MyPhotoAI · pSEO case study

Pollinations or Replicate: which image API works for pSEO at scale

Per-slug AI imagery is one of the cheapest pSEO investments you can make, but the provider choice matters. We ship on Pollinations by default and have evaluated Replicate as the paid alternative. This leaf is the side-by-side comparison across five dimensions, with the leaf-count threshold where the calculation flips from free to paid.

By AlkenaCode engineering team

We ship programmatic SEO surfaces, AI-driven receptionists, and case-study sites for Kenyan SMEs and overseas clients. Every claim on this page traces back to shipping work, not speculation. The audit tool on /audit is the same software we use to evaluate prospective engagements.

Authors of the myphotoai.alkenacode.dev pSEO surface itself. Source code is open at github.com/Kiragu-Maina/alkenacode-family. Production deployments include agents.alkenacode.dev and home.alkenacode.dev.

Last updated:

We have shipped only Pollinations to production across both surfaces. The Replicate numbers in this leaf come from an evaluation run we did during the planning phase of this surface; the per-image price points are taken directly from Replicate's public pricing.

Pollinations or Replicate: which image API works for pSEO at scale

Cost and rate limits

Pollinations is free at the flux model with a soft rate limit of roughly one request per fifteen seconds per IP. Replicate prices per generation; the flux-schnell model is about $0.0035 per image, the flux-dev model is about $0.025. For our 22-leaf surface, Pollinations cost zero dollars and took six minutes; Replicate would have cost about $0.08 at schnell or $0.55 at dev, with no rate limit and a build time of under one minute. The crossover happens around five hundred leaves where Pollinations' rate limit imposes a two-hour wall-clock that may exceed a deploy window.

Quality and aesthetic match

Pollinations flux produces consistent output that suits a technical-diagram aesthetic well. Replicate flux-dev produces sharper output with more reliable composition fidelity to the prompt. For our Vellum & Stars eggshell aesthetic, the difference is small; for surfaces that need photorealism or strong brand consistency, Replicate's flux-dev model is worth the cost. We have not seen a quality gap large enough to recommend Replicate on cost-per-image grounds alone; the recommendation is wall-time-driven.

Reliability and the failure-mode pattern

Across thirty generations on our two production surfaces, we saw one Pollinations soft-failure (a generation that returned 502 after retry) and zero Replicate failures across the evaluation set. Pollinations' failure mode is recoverable: retry after sixteen seconds usually works. Replicate's failures are billed; you pay for failed runs. For a one-shot deploy, the Pollinations soft-failure rate is negligible. For a build pipeline that runs daily, the cumulative failure probability matters; we have not yet stress-tested either provider beyond fifty consecutive generations.

packages/pseo/src/gen-images.tsts
for (const item of inputs.items) {
  const id = item.imageId ?? `${item.slug}-0`;
  const cdnUrl = `${inputs.cdnBase}/${inputs.cdnSubpath}/${item.slug}/${id}.webp`;
  if (await head200(cdnUrl)) {
    console.log(`skip ${item.slug}/${id} (already on CDN)`);
    continue;
  }
  try {
    const png = await generateOne(item.imagePrompt + inputs.styleSuffix);
    const webp = await sharp(png).webp({ quality: 85 }).withMetadata({}).toBuffer();
    const outDir = path.join(tmp, inputs.cdnSubpath, item.slug);
    await fs.mkdir(outDir, { recursive: true });
    await fs.writeFile(path.join(outDir, `${id}.webp`), webp);
  } catch (err) {
    failed.push(`${item.slug}/${id}`);
  }
}
await rsync(tmp, inputs.rsyncTarget);
return { failed };
The per-slug image-generation loop. HEAD-checks the CDN first so re-runs are idempotent; falls through to provider + sharp + write only when an image is missing.

Sources

Every claim on this page traces back to one of the verifiable sources below. Citations are publisher-stable; we avoid blog posts and other URLs that may not survive a year.

  1. Pollinations.ai: free generative image API (Pollinations)
  2. sharp: high-performance Node.js image processing (sharp)
  3. MDN: Cache-Control header (MDN)
  4. RFC 9111: HTTP caching (IETF)
  5. Cloudflare docs: how the cache works (Cloudflare)
  6. web.dev: serving responsive images (web.dev)
  7. web.dev: Largest Contentful Paint (LCP) (web.dev)
  8. Google: page experience signals (Google Search Central)
  9. Google: creating helpful, reliable, people-first content (Google Search Central)
  10. MDN: img srcset attribute (MDN)
Free site auditor

See how the pattern applies to your site

The auditor uses the same evidence bundle described above (robots, sitemap, page samples) and returns a structured opinion. No signup, no email required to see results.