🔔 Image Normalization API

Normalizes each supplied image — auto-rotate by EXIF, resize to a max dimension, convert format, compress, and strip unsafe metadata — and returns the normalized artifact plus before/after dimensions, color mode, MIME, byte sizes, and SHA-256s. Upload-driven; there is no search or discovery source.

Output is one image_normalization row per result; billing is pay-per-event, the value event being one image normalization api verdict (a $0.02 start fee per run, then $0.05 per image normalization api verdict).

No login, no API key and no CAPTCHA solving are involved: the source is read logged-out with an identified contact User-Agent.

Output is one image_normalization row per result; billing is pay-per-event, the value event being one image normalization api verdict (a $0.02 start fee per run, then $0.05 per image normalization api verdict).

No login, no API key and no CAPTCHA solving are involved: the source is read logged-out with an identified contact User-Agent.

📊 Sample Output

Real rows from run cHKPTev94dGgl8TEn on build 0.1.9 (2026-09-23), the same input as the Quick start below — every value is as the source published it (emails masked, long text shortened):

input_ref	source_type	orig_format	orig_mode	out_key	out_mime
data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAA...	data	PNG	LA	norm-0-data.png	image/png

The run finished with the status message: 1 image(s): 1 normalized+charged, 0 unbilled (rejected/blocked/unreachable/unparseable) | image_normalization_check charged=1

✅ What you get

Each row is flat JSON with these fields (from the dataset schema and the sample run; a field the source does not publish for a given row is null):

• input_ref (string/null) — e.g. data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAA...
• source_type (string/null) — e.g. data
• source_url (string/null) — null in every sample row
• orig_format (string/null) — e.g. PNG
• orig_mode (string/null) — e.g. LA
• orig_width (integer/string/null) — e.g. 1
• orig_height (integer/string/null) — e.g. 1
• orig_pixels (integer/string/null) — e.g. 1
• orig_bytes (integer/string/null) — e.g. 70
• orig_sha256 (string/null) — e.g. bf780d1105e7cbb81732af82d45fa3c8a359a130cb5cc6083abdf296e1445e2a
• out_format (string/null) — e.g. PNG
• out_mode (string/null) — e.g. LA
• out_width (integer/string/null) — e.g. 1
• out_height (integer/string/null) — e.g. 1
• out_bytes (integer/string/null) — e.g. 68
• out_sha256 (string/null) — e.g. eba7a7a39459c37cc784afeb2ef1613d0b046b4e1988984fd2f801b568cb7a08
• out_key (string/null) — e.g. norm-0-data.png
• out_mime (string/null) — e.g. image/png
• metadata_stripped (boolean/null) — e.g. True
• operations_applied (string/null) — e.g. ["auto_rotate(exif): no-op","metadata: stripped (no exif/icc/other carried)","en
• warnings (string/null) — null in every sample row
• faults (string/null) — null in every sample row
• record_type (string) — e.g. image_normalization
• source (string/null) — e.g. buyer-supplied images (base64 / data: URL / KVS key / permitted image URL)
• outcome (string/null) — e.g. answer
• input (string/null) — e.g. data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAA...
• outcome_reason (string/null) — null in every sample row
• observed_at (string/null) — null in every sample row
• terminal (string/null) — null in every sample row

✅ What you get

Each row is flat JSON with these fields (from the dataset schema and the sample run; a field the source does not publish for a given row is null):

• input_ref (string/null) — e.g. data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAA...
• source_type (string/null) — e.g. data
• source_url (string/null) — null in every sample row
• orig_format (string/null) — e.g. PNG
• orig_mode (string/null) — e.g. LA
• orig_width (integer/string/null) — e.g. 1
• orig_height (integer/string/null) — e.g. 1
• orig_pixels (integer/string/null) — e.g. 1
• orig_bytes (integer/string/null) — e.g. 70
• orig_sha256 (string/null) — e.g. bf780d1105e7cbb81732af82d45fa3c8a359a130cb5cc6083abdf296e1445e2a
• out_format (string/null) — e.g. PNG
• out_mode (string/null) — e.g. LA
• out_width (integer/string/null) — e.g. 1
• out_height (integer/string/null) — e.g. 1
• out_bytes (integer/string/null) — e.g. 68
• out_sha256 (string/null) — e.g. eba7a7a39459c37cc784afeb2ef1613d0b046b4e1988984fd2f801b568cb7a08
• out_key (string/null) — e.g. norm-0-data.png
• out_mime (string/null) — e.g. image/png
• metadata_stripped (boolean/null) — e.g. True
• operations_applied (string/null) — e.g. ["auto_rotate(exif): no-op","metadata: stripped (no exif/icc/other carried)","en
• warnings (string/null) — null in every sample row
• faults (string/null) — null in every sample row
• record_type (string) — e.g. image_normalization
• source (string/null) — e.g. buyer-supplied images (base64 / data: URL / KVS key / permitted image URL)
• outcome (string/null) — e.g. answer
• input (string/null) — e.g. data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAA...
• outcome_reason (string/null) — null in every sample row
• observed_at (string/null) — null in every sample row
• terminal (string/null) — null in every sample row

Output & billing

One row per image. A completed normalization carries outcome: answer and bills once. A corrupted, oversized, or disallowed-format input, or a blocked / unreachable URL, is delivered as an unbilled status. Push-then-charge: every row is delivered before its charge, so a billing hiccup can only ever undercharge.

