Convert CR2 to JPG — Canon's RAW format, embedded previews, and full demosaic

The short version

1. Drag your .cr2 files onto the converter. Whole folders work for batch.
2. Pick the path: Fast extracts the camera's embedded JPG preview (~1 second per file, identical to what the camera LCD showed). Full demosaics the raw sensor data (~3–8 seconds per file, higher ceiling for editing).
3. Leave quality on 92 unless you know you need smaller. Q92 is visually indistinguishable from the source at roughly half the file size of Q100.
4. Click Convert. Single file → single JPG. Batch → ZIP with EXIF preserved by default.

If you're just getting JPGs out the door, stop here. The rest of this post is for when you want to know what CR2 actually contains, why "extract preview" is a legitimate shortcut, when you need the slow demosaic instead, and how to pick between Adobe Camera Raw, Lightroom, darktable, RawTherapee, and Capture One if you want more control than a converter gives you.

What's actually inside a CR2

CR2 (Canon Raw version 2) is a TIFF/EP-derived container that Canon used from the EOS-1D Mark II (2004) through roughly the 5D Mark IV (2016). The EOS M50 switched to CR3 in 2018, and every Canon mirrorless since the R5 is CR3. If you're reading this, your bodies are almost certainly in the CR2 era — 5D Mark II/III, 6D, 7D Mark II, 80D, 77D, Rebel T5/T6/T7, or similar.

A CR2 file contains three things in one envelope:

• Raw sensor data — 14-bit Bayer-pattern readings from every photosite, undemosaiced, uncolour-mapped, unsharpened. The real payload, usually 20–35 MB on a 20+ MP body.
• An embedded JPG preview — a full-resolution JPG the camera rendered on capture. Same pixels the camera LCD showed you; same JPG you'd get if you shot JPG+RAW. Usually 1.5–4 MB.
• EXIF + MakerNotes metadata — shooting mode, lens, focal length, shutter, ISO, white balance choice, Canon Picture Style, focus points used, and a pile of Canon-proprietary fields that only Canon's DPP reads fully.

The interesting part: the embedded JPG is already a perfectly usable deliverable. The camera baked it using its own colour science, its own noise reduction, its own sharpening, its own picture style. It's exactly what a JPG-only shooter would have ended up with. For social, web, quick client previews, or snapshot-quality archives, extracting that preview is a one-second operation that gives you identical output with no processing decisions required.

Extract preview vs full demosaic — the real decision

This is the split every other blog post skips. Two different operations, two different use cases:

Extract the embedded JPG (fast path)

• Time: ~1 second per file. No demosaicing, no colour conversion — we're literally copying bytes out of a TIFF tag into a JPG wrapper.
• Output: Full-resolution JPG, exactly as the camera rendered it. Canon's colour science, its default sharpening, its default noise reduction.
• When it wins: Wedding proofs, client gallery previews, social media, contact sheets, anything headed to the web without major editing.
• When it doesn't: If the camera nailed neither exposure nor white balance, the embedded preview is stuck with the camera's bad call. You can't recover highlights or shadows from an already-baked JPG the way you could from raw.

Full raw demosaic (slow path)

• Time: ~3–8 seconds per file depending on resolution and algorithm. The converter has to interpolate the Bayer pattern into full RGB, apply a colour matrix, white-balance-correct, and encode.
• Output: JPG rendered from 14-bit raw data. If you want to push exposure ±2 stops, shift white balance 500K, or recover highlights, you're doing it on the raw, not the preview.
• When it wins: The camera got exposure or WB wrong; you're pushing files hard in post; you want the cleanest-possible starting point; you're delivering JPGs as finals, not proofs.
• When it doesn't: The camera nailed it. The embedded preview is already identical to what a hands-off raw render would produce, and a fresh demosaic is extra work for zero gain.

Pro workflow is usually both — extract previews for the client proof gallery (fast), full-demosaic the selects for delivery (slow, higher quality). Our tool exposes both modes; most free online converters silently pick one and don't tell you.

Why shoot CR2 in the first place

If CR2 is so inconvenient, why not just shoot JPG? Three reasons every Canon shooter eventually learns:

• 14-bit depth vs 8-bit. CR2 stores 16,384 luminance levels per channel; JPG stores 256. That's the difference between recovering a blown-out sky and getting a white splotch. JPG is 14 EV stops crammed into a narrow container; raw keeps the full dynamic range available for editing.
• White balance is a decision, not a commitment. Shoot JPG with the wrong WB and you've baked orange into every pixel. Shoot CR2, and WB is metadata — you move a slider in post, the pixels re-render, no quality loss.
• Non-destructive everything. Exposure, shadows, highlights, saturation, sharpness, noise reduction — in raw they're all parameters applied at render time. In JPG they're cooked in. Edit the JPG twice and you've lossily re-encoded once; the CR2 can be re-rendered a thousand times without degradation.

The cost is file size (~25 MB per CR2 vs ~5 MB per equivalent JPG), card-throughput on high burst rates, and the post-processing step you're standing in right now. For working photographers, the trade is always worth it. For vacation snaps, it's often not — which is why most cameras offer JPG-only and RAW+JPG modes.

