Blog

Upscale images for professional printing: Airpic for studios

To upscale images for professional printing you need pixels the client never sent, a colour profile that survives the trip and a way to pay that does not mean creating a user account for every designer. Airpic does the three in the browser: tiled AI upscaling up to 600 megapixels, the ICC profile kept, and bearer credits that one code shares across the studio and the same code reuses as an API key. Here is how it fits a print shop's workflow, from the 1 MP file that arrives by email to the file placed in the layout.

The 1 MP file and the three-metre banner

Every print shop knows the email. The client needs a 300 × 200 cm banner for Thursday, and attached is a JPEG of 1,200 × 850 pixels, about 1 megapixel, saved from their website or forwarded from a phone. Spread across three metres of vinyl, those 1,200 pixels work out at roughly 10 pixels per inch, so each pixel is close to 2.5 mm wide. From the pavement the image reads as a blur; in the hands of the client checking the proof it reads as a mistake.

There have traditionally been three ways out, and none of them is good. You refuse the job and the client goes elsewhere. You accept it and let the layout program interpolate the file up to size, which keeps the same information and spreads it thinner: soft edges, staircase diagonals and a complaint when the banner goes up. Or someone on the team runs the picture through a general AI assistant and gets back a new image that resembles the original but is a reinterpretation at one or two megapixels, often with the lettering changed and faces that belong to nobody. That file is not bigger in any useful sense.

What the job needs is arithmetic, not optimism. A banner read from five metres away needs around 60 ppi; read from three metres, around 100 ppi. For a 300 cm width that means about 7,100 pixels in the first case and 11,800 in the second. From a 1,200 pixel original that is a factor of ×6 to ×10, which in practice means an ×8 or ×16 pass with a model that reconstructs plausible texture instead of stretching the pixels it has. The rest of this article is about doing that with an AI image upscaler built for print sizes, in a way that fits a studio with several designers, a shared budget and jobs that arrive by the dozen.

What an AI tool for print shops has to do differently

A consumer upscaler stops at ×4 and a few thousand pixels, because that is all a screen needs. A print shop needs a tool that keeps going without breaking the file or the colour. These are the five things Airpic's Image Upscaler does for that case.

  • Tiled upscaling up to 600 megapixels. ×2 and ×4 are a single direct pass on inputs up to 32 MP. ×8, ×16 and Max run two passes over a mosaic of roughly 250 tiles that are blended back together without visible seams, on inputs up to 37.5 MP. Max works out the largest output that fits in 600 MP for your particular file: for the 1,200 × 850 example above that is a little over ×24, around 29,000 pixels wide.
  • A mode for large printing. Under “Use”, the option “For large printing” adds fields for the print width and height in centimetres and a viewing distance (in hand, desk, wall, room, poster, banner). The tool shows the ppi you will get and recommends the smallest factor that reaches it, so nobody renders 600 MP for a job that needed 65.
  • Two styles. Extreme sharpness uses the photo model and rebuilds hair, fabric and foliage with macro-like detail; Painterly fill uses the illustration model and produces flat, brush-like areas that suit posters, drawings and soft artwork. For flat vector-style art there is also a separate anime and illustration upscaler.
  • The ICC profile travels with the file. Whatever profile the input carries (sRGB, Adobe RGB, a camera or scanner profile) is kept in the output, and the EXIF orientation is applied. Inputs can be JPG, PNG with transparency and 16-bit depth, WebP, TIFF, HEIC and AVIF. Output is PNG, JPEG or WebP at your choice; results above 32 MP default to top-quality JPEG so the file stays manageable.
  • Faces, when the job has them. The “Enhance faces (AI)” option rebuilds eyes, skin and hair on each detected face after the upscale. It is for portraits that will be read up close, not for a crowd in a landscape: faces under about 100 pixels in the original drift towards a generic face, and the tool warns you when that happens.

Speed matters when there are thirty files in the queue. In our production tests on 3 September 2026, with photos of 0.3 to 1 megapixel, a ×4 pass took between 6 and 9 seconds; with “enhance faces” switched on, about 17 seconds. On 19 September 2026 a 320 × 240 pixel photo went to 1,280 × 960 in 5 seconds. A large ×16 render is a different animal and can take several minutes, and the first job after the engine has gone cold adds 60 to 90 seconds; the progress bar shows an estimate so you can plan around it.

One honest limit before you promise anything to a client: an upscaler reconstructs texture, not meaning. It will not redraw a logo or small type into clean edges. For lettering, ask for the vector or the original artwork, and if that is impossible read how to upscale a pixelated logo first, because the sensible route is different.

