UXPin Merge AI 2.0 - AI that sticks to your design rules
by•
Smarter AI for real-code design. Create and refine layouts faster with components from open-source or custom libraries that follow your design system—so every idea moves closer to production.
Replies
If I’m skeptical of AI UI tools, what’s the “one feature demo” that usually changes minds?
@erin_james2 our 'ah-ha' moment is definitely when a prospect see's AI building a prototype or project using their components.
Seeing the AI stay inside the component system (instead of inventing UI) is the moment it clicks for most folks.
Love the direction. Are you optimizing more for speed of iteration or correctness of system adherenc
@brendon_c_harwell great question Brendon! At the moment, I'd say we're trying to optimise for both - but that's not a real answer haha.
Short-term: speed of iteration with guardrails.
We want you to move fast, but still stay anchored to real components so you’re not generating pretty-but-useless mockups.
Long-term: correctness of system adherence is the real win.
That’s especially true for the custom library beta — the goal is AI that reliably operates inside your tokens/props/patterns so outputs feel “system-native,” not improvised.
In practice, we’re aiming for a workflow where you can explore quickly and trust that refinement won’t quietly drift off-system.
Out of curiosity, which failure would annoy you more?
Fast but occasionally off-system, or slower but extremely strict?
How do you stop the AI from “helpfully” breaking spacing/token rules during refinement?
@bobby_l_benefield Great question. That “helpful drift” is the thing that kills trust fast.
How we’re approaching it in Merge 2.0:
Real component grounding first.
When refinement is operating on code-backed components (not freeform boxes), the AI is already constrained to more “legal” structure.
Library-aware behavior.
With MUI/shadcn/AntD etc., we’re nudging the AI to prefer existing spacing/size patterns those systems already encode, rather than inventing new values.
Refinement > rewrite.
We bias the assistant toward adjusting hierarchy/composition instead of re-authoring styles from scratch, which is where token violations usually sneak in.
Enterprise beta takes this further.
The Custom Library AI track is meant to learn your actual tokens/props conventions via Git so the AI is less likely to “make up” spacing rules.
We’re also very aware this is an area that needs continuous monitoring and tightening, we want “stays on-system by default” to feel boringly reliable.
What’s the biggest improvement from 1.0 to 2.0?
@anaya_backlink
Short answer: the jump from one-shot generation to real, iterative refinement.
V1 was great for “prompt → first draft layout.”
2.0 is about “create + refine” with a proper AI helper, stronger models, and image upload — so you can start from a reference or a real layout and shape it into something shippable instead of restarting from scratch.
If you try one thing to feel the difference, I’d suggest:
generate a layout → then run 1–2 refinement prompts to tighten hierarchy/density.
Curious if you’re seeing better results with certain UI categories — dashboards vs marketing vs admin?
@nomi_butt1 Yes - anecdotally we’re seeing stronger, more “trustworthy” results in structured, component-heavy UIs:
Best so far:
Dashboards (cards, tables, filters, charts, layouts)
Admin/settings (forms, sidebars, multi-panel layouts)
These tend to map cleanly to MUI/shadcn/AntD primitives and benefit a lot from the refinement flow.
A bit more variable:
Marketing pages
They still work well, but the results depend more on your prompt constraints around hierarchy, tone, and component choices.
If you want to stress-test 2.0, dashboards + settings flows are probably the fastest way to see the “real components + refinement” advantage.
What category are you hoping to use this for most?
For custom libraries, how much setup does the Git connection require on the dev side?
@kristina_stankovic Hey Kristina - thanks for the question.
In complete honesty, it's a fair bit of work at this stage. We are working on an automated integration piece which will be a part of our wider rollout of the Enterprise AI offering.
In the interim, for our Enterprise customers, our team completes a boilerplate to ensure your components work nicely within the system and then we assist in the wider integration.
Looks cool. How do you see this fitting into the whole vibe coding realm?
@jozacks Hey Jonathan,
I think in a way, it's almost vibe-design but when integrated with global or custom libraries it has its own inherent rulesets to keep it consistent.
At the moment I love doing early exploration and idea generation before manually refining!
UXPin Merge
Really proud of what the team achieved with Merge AI 2.0. This release is a big step toward making AI genuinely useful in real product workflows—not just for quick mockups, but for creating and refining code that teams can actually ship.
The new refinement flow, image-based generation, and custom library support all came from the same goal: help designers and engineers work with AI without losing their design system or their craft. It’s exciting to finally see those ideas come together.
@uxpinjack let's gooo 🔥
This appears to have significant potential for legacy UI cleanup. Is that a scenario you’ve already evaluated?
UXPin Merge
@nicholas_davidson2 Thanks, Nic! Yes, this is something we’ve looked at. You can upload a screenshot of an older interface or work directly with an existing layout in UXPin, and the step-by-step refinements work surprisingly well for that kind of update.
Any plans for team-level presets so prompts follow company conventions?
UXPin Merge
@teri_morgan Yes! It's on our radar. Shared prompts would be great for keeping teams aligned.