We ran five images through Pixelcut's upscaler at 2x and 4x on August 12, 2026 — a product packshot, a busy photo scene, an AI-generated portrait scene, a line-art graphic and a deliberately degraded photo. Every successful upscale came back in 2.6 to 4.0 seconds, at exactly the requested scale, for $0.10 per image. At 2x, results stayed faithful to the source. At 4x, the tool redrew fine texture: our dress's real lace came back as invented floral outlines — sharp-looking, but no longer the actual fabric. Two of the five inputs were corrupted by our own image hosting before they ever reached Pixelcut, so text graphics and heavily compressed sources remain untested; we say so below rather than guessing. Here is what we measured, what the API currently supports, and when an upscaler is the wrong tool.
What we tested and how
On August 12, 2026 we submitted each image to the Pixelcut upscaler API twice — once at 2x and once at 4x — using a fresh developer account. We recorded submission and completion times, output dimensions, formats and file sizes, then inspected each result against its source on the same display and scored it on a five-point rubric we wrote down before running anything (5 means no material defect found under that inspection; 1 means the output is less usable than the input). Failures were kept, not rerun until they looked good.
Two runs never produced a verdict: our upload pipeline corrupted the line-art PNG and the deliberately degraded JPEG before Pixelcut received them, so those four submissions are excluded as invalid rather than scored — an honest gap we will close in a follow-up round. One further run, the AI-generated image at 4x, was blocked when the account's free credits ran out.
Results
| Input | Scale | Time | Output | Verdict |
|---|---|---|---|---|
| Product packshot, 787×750 | 2x | 3.6 s | 1574×1500 JPEG | Faithful; lace pattern slightly bolder — 4/5 |
| Product packshot | 4x | 4.0 s | 3148×3000 JPEG | Sharp at a glance, but the fabric is redrawn: invented floral texture at 100% — 3/5 |
| Photo scene with model, 750×750 | 2x | 3.7 s | 1500×1500 JPEG | Clean and faithful — 4/5 |
| Photo scene with model | 4x | 4.0 s | 3000×3000 JPEG | Face, jewelry and architecture crisp; dress weave mildly reinterpreted — 4/5 |
| AI-generated scene, 750×750 | 2x | 4.0 s | 1500×1500 JPEG | Natural face, crisp hair detail — 4/5 |
| AI-generated scene | 4x | — | — | Not run: free credits exhausted |
| Line-art graphic; degraded JPEG | both | — | — | Excluded: inputs corrupted before reaching the API |
The one thing to watch: 4x invents texture
The most useful finding was not speed or sharpness — it was fidelity. At 2x the upscaler behaves conservatively: edges tighten, patterns stay recognisably the same. At 4x it starts painting. Our lace dress is the clearest case: the upscaled fabric shows confident, cartoon-outlined flowers that do not exist in the garment, with smeared patches where the pattern lost coherence. Viewed at product-card size it looks excellent; inspected at 100%, it is a different textile.
The practical rule we would follow from this test: use 2x when the image represents a real product a customer will receive, and treat 4x as a look-enhancement for scenes, social posts and backgrounds where nobody will compare threads. If you sell textured goods — lace, knitwear, engravings — check any upscaled image at full zoom before publishing it. If your enhanced photos are headed for a storefront, size them to the platform's requirements too; our guide to Shopify image sizes covers the dimensions that matter there.
What the Pixelcut upscaler supports right now
All of the following was verified against Pixelcut's official documentation on August 12, 2026 — these are vendor facts that can change, so check the linked pages before building on them.
Scale factors are 2x and 4x only. Maximum output resolution is 6000×6000 pixels; inputs are capped at 6000×6000 and 25 MB, with a documented minimum of 64×64. Output is JPEG by default, or PNG when the input contains transparency — our line-art runs confirmed the PNG behaviour. The API is synchronous: a request returns a JSON result_url within seconds, and that link is valid for about an hour, so download promptly. Pricing is 10 credits per upscale at $0.01 per credit — $0.10 per successful image. A new developer account includes 100 free credits (ten upscales — our pilot consumed exactly that), and paid plans start at $10 per month for 1,000 credits. In our runs, failed requests did not consume credits. Sources: the upscale API reference and pricing documentation.
Worth noting: an earlier version of this article claimed scaling "typically ranges from 2x to 8x" and free choice of output format. Neither matches the current documentation, and we have corrected both.
How to use the Pixelcut upscaler API
The flow we actually used:
- Create a developer account and copy your API key.
- Send a POST request to
https://api.developer.pixelcut.ai/v1/upscalewith theX-API-KEYheader, either as JSON ({"image_url": "https://…", "scale": 2}) or as multipart form data with animagefile field. - Read
result_urlfrom the JSON response and download the file within the hour.
We tested the API directly; this article no longer describes the mobile or web app's menus, modes or batch options, because we have not verified them.
When an upscaler is the wrong tool
An upscaler makes the image you have bigger. It cannot recover information that was never captured, and past 2x it will start inventing plausible detail instead. Skip upscaling when the source is smaller than the documented 64×64 minimum, when exact material fidelity at 100% matters more than display size, and when what you actually want is a different image — a new background, a new scene, a styled composition. That last case is generation, not enhancement: ProductAI builds new product scenes from photos you already have, which is a different job than making pixels bigger — our overview of AI product photography explains where that fits in an ecommerce workflow.
Methodology and update note
Benchmark run on August 12, 2026 against Pixelcut's production API (five inputs, ten planned runs, raw logs and unmodified outputs retained internally). Inputs were ProductAI-owned images. Runs were executed and scored by an automated agent under a pre-registered rubric; scores reflect whole-image inspection on one display, and 100%-crop inspection is planned for the follow-up. Known limits of this test: a five-image pilot rather than the full twenty-image set, two categories invalidated by our own tooling, no app-UI coverage, and no competing tools tested. Vendor facts (prices, limits, formats) were verified on the same date against the documentation linked above.