The demosaic algorithm matters (a little)

When you do pick the full demosaic path, the Bayer-pattern interpolation algorithm decides how sharp the output looks, how much moiré appears on fine fabric, and how long the conversion takes. Every raw processor picks a default:

• AHD (Adaptive Homogeneity-Directed) — darktable's, RawTherapee's, and dcraw's default. Good balance, moderate speed.
• AMaZE — RawTherapee's flagship. Slower but cleaner on detail-heavy scenes (fabric, foliage).
• DCB — RawTherapee option, resolves red-channel edges well.
• VNG (Variable Number of Gradients) — older, faster, softer. Default in ancient dcraw builds.
• Proprietary (Adobe, Canon DPP, Capture One) — nobody knows exactly what they're doing; they just look nice.

For 95% of deliveries — client JPG, stock upload, web — the algorithm difference is invisible at web resolution. At 100% crop on fine fabric, you'll see ~5% more visible detail with AMaZE vs VNG. Adobe Camera Raw's demosaic is quietly very good but slower than dcraw's default; our in-browser pipeline uses AHD, which is the best-per-second choice.

White balance — the silent wrecker

CR2 stores the white balance you set on the camera as metadata. A raw processor reads it and applies it at render time. If you set the camera to "Auto" and the sensor guessed wrong under mixed indoor/window light, the embedded preview and any fast-extract JPG will look yellow or blue — there's no recovery, that's baked.

Full demosaic gives you the escape hatch: open the CR2, click a grey card with the WB eyedropper (or nudge the Kelvin slider), and re-render. The pixels are clean; only the WB choice changes. If you shoot a grey card or ColorChecker at the start of each lighting scenario, this is a 30-second fix across a whole batch.

Our converter exposes a single Auto-WB toggle (on by default) and a Kelvin/tint override for the rest. For per-image colour tweaking, use Lightroom or Capture One; they're built for it.

How our tool compares (honestly)

CR2 conversion has a crowded field, with wildly different defaults. Here's what each tool actually does:

If you're a working photographer doing this weekly, Lightroom Classic is the honest answer. If you're technically inclined and want zero subscription, RawTherapee or darktable will do everything a pro tool does given an afternoon of setup. For quick client proofs, a single out-of-camera preview, or a batch you need in the next 60 seconds, our converter is the shortest path.

EXIF, lens correction, and Canon Picture Styles

CR2 files carry a pile of Canon-specific metadata in MakerNotes that most converters discard:

• Picture Style — Standard, Portrait, Landscape, Neutral, Faithful, Monochrome, plus up to three user slots. Only Canon DPP and Adobe (which ships Canon profile emulations) respect these.
• Lens-correction metadata — the lens model, used to apply distortion and chromatic aberration profiles on render. Lightroom and Capture One have libraries of Canon lens profiles; open-source tools rely on lensfun.
• Focus points used — visible in DPP and Photo Mechanic. Not important for JPG export but sometimes load-bearing for culling.
• Dust Delete Data — a recorded sensor-dust map you can apply in DPP to auto-heal dust spots.

For plain CR2 → JPG, standard EXIF (exposure, camera, lens, GPS if you had a GPS accessory) is what matters and our tool preserves it by default. Canon-specific MakerNotes are stripped unless you enable "Preserve all metadata" in advanced settings.

Batch workflow — 400 wedding files in one pass

Cull first, convert second

Converting 400 CR2s at full demosaic is ~30 minutes of wall time. Converting them as previews is ~7 minutes. Converting only the 120 keepers at full demosaic after a 20-minute cull is ~15 minutes total and you threw away the bad-focus frames. Use Photo Mechanic, FastRawViewer, or Lightroom's library module for the cull, then export the selects.

Naming convention

Keep CR2 filenames intact on conversion (our default). Canon'sIMG_0001.CR2 becomes IMG_0001.jpg, so the raw master and the delivered JPG share a name — critical if the client asks for a re-render six months later.

Folder structure

A pattern that scales: /clients/<name>/<date>/raw/for the CR2s, /clients/<name>/<date>/proofs/for preview-extract JPGs, /clients/<name>/<date>/finals/for full-demosaic edited JPGs. Don't delete the CR2s; storage is cheap and you will re-render at least once a year.

Cross-platform raw siblings

CR2 is Canon's old standard. If you also shoot Nikon or shoot with a newer mirrorless, you'll hit the siblings:

• NEF — Nikon's raw format. Same philosophy, different quirks (12-bit or 14-bit depth, compressed or uncompressed). Full guide: Convert NEF to JPG.
• CR3 — Canon's 2018+ raw container, used by every EOS R, EOS M50+, and newer. ISO-BMFF based, smaller, modern.
• DNG — Adobe's open-standard raw. Pixel, Leica, and Hasselblad shoot it natively. Many photographers convert all their CR2/NEF to DNG for archival. Full guide: Convert DNG to JPG.
• ARW — Sony. ORF — Olympus. RAF — Fujifilm. RW2 — Panasonic. Each is conceptually similar; the demosaic math is all the same Bayer interpolation.

