Enlarge drawn artwork up to 16x in the browser with deep convolutional neural networks, choosing an art or photo model and one of five noise levels per job.
Used this tool? Rate it
Used this tool? Rate it
Bigjpg is a browser-based image enlarger that rebuilds a picture at a higher resolution using deep convolutional neural networks instead of stretching existing pixels, reducing noise and jagged edges in the same pass. It does one job, and the whole product is organised around the four settings that job needs.
The model is tuned for drawn artwork rather than photography. The vendor states that anime images and illustrations come out with colours, details and edges nearly perfectly preserved, and describes regular photographs only as supported — a distinction worth taking at face value: two 2026 comparison articles describe the same split, though neither publisher is neutral.
Everything else about the product follows from that focus: a 16x ceiling that only makes sense on clean line art, a noise control with five discrete levels, and a plan structure that meters images per month rather than compute time. The vendor frames the difference against conventional resizing in a general image editor, arguing that a network trained on lines and colour avoids the blur, visible noise, glitter and doubling that ordinary enlargement produces — a claim about what the model preserves, not about resolution recovered from nothing.
art or photo, mapping to cartoon illustration and photograph respectively. The choice selects a differently trained model, not a post-processing filter.-1 through 3, mapping to none, low, medium, high and highest. This is independent of the scale factor, so compression artefacts can be addressed separately from size.1 through 4, mapping to 2x, 4x, 8x and 16x. Free accounts stop at 4x; the 8x and 16x steps require a paid plan.X-API-KEY header.The 16x figure is a dimension multiplier, not a quality promise, and its usefulness depends entirely on source content. One comparison article — a site that discloses earning commission on page links — reports the photo mode producing decent results at 2x and 4x but hallucinating unrealistic texture beyond 8x, while rating the anime mode competitive with waifu2x on clean illustrations. Treating 16x as headroom for line art is the cautious reading.
The same logic applies to the noise control. Because it is exposed as five discrete levels rather than an automatic decision, it is a judgement about the source rather than a quality dial: a clean export needs none, and a heavily re-compressed file needs a high setting applied before the size increases, since enlarging first would magnify the artefacts the model is meant to remove.
art for illustration, manga panels and flat-colour digital work; photo for photographic material, accepting the weaker result that the vendor's own wording and the published comparison articles both describe.The tool fits work where the source is drawn rather than captured and where no software can be installed. Preparing a manga panel or an illustration for print at a size the original file cannot reach is the central case, since flat colour and hard linework are exactly what the model was trained on. Rescuing a small archived illustration — a web-sized image with no surviving master file — is a close second, and the separate noise control matters there because such files usually carry compression damage as well as low resolution.
A third case is programmatic batch work on a paid plan. Because the API takes a source URL and the query endpoint accepts multiple task ids at once, a catalogue of illustration assets already hosted somewhere can be enlarged in bulk without building an upload pipeline first. The retry endpoint accepts the same comma-separated form, which matters because the vendor acknowledges that individual jobs can fail under load — a batch run needs a way to re-drive the failures without resubmitting the whole set.
It is a poor fit for photographic retouching at scale: the vendor calls photographs only supported while calling illustration output nearly perfect, and a competing service's comparison article reports skin texture coming out flat and painted, rating text handling poor. It is also unsuitable for confidential material that must not leave the machine — everything is processed on the vendor's servers.
The primary surface is the website, which works without an account for basic use. Signing in unlocks server-side processing, history, the server-region setting and, on a paid plan, the API.
The vendor states that native applications are published for iOS and Android through the App Store and Google Play, with payment on mobile running through Apple Pay and Google Wallet.
Processing location is a configurable property rather than a fixed one. The default server sits in mainland China, and a signed-in user can switch to a United States server from the settings page to improve download speed — worth setting deliberately, since it affects both latency and where the upload is processed.
The API is a further surface rather than a separate product. It runs against the same service on the same account, so quota, ratio limits and content behaviour are shared with the web interface; what changes is only how jobs are submitted and collected.
There is a usable free tier and three paid tiers, and the paid tiers are one-off purchases covering a fixed number of months rather than recurring subscriptions.
| Plan | Price | Duration | Images per month | Upload ceiling | Max ratio | Queue | Batch mode |
|---|---|---|---|---|---|---|---|
| Free | — | — | 20 | 5 MB | 4x | Slow, shared server | No |
| Basic | 6 USD | 2 months | 500 | 50 MB | 16x | Top priority | Yes |
| Standard | 12 USD | 6 months | 1000 | 50 MB | 16x | Top priority | Yes |
| Premium | 22 USD | 12 months | 2000 | 50 MB | 16x | Top priority | Yes |
The paid tiers differ only in duration and monthly quota. Upload ceiling, maximum ratio, queue priority, high-performance server, offline enlarging, parallel enlarging and batch mode are identical across all three, so the choice is purely a question of how many images per month you need and for how long.
Each purchased plan carries its own quota cycle with its own reset time, and the account page shows current-period quota, remaining quota and the next reset moment per plan in your local time zone. That matters if plans overlap, because the cycles do not merge into one pooled balance.
Web payment runs through Stripe, WeChat Pay and Alipay. The vendor additionally states it accepts BTC, ETH and USDT, but by email arrangement rather than as a checkout option, so cryptocurrency is a supported route rather than a self-service one. Mobile purchases go through Apple Pay and Google Wallet inside the apps.
Two 2026 comparison articles place the product against waifu2x, Upscayl and Topaz Gigapixel. Both credit it with the highest scale factor of any browser tool they list and an anime mode competitive with waifu2x, and both conclude photographic sources are better served elsewhere — one rating its photo texture only fair and text handling poor, the other reporting real photographs come out flat and painted. Neither is disinterested: one discloses commission on page links, the other is published by a competing service that ranks itself highly in the same article.
The split those articles describe is by content type rather than price. For drawn material in a browser they treat it as a first choice; for photographs, or where files must stay local, they point to installed alternatives. Given who publishes them, treat that as a hypothesis to check on your own files rather than a settled result — though it matches the vendor's own wording.
Both are accepted and there is a dedicated photo style, but the vendor itself describes anime and illustration as nearly perfectly processed while calling photographs merely supported. Two third-party comparison articles call photographic output flat, painted and weaker beyond 4x — one carries an affiliate disclosure, the other is published by a competing service.
It is a dimension multiplier reached through discrete steps of 2x, 4x, 8x and 16x, not a continuous slider. Free accounts are limited to 4x. One comparison article reports the photo mode beginning to invent unrealistic texture past 8x, so the top steps are most defensible on clean line art.
Not to try it, but signing out costs you real functionality. Offline processing, history and the server-region setting all require being signed in, and without an account you must keep the browser open or lose the result.
They are deleted automatically within 24 hours on free accounts and within 72 hours on premium accounts. Image links are encrypted, and the vendor states that unless you share a link, nobody else can download your source or result.
Yes, with documented submit, query and retry endpoints, an X-API-KEY header and parameters for style, noise and scale. It takes the source image as a URL rather than an upload, and keys are available to paying members only.
Only duration and monthly quota: 2 months and 500 images, 6 months and 1000, or 12 months and 2000. Upload ceiling, maximum ratio, queue priority, server class, offline and parallel enlarging and batch mode are the same on all three.
On the vendor's servers, with mainland China as the default location. A signed-in user can switch to a United States server in the settings page to improve download speed, so the region is a choice rather than a fixed property.
The vendor states that failures can occur depending on your network and how many users are online, and asks you to try again. There is no documented automatic retry on the web flow, though the API exposes a retry endpoint that accepts several task ids at once.