Shared credits for a team: one code for the whole studio

Most online tools assume one user, one email, one subscription. A studio is four designers on three machines and an account manager who approves the spend. Airpic has no accounts at all: a purchase gives you a credit code starting with 0TK, and the code is the balance. Whoever pastes it in the browser can download with it. No seats, no invitations and no password to reset when someone leaves.

In practice this is how studios use it:

  • One code, pinned where the team can find it. Paste it once in the tool on each workstation; the browser remembers it. Everyone previews for free and spends a credit only when a full-resolution file is downloaded.
  • Separate codes when you need to know who spent what. A bearer code has no audit trail by design. If you bill clients per job, buy a pack per client or per designer and the balance of each code is the record.
  • Treat the code like petty cash. Anyone holding it can spend it, so keep it out of documents that leave the building, and check the balance when the pack starts running low.

The packs that matter for print are the large-format ones: the Large Format Pack, Print Pro, and Print 100 + API, which also covers API access. Large format means output up to 600 MP, which is any ×8, ×16 or Max render; ×2 and ×4 results up to 32 MP count as standard images. Standard images come to €0.099 per image in the 100-image pack and €2 for 12 images in the Mini pack; large-format downloads start from €0.49 per image in the biggest print pack. Everyone, with or without a code, gets 1 free full download a month, credits never expire, and if a job fails the credit refunds itself. The current price of each pack is on the image upscaler page rather than here, because prices change and articles do not.

Why this matters for a print shop specifically: a client file goes in, is processed on a GPU in the European Union, and is deleted an hour after download (one day at most if nobody downloads it). Nothing is used for training. For studios handling unreleased campaign artwork or other people's family photographs that is not a footnote, and there is a longer explanation in AI image processing in the EU.

An image upscaling API that takes the same code

Once the volume passes a handful of files a day, nobody wants to drag and drop. Airpic's public API is the same set of endpoints the web page calls, and the credit code you already bought is the API key: you send it as a bearer token. No developer registration, no separate key to rotate, no SDK. The documentation lives at airpic.app/developers/ and the whole thing is three HTTP calls.

  1. Create the job. POST the file to /api/jobs with the field tool=upscale and an options JSON. For print work the options you care about are mode (x2, x4, x8, x16 or max250), style (sharp or paint), faces (true or false) and format (png, jpg or webp). The response is a job id.
  2. Poll the status. GET /api/jobs/JOB_ID every second or two until it reports done. The documented flow is polling, so your script waits. When the job finishes you get a free preview URL (up to 1,600 pixels on the long side) and the download URL.
  3. Download the full file. GET /api/download/JOB_ID with an Authorization header carrying Bearer and the code. The first download of a job spends one credit of the right kind, standard for x2 and x4, large format for x8, x16 and max250; downloading the same job again is free. Without balance you get a 402 and a field telling you which kind of credit is missing, and you can query a code's balance at any time.

A flow that works for a studio looks like this:

  • Receive. Client files land in a job folder (upload form, shared drive, an email rule). A small script reads the pixel dimensions and the order's print size and viewing distance.
  • Upscale. The script picks the smallest mode that reaches the target ppi, posts the file with style sharp for photographs or paint for illustrations, and stores the job id against the order number.
  • Review. A designer opens the preview URL in the browser, which costs nothing, and checks the areas that matter at 100 %: skin, lettering, repeated patterns. If the style was wrong, the job is rerun with the other style before any credit is spent.
  • Lay out. Only approved jobs are downloaded, and that is the moment a credit is used. The full file, ICC profile intact, is placed in the layout at final size and goes to the RIP like any other asset.

The same three calls work for every tool, so the pipeline that upscales can also remove the background from a product photo before upscaling it for a point-of-sale display, or compress images for the client's online proof, which is free and never spends a credit. Keep the code on your server: it is a bearer token, and anything that can read it can spend it, so it does not belong in a browser page or a mobile app binary.

Step by step: upscale an image for print in Airpic

This is the manual route, for the one-off job or for the designer who wants to see what the API is doing. The example is the 1,200 × 850 pixel JPEG and the 300 × 200 cm banner from the beginning.

