Dresma powers 50K+ brands with AI visual content — here's how it could run end-to-end on Cloudflare's developer platform

What if Dresma's AI content engine ran on Cloudflare?

Dresma generates AI lifestyle shots, on-model imagery, videos, infographics, and ad creatives for 50K+ ecommerce brands — Fabletics, Luxottica, Champion, Puma, Nautica, Palm Angels, Off-White, Sperry, Zenni, World Rugby Shop. Every workflow is compute-heavy, storage-heavy, and latency-sensitive. Here's what the same engine looks like on Cloudflare's developer platform.

Workers AI AI Gateway R2 Vectorize Workers Images Stream
dresma-stack-comparison.sh
# Current: AWS + Webflow + Google Cloud
Marketing site: Webflow (cdn.webflow.com)
SaaS app: Google App Engine (app.dresma.com)
Backend API: AWS EC2 (us-east-2)
Asset CDN: CloudFront (2× distributions)
Image storage: S3 + egress fees
Mobile backend: Firebase
# Proposed: Cloudflare developer platform
Marketing site: Pages (edge-served)
SaaS app: Pages + Workers (330+ cities)
Backend API: Workers (~5ms cold)
Asset CDN: R2 + Images ($0 egress)
Image storage: R2 (S3-API)
AI orchestration: AI Gateway + Workers AI
$ echo "One platform, edge-native, no egress."
From multi-cloud sprawl → one developer platform

Dresma powers visual content for 50K+ brands

Fabletics Luxottica Champion Puma Nautica Palm Angels Off-White Sperry Zenni World Rugby Shop
Track → Curate → Create → Human QC

Dresma's four-step workflow, mapped to Cloudflare primitives

Each step of the workflow has a clean Cloudflare equivalent. The result is a developer platform that's not just cheaper — it's structurally better-suited to AI content at scale.

Step 1

Track

Scan top-performing visuals across Amazon, Instagram, Pinterest, competitor sites. Index for similarity search.

Workers + Vectorize
Workers crawl + transform; Vectorize indexes embeddings of millions of visuals. Native at-edge inference for embedding generation.
Step 2

Curate

Build a data-backed mood board personalized to the brand. Surface high-performing references from the index.

Vectorize + D1 + Workers AI
Vector similarity search at the edge. D1 stores brand profile + history. Workers AI ranks and re-ranks results in <50ms.
Step 3 — biggest play

Create

Generate AI lifestyle shots, on-model images, videos, infographics, ad creatives. Multi-provider: Stable Diffusion, OpenAI, Anthropic, custom.

AI Gateway + Workers AI + R2
One observability + caching + rate-limit layer across providers. R2 stores generated assets with $0 egress. Up to 40% cost reduction through cached identical prompts.
Step 4

Human QC

Review against brand guidelines. Adjust if needed. Ship to delivery channels (Amazon, Shopify, marketplaces).

Pages + Workers + Queues
QC UI on Pages, edge-served globally. Workers run brand-rule validators. Queues handle delivery fan-out to Amazon, Shopify, Walmart, Kaufland, Mercado Livre.

Why Cloudflare's developer platform fits Dresma specifically

Four structural properties of the platform that map directly to how Dresma operates today.

AI Gateway — multi-model orchestration that pays for itself

Dresma runs multiple AI providers behind the scenes — Stable Diffusion for backgrounds, OpenAI/Anthropic for copy and reasoning, custom models for brand-specific outputs. AI Gateway sits in front of all of them: one observability + caching + rate-limit layer. Identical or near-identical prompts (common in batch product photography) are cached, cutting inference cost by 30-50% with zero code changes.

Inference cost reduction without changing models

R2 — zero-egress storage for visual assets

Every Dresma workflow produces images and videos. Today these live on S3 + CloudFront with egress fees on every brand customer download, every Amazon A+ ingest, every Shopify CDN pull-through. R2 is S3-compatible — boto3 and aws-sdk code paths work unchanged. Zero egress. Multiply by 50K brands and tens of millions of assets/year, and the math is meaningful (see calculator below).

Drop-in for boto3 / aws-sdk

Vectorize — at-edge similarity search for mood boards

Dresma's "data-backed mood board" feature is a vector similarity problem at heart: find the top-performing visuals similar to this brand's style, across Amazon / Instagram / competitors. Vectorize stores embeddings at the edge, queries return in <50ms globally. Workers AI generates embeddings in the same control plane — no separate inference vendor needed.

<50ms semantic search at the edge

Images + Stream — output-channel optimization

Every Dresma output needs to be resized, format-converted, and channel-optimized for Amazon, Shopify, Walmart, Kaufland, Mercado Livre. Cloudflare Images handles per-request transformation on the fly (no pre-rendering, no S3 variants); Stream handles AI-generated video delivery globally. Replaces a custom image-processing pipeline with one managed service.