⚙️ Sample inputs

1. Quick start — the Store example (this is what the sample above came from)

{
  "images": [
    "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mP8z8BQDwAEhQGAhKmMIQAAAABJRU5ErkJggg=="
  ],
  "targetFormat": "PNG",
  "maxDimension": 2048,
  "quality": 85,
  "stripMetadata": true,
  "maxBytes": 26214400,
  "maxPixels": 40000000,
  "userAgent": "Mozilla/5.0 (compatible; NexGenWatchBot/1.0; +https://apify.com/nexgenwatch)"
}

The sample run charged exactly: 2 × $0.02 apify-actor-start + 1 × $0.05 image_normalization_check = $0.09 on the Free tier — every delivered row was billed.

This actor takes no size cap or filter beyond the input above, so the Store example is the one input to run; put it on a schedule to keep the data current.

🧾 JSON sample record

One real record from run cHKPTev94dGgl8TEn, exactly as it lands in the dataset (emails masked, long text shortened):

{
  "input_ref": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAA...",
  "source_type": "data",
  "source_url": null,
  "orig_format": "PNG",
  "orig_mode": "LA",
  "orig_width": 1,
  "orig_height": 1,
  "orig_pixels": 1,
  "orig_bytes": 70,
  "orig_sha256": "bf780d1105e7cbb81732af82d45fa3c8a359a130cb5cc6083abdf296e1445e2a",
  "out_format": "PNG",
  "out_mode": "LA",
  "out_width": 1,
  "out_height": 1,
  "out_bytes": 68,
  "out_sha256": "eba7a7a39459c37cc784afeb2ef1613d0b046b4e1988984fd2f801b568cb7a08",
  "out_key": "norm-0-data.png",
  "out_mime": "image/png",
  "metadata_stripped": true,
  "operations_applied": "[\"auto_rotate(exif): no-op\",\"metadata: stripped (no exif/icc/other carried)\",\"encode: PNG q=n/a\"]",
  "warnings": null,
  "faults": null,
  "record_type": "image_normalization",
  "source": "buyer-supplied images (base64 / data: URL / KVS key / permitted image URL)",
  "outcome": "answer",
  "input": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAA...",
  "outcome_reason": null
}

🔧 How it works

Transport. Plain HTTPS from the Apify platform, no proxy. robots.txt is read first and a disallowed path is never fetched. Every request carries an identified contact User-Agent.

Terminal states. A run ends NORMAL, CAPPED (your cap was reached), PARTIAL (something was withheld and the message says what), GENUINE_EMPTY (the source was read and truly had nothing in scope) or BLOCKED (the source refused or changed shape — the run FAILS loud and bills nothing). A zero-row run is never reported as a silent success.

Charging. Each image normalization api verdict is charged at the moment it is pushed (image_normalization_check); a row that fails to charge is not delivered, so the dataset count always equals the charged count.

🔧 How it works

Transport. Plain HTTPS from the Apify platform, no proxy. robots.txt is read first and a disallowed path is never fetched. Every request carries an identified contact User-Agent.

Terminal states. A run ends NORMAL, CAPPED (your cap was reached), PARTIAL (something was withheld and the message says what), GENUINE_EMPTY (the source was read and truly had nothing in scope) or BLOCKED (the source refused or changed shape — the run FAILS loud and bills nothing). A zero-row run is never reported as a silent success.

Charging. Each image normalization api verdict is charged at the moment it is pushed (image_normalization_check); a row that fails to charge is not delivered, so the dataset count always equals the charged count.

What you submit

images — images (base64, data: url, kvs key, or permitted image url). Each item is an image supplied as a base64 string, a data: URL, a key naming an entry in the run key-value store, or a permitted direct http(s) image URL (URL items pass the per-origin runtime gate — robots + SSRF — before any fetch; there is no search/discovery source). Per image ONE normalized artifact (stored in the run key-value store) plus before/after dimensions, color mode, MIME, byte sizes, SHA-256s, the operations applied, and warnings. Operations: auto-rotate by EXIF, resize to a max dimension, convert format, compress, and strip unsafe metadata. Guards: format allowlist, a pixel-count cap, and a decompression-bomb guard. A completed normalized image bills; a corrupted, oversized, or disallowed-format input, or a blocked/unreachable URL, is an unbilled status.

