title: How I Optimized My AI Image App from 3s to 300ms with Next.js & Supabase

Richard — Thu, 08 Jan 2026 08:02:27 +0000

Hello Developers! 👋

I recently launched Nanobanan Editor, an AI-powered image editing tool focusing on natural language prompts.

While building the MVP was fun, I hit a massive roadblock: Performance. Specifically, my "Community Feed" page was taking 3-5 seconds to load. For a user-facing gallery, that's unacceptable.

In this post, I want to share how I diagnosed the bottleneck and optimized it to under 300ms (a 10x improvement) using client-side rendering strategies and database indexing.

The Stack 🛠️

Framework: Next.js 14 (App Router)
Database: Supabase (PostgreSQL)
Styling: TailwindCSS
Deployment: Vercel

The Problem: Traditional SSR Bloat 🐢

Initially, I implemented the community page using standard Server-Side Rendering (SSR) in Next.js.

// ❌ The slow way (Simplified)
export const dynamic = "force-dynamic"

export default async function CommunityPage() {
  // Blocking current thread to fetch database
  const images = await supabase.from('generations').select('*')...

  return <Gallery images={images} />
}

Why it was slow:**

Blocking: The HTML wouldn't stream until the database query finished.

Complex Query: I was querying a large dataset without proper indexes.

Network: The server-to-database round trip added latency for every single request.

The Solution: CSR + SWR + Indexes 🚀

I decided to pivot from SSR to Client-Side Rendering (CSR) with a "Stale-While-Revalidate" strategy.

Step 1: Switch to CSR with Skeleton Loading

Key change: Show the UI immediately (skeletons), then fetch data.

"use client"
import useSWR from 'swr'

export default function CommunityPage() {
    // Non-blocking fetch
    const { data, isLoading } = useSWR('/api/community/feed', fetcher)

    if (isLoading) return <SkeletonGrid /> // Instant feedback

    return <Gallery images={data.images} />
}

Step 2: Intelligent Caching with SWR

I used swr to handle caching. If a user visits the community page, leaves, and comes back 10 seconds later, it loads instantly from the cache without hitting the API.

const { data } = useSWR('/api/community/feed', fetcher, {
    dedupingInterval: 60000, // Reuse data for 60s
    revalidateOnFocus: false // Don't re-fetch just because I clicked a tab
})

Step 3: Database Indexing (The Real MVP)

This was the biggest win. I analyzed my SQL query:
SELECT * FROM generations WHERE is_public = true ORDER BY created_at DESC LIMIT 12

I realized I was doing a sequential scan. I added a composite index in Supabase:

CREATE INDEX idx_generations_public_created 
ON generations (is_public, created_at DESC) 
WHERE is_public = true;

Result*: The database query time dropped from *~500ms** to ~15ms.

The Results 📊

First Load: ~300ms (Skeleton UI visible instantly)
Repeat Visit: < 50ms (Cache hit)
Lighthouse Score: Jumped from 65 to 95.

Try it out

You can experience the speed difference live here:
👉 **[Nanobanan Editor Community]

(https://www.nanobanan.tech/)**

This journey taught me that while SSR is powerful, sometimes good old Client-Side Rendering with a smart caching strategy provides a snappier UX for feed-based pages.

Let me know what you think of the app! Also, happy to answer any questions about the Next.js + Supabase stack in the comments. 👇

webdev #javascript #programming #showdev

Forem: Richard

title: How I Optimized My AI Image App from 3s to 300ms with Next.js & Supabase