One service, every channel
Interactive Visual Asset Storage Calculator

AWS S3 + CloudFront vs. Cloudflare R2 for Dresma's asset library

Pre-populated with Dresma-scale defaults (50K brands × generated images per brand). Adjust to match your real numbers.

Your inputs

Monthly figures, applied as steady-state.

AI-generated images, videos, lifestyle shots, on-model imagery, ad creatives across 50K brands.

Brand customers pulling assets, Amazon/Shopify/Walmart ingest, CDN distribution, mobile app downloads.

PUT / COPY / POST — every AI-generated image, video frame, infographic upload.

GET — every asset retrieval, CDN cache miss, A+ content fetch, app preview load.

AWS S3 + CloudFront
Current state
Storage $4,600
Egress $7,200
Write ops $250
Read ops $120
Monthly
$12,170
≈ $146,040/year
Cloudflare R2
Proposed
Storage $3,000
Egress (zero) $0
Class A ops $225
Class B ops $108
Monthly
$3,333
≈ $39,996/year
Estimated Annual Savings
$106,044
73% lower TCO vs. AWS S3 + CloudFront — compounds as brand-customer reads scale

Calculation based on AWS S3 + CloudFront public list pricing vs. Cloudflare R2 public list pricing as of 2025. Excludes additional value from AI Gateway inference caching, edge-served Workers, and Vectorize replacing a self-hosted vector DB.

Real-world savings typically exceed the calculated figure because CloudFront egress isn't a flat rate — origin-shield pull-through, regional egress, and per-distribution charges add layers. R2 charges zero across all of them.

Architecture: Dresma on Cloudflare end-to-end

Every layer of Dresma's current stack maps to a Cloudflare primitive. Multi-cloud sprawl → one developer platform.

Marketing site
Cloudflare Pages

Replaces Webflow + cdn.webflow.com. Git-driven deploys, branch previews, edge-served globally with no origin server.

Replaces: Webflow
SaaS app + API
Workers + Pages

Replaces app.dresma.com (Google App Engine) + api.dresma.com (AWS EC2 us-east-2). 5ms cold starts. Edge-served in 330+ cities.

Replaces: GAE + AWS EC2
Visual asset storage
R2 + Images + Stream

Replaces S3 + 2× CloudFront distributions. Zero egress, S3-compatible API, on-the-fly image transforms, video delivery.

Replaces: S3 + CloudFront ×2
AI orchestration
AI Gateway + Workers AI

One layer over every model provider (image gen, copy gen, infographic gen). Caches identical prompts for 30-50% inference cost cut.

New capability — net-new
Vector / mood board search
Vectorize

Embedding store for top-performing visuals indexed across Amazon, Instagram, competitors. <50ms similarity queries at the edge.

Replaces: self-hosted vector DB
Brand profile DB
D1 + Hyperdrive

D1 for brand profiles, workflow history, QC state. Hyperdrive accelerates any AWS RDS / Postgres queries from edge Workers.

Augments: existing DB layer
Workflow orchestration
Queues + Workers

Track → Curate → Create → QC pipeline. Queues handle fan-out: each created asset gets resized, distributed to Amazon, Shopify, Walmart, Kaufland, Mercado Livre.

Replaces: custom orchestrator
Bot Mgmt + WAF
App Security

Protect the AI generation API from prompt-abuse scraping (free-tier exploitation, brand-IP theft) and protect the customer-facing dashboard.

Augments: existing infra
Mobile backend
Workers + D1

Dresma's mobile app (formerly DoMyShoot) runs on Firebase. Workers can replace mobile API + auth; D1 for offline-sync state.

Optional: replace Firebase

The Developer Platform Business Case for Dresma

Three measurable outcomes — cost, capability, and consolidation.

70%+
Asset Storage TCO Reduction

R2 vs. S3 + CloudFront, calculated on Dresma-scale storage + egress + ops. Compounds as brand-customer reads scale.

30-50%
Inference Cost Cut via Caching

AI Gateway caches identical or near-identical prompts. In batch product photography workflows, prompts repeat constantly — caching is structural savings.

5 → 1
Vendor Consolidation

Webflow + Google App Engine + AWS EC2 + CloudFront + Firebase → one developer platform. One bill, one dashboard, one identity layer.

Ready to walk through this for Dresma?

Let's map Dresma's actual workflow + cost model and build a concrete migration path. Workshop format, 60-90 min, with your engineering team.

About this page

Prepared as a Cloudflare Developer Platform conversation primer for Dresma. Built and maintained by Matt Holscher, Digital Native Sales Team, Cloudflare. mattholscher@cloudflare.com · +1 (415) 699-7053