA me, for a sanity check, then reached out to the Strapi founders. But being "family" only gets you the meeting; it doesn’t close the deal.

The conversation quickly turned into a deeper one about Fimo.ai, a new product Strapi was preparing to launch. Not another vibe-coding website builder, but an AI-native CMS where content, media, and structure are created, edited, and iterated continuously with AI in the loop.

The timing mattered. Fimo was still being shaped, which meant foundational decisions, including how image generation would work, were not locked yet. This was not about swapping models later. It was about choosing the right base before real users showed up.

The Problem: "Vibe-Coding" vs. Real Work

In a real AI-native CMS, AI can't just be a sidecar or a "magic" button that generates a random, polished-but-soulless stock photo. If you’re building a real brand, those generic, over-saturated AI images are a dealbreaker. They break the "context" of the site.

That might be acceptable for a mockup. Inside a CMS, it breaks the experience.

Once you add real usage on top of that, the problems compound. Latency spikes. Output inconsistency. Pricing that works for a prototype but not at scale. What looked fine in a demo quickly becomes a blocker in production.

In a CMS, “good-looking” is not enough. Images need to belong.

When the Fimo team started building, they hit the wall that every AI founder hits:

• The "Speed vs. Quality" Trap: You can have fast (FLUX Schnell), but the quality is often "meh."

• The "Luxury" Trap: You can have great quality (Nano Banana), but it’s slow, expensive, and a nightmare to scale when you have thousands of users hitting the API.

• The "Math" Trap: Some models (like Qwen) are interesting, but the pricing complexity makes you want to hire a full-time accountant just to figure out your margins.

Cutting through the BS

We skipped the slide decks. In the startup world, if you can’t show it in a playground, it doesn’t exist.

The Fimo team jumped into our Playground and played with P-Image and P-Image-Edit hands-on. Generated images. Edited them. Pushed prompts. Tried to break things. The speed stood out immediately. Then the consistency. Then the fact that generation and editing behaved the same way across use cases.

Only after that did they move to testing through our API.

Around the same time, Strapi was also using fal.ai. It worked, but currency and operational constraints eventually pushed them to move forward directly with our API. Hope there are no hard feelings, folks. And by the way, we can deploy our models on your platform anytime 😉

Why this actually matters

At the end of the day, Strapi chose Pruna because we both care more about shipping than debating AI theory.

Fimo.ai is launching with image generation and editing that feels like a first-class citizen. It’s not an experiment hidden behind a toggle; it’s a foundational part of how the CMS works. We helped them avoid the "technical debt" of picking a slow/expensive model early on, which usually leads to a painful rewrite six months later.

What’s next

Fimo.ai is just getting started. As they add more agents and optimization layers, the foundation we built together is going to get a real workout. We’re not just watching from the sidelines; we’re in the trenches with them.

Want to see what an AI-native CMS actually looks like? Upvote and check out Fimo.ai on Product Hunt here!

And if you’re tired of "demo-only" AI and need your features to actually scale without costing a fortune, we should probably chat!