Works well / doesn't work

Works well

• CR2s from every Canon body 2004–2018 (1D Mark II onward, up through 5D Mark IV)
• Fast preview-extract for proof galleries and client contact sheets
• Full demosaic for final JPG delivery at Q92 or Q95
• EXIF preservation for catalog-based downstream tools
• Batches up to ~200 CR2s per session on modern laptop hardware
• Auto white balance and Kelvin override on full demosaic path

Doesn't work (well) yet

• CR3 (Canon's 2018+ format) — on the roadmap, currently requires Lightroom or DPP.
• Canon Picture Style replication — we ship a neutral profile; Lightroom's Canon-matched profiles are closer to out-of-camera looks.
• Lens correction — distortion and CA profiles are not applied; open in Lightroom/Capture One if you need them.
• Sensor-dust-spot removal — DPP-only; we don't read the Dust Delete Data field.
• Dual Pixel RAW (5D Mark IV) — we decode the primary subframe; the secondary parallax frame is ignored.

Tips for the best result

• Pick preview-extract for speed, full demosaic for flexibility. Most client proofs don't need raw processing; final deliverables often do.
• Set quality to 92, not 100. Q100 is ~45% larger than Q92 with no visible improvement on photographs. Q95 is the middle ground.
• Preserve EXIF. Downstream catalog tools (Lightroom, Capture One, Photo Mechanic, even macOS Photos) read shooting info from EXIF; stripping it orphans the JPG.
• Keep the CR2s. Storage is $0.02/GB/month on S3. Re-rendering six months later with improved tools — Adobe AI Denoise, Capture One 25's new profiles — gives you meaningfully better output from the same capture.
• Cull before you convert. Three minutes per file on 400 shots is 20 hours. Thirty seconds per file on the 120 keepers is one hour. Photo Mechanic or FastRawViewer for cull, then batch-convert the selects.
• If your CR2 won't open anywhere, it's probably CR3. Check the file extension case-insensitively. CR3 needs a 2019+ raw processor or conversion through DNG.

Common questions

Can I convert CR2 without Photoshop or Lightroom?

Yes — that's the whole point of this page. Our converter accepts CR2 drops with no install and no subscription. Canon's own Digital Photo Professional (DPP) is also free and handles CR2 natively, if you'd rather a desktop tool. Open-source RawTherapee and darktable are free and more powerful.

Will I lose quality converting CR2 to JPG?

Yes, by definition. CR2 is 14-bit raw; JPG is 8-bit with lossy DCT compression. For preview-resolution delivery, the loss is invisible — JPG at Q92 from a CR2 is pixel-indistinguishable from the in-camera JPG. For pushed edits (±2 EV exposure, large WB shifts) do the editing on the raw and export JPG only at the end; don't edit the JPG.

Is "extract preview" real conversion?

It's extracting the JPG the camera already rendered at capture time. That's as legitimate as any conversion — it's just that Canon's in-body JPG engine made the rendering choices instead of your raw processor. For hands-off delivery ("give me what the camera saw"), it's the right answer. For editing latitude, demosaic the raw yourself.

How big will the JPG be compared to the CR2?

Typical 24 MP Canon body: ~25 MB CR2, ~7 MB JPG at Q92, ~3 MB JPG at Q85. Preview-extracted JPGs tend to land at Q90-ish (Canon's default) and come in around 5–6 MB. JPG is 3–5× smaller than CR2 for essentially all Canon bodies.

Can I batch a whole SD card?

Yes. Drop the folder onto the converter; we process CR2s sequentially in a web worker and hand back a ZIP. Browser memory is the practical cap — ~200 full-demosaic CR2s is comfortable; ~500 preview-extracts is fine. If the card has 1,500 frames, split the folder in two or use Lightroom's Export dialog.

Does CR2 have white balance baked in?

No. White balance is stored as metadata; a raw processor applies it at render time. If you set the camera to Auto and it guessed wrong, full demosaic lets you fix it losslessly — preview-extract is stuck with the camera's call because the embedded preview is already rendered.

My CR2 won't open — is the file corrupt?

Probably not. Most likely: (a) you have a CR3 file (2018+ Canon body) and the tool you're using only speaks CR2, or (b) your version of Photoshop / Camera Raw predates your camera body and the demosaic profile isn't in the version you have. Update Camera Raw, or run the file through Adobe's free DNG Converter as a compatibility shim.

Ready?

Open the converter →. Drop your CR2s, pick preview-extract for speed or full demosaic for headroom, leave quality on 92. Free, in your browser for metadata work, server-side for the demosaic pipeline, no watermark, no sign-up. For Nikon raw, jump to NEF to JPG; for Adobe's open-standard DNG, DNG to JPG; if your photos actually want web-ready smaller files, chain the output through our JPEG compressor. Comparing the pro raw tools? We've sat them side by side at Adobe's ecosystem and on the broader tool comparisons index.