Setting Up a Monorepo That Doesn't Make You Want to Quit

I've seen the pattern in almost every codebase I've inherited: users upload images, the app stores them raw, and the frontend serves them as-is. Someone uploads a 8MB DSLR photo as their profile picture. It gets resized to 40x40 pixels in CSS. The browser still downloads all 8MB.

It takes maybe a day to build a proper pipeline. Here's exactly how I do it.

The Goals

1. Accept uploads in any common format
2. Validate on the client before wasting upload bandwidth
3. Process server-side: resize, convert to WebP/AVIF, strip metadata
4. Store originals and processed variants separately
5. Serve via CDN with proper caching headers
6. Generate srcset-compatible URLs for responsive images

Client-Side Validation First

Don't wait for a failed upload to tell the user their 4K raw file is too big:

typescript
// hooks/useImageUpload.ts
interface UploadOptions {
  maxSizeMb?: number
  acceptedFormats?: string[]
  maxDimension?: number
}

const DEFAULT_OPTIONS: Required<UploadOptions> = {
  maxSizeMb: 10,
  acceptedFormats: ['image/jpeg', 'image/png', 'image/webp', 'image/gif'],
  maxDimension: 8000,
}

export async function validateImageFile(
  file: File,
  options: UploadOptions = {}
): Promise<{ valid: true } | { valid: false; reason: string }> {
  const opts = { ...DEFAULT_OPTIONS, ...options }

  if (!opts.acceptedFormats.includes(file.type)) {
    return {
      valid: false,
      reason: `Format not supported. Please upload a JPG, PNG, or WebP image.`,
    }
  }

  const sizeMb = file.size / (1024 * 1024)
  if (sizeMb > opts.maxSizeMb) {
    return {
      valid: false,
      reason: `File too large (${sizeMb.toFixed(1)}MB). Maximum size is ${opts.maxSizeMb}MB.`,
    }
  }

  // Check actual image dimensions
  const dimensions = await getImageDimensions(file)
  if (dimensions.width > opts.maxDimension || dimensions.height > opts.maxDimension) {
    return {
      valid: false,
      reason: `Image too large (${dimensions.width}×${dimensions.height}px). Maximum is ${opts.maxDimension}px.`,
    }
  }

  return { valid: true }
}

function getImageDimensions(file: File): Promise<{ width: number; height: number }> {
  return new Promise((resolve, reject) => {
    const url = URL.createObjectURL(file)
    const img = new Image()

    img.onload = () => {
      resolve({ width: img.naturalWidth, height: img.naturalHeight })
      URL.revokeObjectURL(url)
    }

    img.onerror = () => {
      reject(new Error('Could not read image dimensions'))
      URL.revokeObjectURL(url)
    }

    img.src = url
  })
}

And a preview before upload:

typescript
export function useImagePreview(file: File | null) {
  const [preview, setPreview] = useState<string | null>(null)

  useEffect(() => {
    if (!file) {
      setPreview(null)
      return
    }

    const url = URL.createObjectURL(file)
    setPreview(url)

    return () => URL.revokeObjectURL(url)
  }, [file])

  return preview
}

Server-Side Processing with Sharp

sharp is the standard Node.js image processing library. It's fast, handles all formats, and runs native binaries:

bash
npm install sharp
npm install -D @types/sharp

typescript
// lib/image-processor.ts
import sharp from 'sharp'

export interface ImageVariant {
  name: string
  width: number
  height?: number  // If not provided, aspect ratio is preserved
  format: 'webp' | 'avif' | 'jpeg'
  quality: number
}

export const AVATAR_VARIANTS: ImageVariant[] = [
  { name: 'sm', width: 40, format: 'webp', quality: 80 },
  { name: 'md', width: 80, format: 'webp', quality: 80 },
  { name: 'lg', width: 160, format: 'webp', quality: 85 },
  { name: 'sm', width: 40, format: 'jpeg', quality: 85 }, // JPEG fallback
]

export const POST_IMAGE_VARIANTS: ImageVariant[] = [
  { name: 'thumbnail', width: 400, format: 'webp', quality: 80 },
  { name: 'card', width: 800, format: 'webp', quality: 82 },
  { name: 'hero', width: 1200, format: 'webp', quality: 85 },
  { name: 'hero-avif', width: 1200, format: 'avif', quality: 75 },
  // JPEG fallbacks
  { name: 'thumbnail', width: 400, format: 'jpeg', quality: 85 },
  { name: 'card', width: 800, format: 'jpeg', quality: 85 },
  { name: 'hero', width: 1200, format: 'jpeg', quality: 88 },
]

export interface ProcessedVariant {
  variant: ImageVariant
  buffer: Buffer
  size: number
}

export async function processImage(
  input: Buffer,
  variants: ImageVariant[]
): Promise<ProcessedVariant[]> {
  // Strip EXIF metadata (GPS data, camera info, etc.)
  const base = sharp(input).rotate() // rotate() applies EXIF orientation then strips it

  const results = await Promise.all(
    variants.map(async (variant) => {
      let pipeline = base.clone().resize({
        width: variant.width,
        height: variant.height,
        fit: 'cover',
        position: 'entropy', // crops to the most interesting part
        withoutEnlargement: true, // don't upscale small images
      })

      if (variant.format === 'webp') {
        pipeline = pipeline.webp({ quality: variant.quality })
      } else if (variant.format === 'avif') {
        pipeline = pipeline.avif({ quality: variant.quality })
      } else {
        pipeline = pipeline.jpeg({ quality: variant.quality, mozjpeg: true })
      }

