Image upscaling API with no sign-up

What signing up for an image API usually involves

Most image APIs put five steps between you and the first result. You create an account with an email address and a password. You verify the email, sometimes a phone number too. You generate an API key in a dashboard and learn where it must and must not appear. You add a card, because the free tier runs out before you have finished testing. Finally you pick a plan, and from that moment a counter runs whether you send one request or ten thousand.

None of that is about the pixels. It exists so the provider can identify you, bill you monthly and switch the key off when the plan lapses. For a developer who wants to upscale a batch of product photos once a quarter, or wire background removal into an internal tool, the overhead is out of proportion to the job.

Airpic's image upscaling API drops all of it. There is no sign-up and no developer dashboard. You buy a credit pack on the website, you receive a code that begins with 0TK-, and that code is your token. You send it in a standard Authorization: Bearer header, exactly as you would send an API key, and it works from the first call. The website and the API share the same credits, the same tools and the same job ids: a job you create from a script can be inspected in the browser, and a code you bought for the browser can be used from a server.

This is the same approach as the rest of the service, described in AI photo editor with no sign-up: no email, no password, nothing stored about you beyond the balance of a code. Your files follow the same rule. They are processed on a GPU in the European Union and deleted at most one hour after you download the result, or one day if you never do, and they are never used to train models.

One consequence is worth stating plainly before you build on it: because there is no account, there is no password reset. The code is the balance. Treat it like cash, keep it in your server's secrets store, and never ship it inside a browser bundle or a mobile app, where anyone who opens the developer tools can read it and spend it.

What you can call: upscale image API, background removal API and the rest

Every tool on the website is reachable with the same request shape: one multipart POST to the jobs endpoint with the file, the tool name and a JSON string of options. The options are the controls you see on the web page, with the same names and the same ranges. Nothing is held back for a higher tier, because there is no tier.

• Upscale image API. Scale ×2 or ×4 in a single pass, or ×8, ×16 and a maximum mode that tiles the image up to 600 megapixels for large-format print. A style switch chooses between sharp detail and a more painterly fill for very poor sources, and a faces flag runs face enhancement on every detected face. The web version is the tool to upscale an image; the API gives you the same engine with the same output.
• Background removal API. Edges are computed at full resolution, hair included, with no halo. You choose a transparent, white or black background, or any colour as a hex value, plus the output format. On the site it is the tool to remove the background from a photo, and it needs no registration in either form.
• Compress. JPEG, WebP, AVIF or PNG output, a quality value, a target size in KB, a maximum side in pixels, and an option to strip metadata while keeping the ICC profile. It runs on CPU and finishes in seconds. The web tool to compress images is free in full, and so is the API call.
• Restore, denoise, unblur, colourise, erase. Each takes a strength value between 0.25 and 1. Restoration accepts a faces flag, optional recolouring and up to three reference photos of the same person to guide the likeness; the web tool to restore old photos exposes the same controls. Object removal takes a second form field with a PNG mask the size of the image, painted pixels over transparent. Anime and flat illustration have their own upscaler.
• Video. Transcription in 99 languages returning a ZIP with TXT, SRT, VTT and JSON; video to document with a summary and timestamped chapters as DOCX, PDF or Markdown; and shorts, 9:16 clips with word-level captions. The web entry point is the tool to transcribe video to text.

Images arrive as JPG, PNG (with alpha and 16-bit), WebP, TIFF, HEIC or AVIF. EXIF orientation is applied and the ICC profile is preserved in the output, so a photo shot on a phone comes back the right way up and in the right colour space without a conversion step on your side. Files over about 90 MB, which in practice means video, use a chunked upload: you ask for an upload id, send the parts one PUT at a time, confirm the list of parts, and then create the job with the upload id instead of the file.

The exact option names, value ranges and size limits per tool are kept in one place, the developers page at airpic.app/developers, so that the table there is always the current one. Treat this article as the map and that page as the reference. If a parameter is not listed there, do not assume it exists.

How a pay-as-you-go AI image API is billed here

The billing rule is the same as on the website and it fits in one sentence: processing is free, the preview is free, and the first full download of a job costs one credit. Nothing else is metered. There is no monthly minimum, no charge for jobs that fail, and no fee for polling.

In practice that changes how you write the client:

1. Create the job with the file and the options. No credit is touched.
2. Poll the status. When it reports done, the response includes the output width and height, the extension, a download URL and a preview URL. The preview, up to 1600 pixels on the long side, can be fetched without any token. You can look at it, run your own quality check on it, or show it to a user before anything is paid.
3. Download the full file with the bearer header. This is the only step that spends a credit.

Credits come in three kinds, and the API uses the same ones the website does: standard for ×2, ×4 and the single-image tools, large format for ×8, ×16 and the maximum mode, and video for the three video tools. If the code does not hold the right kind, the download returns HTTP 402 with a field telling you which kind is missing, and nothing is deducted. Downloading the same job again later is free, because the job is marked as paid. A balance call returns the remaining standard, large-format and video credits of a code at any time, which is useful for a dashboard or an alert when a batch is about to run dry.