The runtime source gate (per submitted target)

You choose the targets, so the source contract is enforced at run time — for every URL-mode item (base64 / data: URL / key-value-store items are local and are not fetched) — per origin, before any page is read:

• robots.txt fetched once per origin and honored; no robots / 404 = permitted, a disallowed path or an unavailable/forbidden robots file = BLOCKED.
• SSRF guard — every host is resolved and must be public; private, loopback, link-local, reserved and cloud-metadata addresses are refused before a socket opens.
• HTTP — 403 / 429 / 5xx = BLOCKED; DNS / timeout / connection faults = UNREACHABLE.

A BLOCKED or UNREACHABLE target is delivered as an unbilled status row — never a broken-site verdict, and never charged.

Untrusted-file guards

Every image is treated as untrusted: a byte-size cap and a pixel-count cap (checked from the declared dimensions before the image is decoded) guard against decompression bombs, and only an allowlisted set of input formats is accepted.

Why this and not the obvious alternative

a deterministic normalized image artifact (auto-rotate, resize, convert, compress, strip unsafe metadata) with before/after hashes and dimensions and hard pixel/byte bomb guards — an upload-driven normalization API, not an image search or scraper.

What is not done. No login, no cookie or CAPTCHA bypass, no private or personal-account data, no browser automation.

What is not done. No login, no cookie or CAPTCHA bypass, no private or personal-account data, no browser automation.

💰 Pricing example

Event	Free	Bronze	Silver	Gold
Actor Start (apify-actor-start)	$0.02	$0.02	$0.02	$0.02
Image Normalization API verdict (image_normalization_check)	$0.05	$0.04	$0.04	$0.03

Worked at the live Free-tier price:

• 1 image normalization api verdicts: $0.02 start + 1 × $0.05 = $0.07
• 25 image normalization api verdicts: $0.02 start + 25 × $0.05 = $1.27
• 100 image normalization api verdicts: $0.02 start + 100 × $0.05 = $5.02

A run that delivers zero rows charges the $0.02 start fee only. A BLOCKED run (source refused) fails loud and charges no value event. The start fee is charged once per GB of run memory; the default run memory is 4096 MB.

Yield on the sample run: 1 image(s): 1 normalized+charged, 0 unbilled (rejected/blocked/unreachable/unparseable) | image_normalization_check charged=1.

⚖️ Legal & ToS

This actor reads public data only. It collects only what the source publishes to any visitor, keeps to the source's robots rules (checked on every run), identifies itself with a contact User-Agent, and does not access accounts, private data or anything behind authentication. Use the output in line with the source's terms and your local law; the intended use is B2B research and monitoring.

❓ FAQ

Q: Do I need an API key or a login?
A: No. the input schema has no key field and the actor carries no secrets.

Q: Why did my run return 0 rows?
A: Read the run's status message. GENUINE_EMPTY means the source was read and had nothing in scope for your input; BLOCKED means the source refused and the run failed without billing a value event — retry later or narrow the input. A zero-row run bills the start fee only.

Q: How fresh is the data?
A: Every run reads the source live at run time; nothing is cached between runs. Put it on a schedule for a continuous feed.

Q: What formats can I export?
A: The dataset downloads as JSON, CSV, Excel, XML or RSS from the run's Dataset tab or the Apify API, and any run can push to a webhook or integration.

Q: How is this different from the other NexGen Watch actors actors?
A: Same output shape and billing model; this one covers Image Normalization API. The siblings under Related Actors cover the other sources or slices — run several on one schedule for a combined feed.

Q: Are there rate limits?
A: The actor paces itself against the source and honours its robots rules; there is no per-buyer limit beyond your Apify plan's concurrency.

🆘 Troubleshooting

• Run FAILED with BLOCKED → the source refused the request or changed its page shape → nothing was billed beyond the start fee; retry after a while, and if it persists open an Issue with the run id.
• Fewer rows than expected → the source had fewer items in scope → widen the input.
• Input validation error on start → a field is outside the schema's allowed values → start from the Quick start block and change one field at a time.
• Run TIMED-OUT → a very wide request on a slow day → raise the run timeout in Run options or narrow the input; what was delivered before the timeout is still in the dataset.

🔗 Related Actors

• DOL H-2A Labor Certifications (FY2026 Q3, quarter-to-date) — same family, different source
• DOL H-2B Labor Certifications (FY2026 Q3, quarter-to-date) — same family, different source
• Hong Kong New Company Registrations Scraper — Scrape newly registered Hong Kong companies: name, company number, incorporation date and type. One record per result. Pay per result
• IPEDS 12-Month Enrollment (EFFY2024) — same family, different source
• IPEDS Academic Libraries (AL2023) — same family, different source
• IPEDS Degree & Certificate Completions (C2024_A) — same family, different source
• 🏢 About NexGenData — NexGen Watch is NexGenData's fleet of 281 public data actors built on official sources, pay-per-result. Browse the catalog at apify.com/nexgenwatch.

⭐ Found this useful?

If this actor saved you a manual check, a quick review on the Apify Store helps other teams find it. Feature request or a source that changed? Open it from the Issues tab — every one is read.