      const buffer = await pipeline.toBuffer()

      return {
        variant,
        buffer,
        size: buffer.length,
      }
    })
  )

  return results
}

Storage with S3-Compatible Object Storage

typescript
// lib/storage.ts
import { S3Client, PutObjectCommand, DeleteObjectCommand } from '@aws-sdk/client-s3'
import { randomUUID } from 'crypto'

const s3 = new S3Client({
  region: process.env.AWS_REGION!,
  credentials: {
    accessKeyId: process.env.AWS_ACCESS_KEY_ID!,
    secretAccessKey: process.env.AWS_SECRET_ACCESS_KEY!,
  },
})

const BUCKET = process.env.S3_BUCKET!
const CDN_URL = process.env.CDN_URL! // e.g., https://cdn.yourapp.com

export interface StoredFile {
  key: string
  url: string
  size: number
}

export async function uploadFile(
  buffer: Buffer,
  options: {
    folder: string
    filename?: string
    contentType: string
    cacheControl?: string
  }
): Promise<StoredFile> {
  const filename = options.filename ?? randomUUID()
  const key = `${options.folder}/${filename}`

  await s3.send(
    new PutObjectCommand({
      Bucket: BUCKET,
      Key: key,
      Body: buffer,
      ContentType: options.contentType,
      CacheControl: options.cacheControl ?? 'public, max-age=31536000, immutable',
      // Immutable because filenames include content hash
    })
  )

  return {
    key,
    url: `${CDN_URL}/${key}`,
    size: buffer.length,
  }
}

The Upload API Route

typescript
// routes/uploads.ts
import multer from 'multer'
import { processImage, AVATAR_VARIANTS } from '../lib/image-processor'
import { uploadFile } from '../lib/storage'
import { createHash } from 'crypto'

// Store uploads in memory — we process and discard immediately
const upload = multer({
  storage: multer.memoryStorage(),
  limits: { fileSize: 10 * 1024 * 1024 }, // 10MB
})

app.post(
  '/api/upload/avatar',
  requireAuth,
  upload.single('image'),
  async (req, res) => {
    if (!req.file) {
      return res.status(400).json({ error: 'No file provided' })
    }

    // Content-hash the filename so cache busting works automatically
    const hash = createHash('md5').update(req.file.buffer).digest('hex').slice(0, 8)

    const variants = await processImage(req.file.buffer, AVATAR_VARIANTS)

    // Upload all variants in parallel
    const uploaded = await Promise.all(
      variants.map(({ variant, buffer }) =>
        uploadFile(buffer, {
          folder: `avatars/${req.user.id}`,
          filename: `${variant.name}-${hash}.${variant.format}`,
          contentType: `image/${variant.format}`,
        })
      )
    )

    // Build the response: variant name → URL mapping
    const urls = Object.fromEntries(
      variants.map((v, i) => [
        `${v.variant.name}_${v.variant.format}`,
        uploaded[i].url,
      ])
    )

    // Save to DB
    await db.user.update({
      where: { id: req.user.id },
      data: { avatarUrls: urls },
    })

    res.json({ urls })
  }
)

Serving Responsive Images

With variants stored, your <img> tags can use srcset for responsive delivery:

typescript
// components/Avatar.tsx
interface AvatarProps {
  urls: Record<string, string>
  name: string
  size?: 'sm' | 'md' | 'lg'
}

export function Avatar({ urls, name, size = 'md' }: AvatarProps) {
  const pxMap = { sm: 40, md: 80, lg: 160 }
  const px = pxMap[size]

  // Prefer AVIF > WebP > JPEG
  return (
    <picture>
      {urls.sm_webp && (
        <source
          type="image/webp"
          srcSet={`${urls.sm_webp} 40w, ${urls.md_webp} 80w, ${urls.lg_webp} 160w`}
          sizes={`${px}px`}
        />
      )}
      <img
        src={urls[`${size}_jpeg`] ?? urls.md_jpeg}
        width={px}
        height={px}
        alt={`${name}'s avatar`}
        loading="lazy"
      />
    </picture>
  )
}

For post hero images with AVIF:

tsx
<picture>
  <source
    type="image/avif"
    srcSet={`${urls.thumbnail_avif} 400w, ${urls.card_avif} 800w, ${urls.hero_avif} 1200w`}
    sizes="(max-width: 640px) 400px, (max-width: 1024px) 800px, 1200px"
  />
  <source
    type="image/webp"
    srcSet={`${urls.thumbnail_webp} 400w, ${urls.card_webp} 800w, ${urls.hero_webp} 1200w`}
    sizes="(max-width: 640px) 400px, (max-width: 1024px) 800px, 1200px"
  />
  <img
    src={urls.hero_jpeg}
    alt={post.imageAlt}
    width={1200}
    height={630}
    loading="eager" // Hero image load eagerly
    fetchPriority="high"
  />
</picture>

AVIF is significantly smaller than WebP for the same quality, but browser support is slightly narrower (still 90%+ globally). The <picture> element handles the fallback gracefully.

What This Gets You

A raw 3MB JPEG profile photo becomes three WebP variants totaling roughly 60KB and three JPEG fallbacks totaling about 120KB. The browser downloads the appropriate size and format on a modern browser viewing a large avatar, that's a 30KB WebP file instead of 3MB of original data.

For post images, AVIF hero images are typically 40–60% smaller than the equivalent JPEG. At scale, that's meaningful bandwidth and meaningful load time.

The pipeline is a day of work. The payoff is permanent.