  1. Open the image upscaler and drop the file in. A file card shows the name, the input dimensions and the output dimensions, which update as you change the factor. JPG, PNG, TIFF, WebP, HEIC and AVIF are accepted, and the EXIF orientation is applied, so a portrait phone photo arrives upright.
  2. Under Use, pick “For large printing”. Type 300 in “Width in cm” and 200 in “Height in cm”, and set “Viewing distance” to “Banner (5 m or more)”. The tool shows the ppi you would get and recommends the smallest factor that reaches the target. For this file the recommendation is ×8: 9,600 × 6,800 pixels, about 65 MP and roughly 80 ppi across the 3 m width, fine for a banner read from across a street. If the client will stand within three metres, change the distance to “Poster (3 m)” and the recommendation moves to ×16: 19,200 × 13,600 pixels, 261 MP, about 160 ppi. The quick presets (Canvas 60×90, Banner 100×200) fill in common sizes.
  3. Open Advanced settings and choose the style. Extreme sharpness for a photograph (this one), Painterly fill for an illustration or a design with flat colour and soft gradients. Leave “Enhance faces (AI)” off unless the banner is a portrait read up close; on a small face it adds time and can replace the person with a plausible stranger. Set the output format: PNG if the result stays under 32 MP and you need it lossless, otherwise JPEG, which is what a 261 MP render will be delivered as in any case.
  4. Press Process image. Processing and previewing are free. The bar shows an estimate: seconds for ×4, several minutes for a big ×16. When it finishes, a before-and-after slider appears with a reduced preview; use “View at 100 %” to inspect at pixel level the parts the client will look at. If the texture looks painted where it should look photographic, go back with “Redo with other settings” and switch the style. Nothing has been charged yet.
  5. Download. The first full-size download each month is free; after that the download uses one credit from your code (large format for ×8, ×16 and Max; standard for ×2 and ×4). The file keeps the ICC profile of the original. Place it in the layout at 300 × 200 cm and send it to the RIP as usual; the one thing Airpic does not do is convert to CMYK, and what happens to your colours in print explains why that step still belongs in your prepress, not in the upscaler. An hour after the download the file is gone from our side.

The viewing-distance rule above is also the correction to the most repeated piece of advice in this trade. “Everything at 300 ppi” is right for a brochure in the hand and wrong for a banner: a 300 × 200 cm banner at 300 ppi would be over 800 megapixels, beyond the 600 MP ceiling and far beyond what the job needs. Pick the ppi for the distance and let the factor follow. The full method, with a calculator for other sizes, is in the guide to upscaling an image for large-format printing; when the file is a client's scanned print rather than a digital original, restore the old photo first and upscale the result. Then open the image upscaler and send the banner on Thursday.

FAQ

Can you upscale a 1 MP image for a 3 metre banner?

Yes, if it is a photograph. A 1,200 × 850 pixel file taken to ×16 becomes 19,200 × 13,600 pixels, about 261 megapixels, which is around 160 ppi across a 3 m width; a banner read from 3 m needs about 100 ppi and one read from 5 m about 60. The model reconstructs texture (skin, fabric, foliage, sky); it does not redraw small text or logos into clean edges, so ask for vector artwork for those.

Does the AI upscaler keep the ICC colour profile?

Yes. The profile embedded in the input (sRGB, Adobe RGB or a scanner profile) is written into the output, and the EXIF orientation is applied. It does not convert RGB to CMYK: that conversion stays in your layout software or RIP, as with any other asset.

Can several people in a studio share Airpic credits?

Yes. There are no user accounts: a purchase gives you a bearer code starting with 0TK, and anyone who pastes it into the tool can download with it. If you need to know who spent what, buy a separate pack per designer or per client; each code's balance is the record. Anyone holding a code can spend it, so treat it like petty cash.

Is there an API to upscale images in bulk?

Yes. The public API uses the same endpoints as the web page and the credit code as a bearer token: one POST to create the job, a GET to poll its status and a GET to download the full file, which is when a credit is spent. The Print 100 + API pack is the one designed for it, and the documentation is at airpic.app/developers/.

We measured it

  • A 320 × 240 px photo upscaled ×4 to 1280 × 960 px in 5 seconds. (19-09-2026)
  • With 0.3 to 1 megapixel photos, ×4 took 6 to 9 seconds; with face enhancement on, about 17 seconds. (03-09-2026)

Measured on airpic.app with the GPU already running. If nobody has used it for a while, the first job takes 40 to 90 seconds longer to start.

Upscale your first print file free

Keep reading

Image upscaling API with no sign-up

Airpic's image upscaling API needs no sign-up, API key or card: a credit code is your bearer token. What you can call, how you pay per download, the limits.

Enhancing a photo with an AI chatbot: why it changes your picture

Asking an AI chatbot to enhance a photo returns a small, redrawn image with new faces and text. See how restoration differs and when to use each.

Upscale an image without Photoshop

Upscale an image without Photoshop, remove noise, erase objects or cut out the background in your browser. No account or install, and when an editor wins.