Initializing... drag & drop files here
Supports: RAF
A RAF is a Fujifilm raw capture: the unprocessed sensor readout, 14 bits per photosite on X-series bodies, and unopenable by anything that is not a raw developer. WebP is a delivery format for the web — small, widely supported in current browsers, and finished. Converting one to the other is a one-way trip from "still editable" to "ready to publish", and the only real decision on the way is whether you want the lossy or the lossless variant of WebP at the other end.
The short version: convert to WebP when the photo is going onto a web page or into an app, keep the .RAF as your master, and leave the "Lossless?" control on "No" unless you know why you are turning it on.
| Property | RAF (Fujifilm raw) | WebP |
|---|---|---|
| What it holds | One undeveloped sensor readout, mosaic form | A finished RGB image, optionally with alpha |
| Origin | Fujifilm; spec sheets call it "RAF original format" | Google; bitstream-compatible with VP8 |
| Bit depth | 14-bit on X-series; 14 or 16-bit on GFX | 8 bits per channel |
| Compression | Uncompressed, compressed or lossless-compressed in camera | Lossy or lossless, your choice |
| Maximum dimensions | Whatever the sensor is — 11648 x 8736 on a 102 MP GFX100 II | 16383 x 16383, because the bitstream uses 14 bits for each axis |
| Transparency | No — a sensor records opaque light values | Yes, in both lossy and lossless modes |
| Editing latitude | Full: exposure, white balance, highlight recovery still adjustable | None; the render is baked in |
| Browser support | None | Chrome 32+, Firefox 65+, Edge 18+, Opera 19+, Safari 16+, iOS Safari 14+ (caniuse) |
| Typical size | 40-80 MB for a 40 MP frame | A fraction of a megabyte to a few megabytes |
.RAF files onto the page or click "+ Add Files" to browse. Queue as many as you like; they all convert with the same settings.| Lossy WebP (the default) | Lossless WebP | |
|---|---|---|
| Best for | Photographs — which is what a RAF always is | Graphics, screenshots, flat colour, anything that must survive re-encoding untouched |
| Size on a photograph | Small; 25-34% under a comparable JPEG per Google's figures | Very large — lossless coding has nothing to exploit in sensor noise and fine detail |
| Size vs PNG | Not comparable | About 26% smaller than the same image as PNG, per Google |
| What Quality Preset does | Sets the real fidelity knob; the difference between a mid and a high preset is clearly visible | Effectively nothing to the picture, and very little to the size — see below |
| Transparency | Supported | Supported |
The Quality Preset behaves completely differently in the two modes, and this is the trap on this page. In lossy mode it is the quality factor the encoder documents, and it does exactly what you would expect. In lossless mode it is not a fidelity control at all: we measured the full span from Q1 to Q100 on one test image through this pipeline, and every output was pixel-identical to every other, with file size moving only from 6,200 bytes down to 5,912. The same span in lossy mode produced substantial, easily measurable differences. So if you have switched "Lossless?" to "Yes", stop adjusting the preset — it is not going to change your picture.
For a rendered Fujifilm photograph, lossy at the "Very High" preset is almost always the right answer. Lossless WebP on a 40 MP sensor image produces a file so large that the format's whole advantage disappears.
Almost certainly not. Lossless compression works by finding exact repetition, and a photograph from a 40 MP sensor is essentially free of it — every pixel differs slightly from its neighbours because of real detail and sensor noise. The result is a file that can be many times larger than a lossy WebP of the same image while looking identical at normal viewing sizes. Lossless earns its place on graphics, logos, screenshots and diagrams, and on images that will be re-encoded repeatedly. For a developed RAF headed to a web page, leave "Lossless?" on "No (Recommended)".
Because in lossless mode the quality value is not a fidelity setting. The underlying encoder documents Q as the quality factor for lossy encoding — in lossless mode it influences how hard the encoder works rather than what it keeps. We measured this directly: across Q1 to Q100 with lossless enabled, every output file on our test image was pixel-identical, and the total size difference over that entire range was under 5%. In lossy mode the same range changes the image substantially. Practical takeaway: choose the mode first, then only bother with the preset if you chose lossy.
Google's published figures are 25-34% smaller than a comparable JPEG at an equivalent SSIM quality index for lossy WebP, and 26% smaller than PNG for lossless WebP. Those are the numbers to plan around, not guarantees for a specific image — how much you actually save depends on the content, and a noisy high-ISO frame compresses worse than a clean studio shot in any format. For the RAF case the more useful comparison is against the source: a raw file that started at 40-80 MB commonly lands under a megabyte as a lossy WebP at a high preset.
Yes, comfortably. WebP is bitstream-compatible with VP8 and uses 14 bits for each dimension, which caps an image at 16383 x 16383 pixels. The largest raw frame Fujifilm currently produces — 11648 x 8736 from the 102 MP GFX100 II — is well inside that, as is the 7728 x 5152 frame from a 40.2 MP X-T5. You would need to stitch a panorama before the ceiling became a real concern, and if you ever did hit it, scaling down under "Image resolution" is the fix.
Transparency is not a question here: a camera sensor records opaque light values, so a RAF has no alpha channel and the WebP comes out fully opaque. WebP itself does support alpha in both lossy and lossless modes, which matters if you later composite the image, but nothing is lost in the conversion because there was nothing to lose. Metadata is a different story — nothing in this conversion asks the encoder to strip it, but Fujifilm's shooting data lives in maker-note structures that a generic renderer has no obligation to carry across a demosaic. Keep the .RAF if the shot data matters to you.
Two separate reasons, and neither is a fault in the file. Fujifilm's Film Simulations — Provia, Velvia, Astia, Classic Chrome, Acros and the rest — are instructions applied when a renderer develops the raw data, not pixel values stored inside the RAF, so a neutral development ignores them. Separately, most X-series bodies use the X-Trans colour filter array, a 6 x 6 non-Bayer pattern that every renderer demosaics with its own algorithm, so no two converters produce identical fine detail. If the camera look matters, apply it in a Fujifilm-aware developer, export a finished image, and convert that instead.
It lets you name a size — in bytes, kilobytes or megabytes — and have the converter work towards it rather than towards a quality preset. It is the practical choice when you have a hard budget, such as a content management system that rejects anything over a certain size, or a page-weight target. There is also an "Auto Scale" switch alongside it: with a big raw frame and a small target, reaching the size by quality alone would look poor, so allowing the converter to reduce the pixel dimensions as well usually gives a much better-looking result than forcing full resolution into too few bytes.
Yes, and permanently. A RAF holds what the sensor measured — 14 bits per photosite on X-series bodies, up to 16 on GFX — which is exactly why exposure, white balance and highlight recovery stay adjustable while it is raw. Rendering to WebP demosaics that data and writes 8 bits per channel with the current interpretation baked in. Keep the original .RAF as your master and treat the WebP as a delivery copy. Your file travels over an encrypted connection, is developed and encoded on our servers, and is deleted automatically after a few hours; it is never shared or made public. To shrink WebP files you already have, compress WebP works on existing files without a format change.