Credits never expire. A pack bought for a one-off project is still valid a year later, which is the whole point of the model described in pay per image, not per month. The free full download of the month, counted per IP address, applies to API calls too, so you can run a complete end-to-end test before buying anything. The large-format print pack is the one sold with API access in its name, and it is the natural choice for batch work because it carries the kind of credit the ×8, ×16 and maximum modes need; but every code behaves the same way as a token, whatever pack it came from.

Compression is the exception in the other direction: it is free in full, processing and download, from the website and from the API alike, because it runs on CPU in seconds. Current pack sizes and prices are on the Why Airpic page; they are not repeated here so that this article does not go stale when they change.

Honest limits: hourly quota, daily GPU cap and cold starts

An API with no account cannot know who you are, so its fairness rules are attached to what it can see: your IP address and the code you present. Design around these three limits from the start.

• 20 jobs per hour per IP address. This is the ceiling for anonymous use and it protects the GPU pool from scripts that loop on errors. For a nightly batch of a few hundred images it means spreading the work across hours, or asking for a higher limit: write to the support address with your code and the quota is raised for it. Keep at most two jobs in flight at once; the workers are shared with everyone else.
• A daily cap on total GPU time. The whole service has a budget of GPU seconds per day. If it is reached, new jobs are refused with a clear error until the counter resets the next day, and no credit is spent. A client that cannot tolerate a refused job should catch that response and retry later rather than treat it as a bug.
• Cold starts. The models live on serverless GPUs that shut down when idle. The first job after a quiet period pays for a worker to boot and load its weights, which adds tens of seconds, typically 60-90, on top of the processing time. Once warm, the same job runs in seconds. Poll every one or two seconds and set a timeout measured in minutes, not seconds.

Two more things that are limits in practice rather than quotas. Large-format modes on big inputs take several minutes: ×16 on a 2 megapixel photo produces more than 500 megapixels, and that is a lot of tiles. And the megapixel ceilings per tool (around 32 megapixels for the direct tools, lower for unblur and restoration) are enforced before the job starts, so check dimensions on your side to avoid a wasted round trip.

Here is the piece of common advice that is wrong for an API client: "always upscale ×4 and downsize afterwards". On a website that costs you a click. In a batch it costs a large-format credit every time you need more than ×4, and minutes of waiting per image. Pick the smallest factor that reaches your target size. For a 1200 pixel product shot that needs to become 2400 pixels, ×2 is the right answer and it is a standard credit. For scale, in our tests on 3 September 2026 photos of 0.3 to 1 megapixel took 6 to 9 seconds at ×4 on a warm worker, and around 17 seconds with face enhancement switched on; a batch of five hundred at ×4 is roughly an hour of warm GPU time spread over the hourly quota, not an afternoon.

A minimal example in words, then try it by hand

Here is the whole integration without a single line of SDK, because there is none to install. Any HTTP client that can send a multipart form will do: curl, a Python script, a serverless function, a spreadsheet macro.

1. Upload. Send a POST to the jobs endpoint with three form fields: the image file, the tool name (for upscaling it is upscale) and an options field containing a small JSON string, for example mode set to x4 and style set to sharp. Add a language field if you want error messages in a particular language. The response is a JSON object with a job id. Store it.
2. Check the status. GET the job by its id every one or two seconds. You will see queued, then processing, then done or error. On done, read the output block: width and height of the result, the file extension, the download URL and the preview URL. Fetch the preview if you want to inspect the result; it costs nothing.
3. Download. GET the download URL with the header Authorization: Bearer followed by your 0TK code. The first time, one credit of the right kind leaves the code and the full-resolution file streams back. If the code is empty you receive a 402 with the kind of credit needed; nothing is lost and the result stays available for up to a day, so you can top up and download. Delete the file on your side when you are done; Airpic deletes its copy one hour after the download.

That is the complete loop. For the per-tool option names, the chunked upload for video, the balance call and the mask field for object removal, the developers page at airpic.app/developers has the current table and a curl example of each call.

Before you write any code, run the same job once by hand so you know what a good result looks like at your settings. In our tests on 19 September 2026, a 320 × 240 pixel photo upscaled ×4 to 1280 × 960 pixels in 5 seconds on a warm worker; that is the speed to expect once your batch is running. The steps in the browser:

1. Open the tool to upscale an image and drop a file from the batch you plan to process. A card shows the input size and the output size at the current factor, so you can see at once whether ×2 already reaches your target.
2. Under Use, leave Standard unless the result goes to print. Open Advanced settings only if you want the painterly style for very low-quality sources or a specific output format; the defaults are what the API uses when you send no style or format option.
3. Tick Enhance faces only if the photo contains people. On products, text or illustrations it has nothing to find and only adds seconds to every job in the batch.
4. Press Process. The preview appears in the comparator with a divider you can drag; View at 100% shows real pixels rather than a scaled view. If this is good enough, the API will produce the same file from the same options.
5. Download once to confirm the full-resolution output; this uses the free download of the month or one credit. Then copy the exact option values into your script, and the first automated run should hold no surprises.

FAQ

Keep reading