This might be the easiest way to improve the image quality of your Nikon D3500’s RAW files. And if you’re on the fence about “upgrading” to a fancy new mirrorless model, this might buy you more time. It also changes the math if you’re thinking about picking one up used: Nikon D3500 bodies go for very little now, and this is a cheap way to close a lot of the gap to a newer camera in terms of image quality.
It doesn’t involve hacking your camera or sending it in for repairs. Instead, it’s about running the RAW files through one of the new AI-based processing apps. In these examples, I’m using DxO PureRAW, which has become my app of choice for this type of thing in my own workflow.
I’ve run some of my D3500 photos through PureRAW, and you can see the results below, with before/after sliders and 100% crops. (And if you’d like an even closer look, you can also download the RAW/DNG files; I have more information on how you can get those at the bottom of the page.) It’s the closest thing to a sensor upgrade you can get without buying a new camera.
Tools like DxO PureRAW and Lightroom’s newer generation of denoise tools work on the sensor data before it has been demosaiced and baked into the pixels, and that’s where these newer machine-learning models have the advantage over the older generation. You can still run the older style of noise reduction on JPEGs and TIFFs (using other apps, not PureRAW), but you’re working with already-cooked data, and the results are a different league. So this is the payoff for deciding to keep all those RAW files from years ago.
You’ll see the most dramatic improvements in high-ISO frames and in shadow detail. That’s where older RAW files degrade most visibly. In many of today’s cameras, ISO 3200 looks fantastic. But in many older digital cameras, ISO 3200 was the bleeding edge and came with pretty heavy noise and graining.
But the improvements aren’t only limited to high ISOs. Even at the camera’s base ISO you can see useful improvements, albeit usually more subtle.
The Nikon D3500: Sensor and ISO range
A quick look at what we’re working with here.
The D3500 has a 24.2-megapixel APS-C CMOS sensor measuring 23.5 × 15.6 mm, recording 6000 × 4000-pixel images; that works out to a pixel pitch of about 3.9 µm. Its base ISO is 100, and the standard sensitivity range runs from ISO 100 to 25600, with no separate extended setting. It came out in 2018, so there are plenty of RAW files from it sitting in archives by now, including my own.
One thing worth knowing before you start: the D3500 records 12-bit compressed NEF files, and that’s the only RAW option it has. There’s no 14-bit setting, and no lossless or uncompressed choice buried in the menu. That’s less headroom than a higher-end Nikon gives you, and it’s part of why the shadows are where these files run out of road first — and where the denoising has the most to work with.
The examples below run from ISO 25600 down to ISO 100, highest first, since that’s where the differences are most obvious.
| Sensor | 24.2 MP APS-C CMOS 23.5 × 15.6 mm |
| Resolution | 6000 × 4000 pixels |
| Pixel pitch | ≈ 3.9 µm |
| Base ISO | 100 |
| Native ISO range | 100 – 25600 |
| Extended ISO (low) | none |
| Extended ISO (high) | none |
| ISO in these examples | 25600 – 100 |
And a reminder that PureRAW and its competitors only work on RAW files.
Before and after: ISO 25600 down to ISO 100
Shot 1: ISO 25600
Shot 2: ISO 4500
Shot 3: ISO 3200
Shot 4: ISO 720
Shot 5: ISO 320
Shot 6: ISO 100
Verdict
For an entry-level model, the Nikon D3500 doesn’t do too badly with higher ISOs on its own, but you can see in these examples that the denoising has a pretty dramatic effect, making a much larger range of the D3500’s ISO range usable in a very real sense.
As usual, the effect at the base ISO 100 is subtle, and it’s probably not worth doing if all you shoot is in the ISO 100 to 400 range.
But if you’re shooting at ISOs above that — and aren’t most of us doing that at least some of the time? — running the RAW files through a denoising app like PureRAW can make a big difference and salvage image files that you might otherwise reject. So it’s well worth a try on your own images (there’s a free trial version you can download).
The amount of difference that the distortion and vignetting functions have depends on the lens you’re using. In these examples, with the kit D3500 lens, the effect is pretty subtle. It’ll be much more pronounced with some other lenses. And that, after all, is a real strength of DxO’s suite of apps, because they’re built on what brought DxO to prominence: their extensive database of camera and lens profiles, with the results tailored for each specific camera and lens combination.
You can try it on your own images
I’ve found PureRAW to be very useful in my own workflow and have been a paying customer since version 3 in 2023.
But it’s all well and good to see how it works on a small sample of my images; the real test is what effect it’ll have on your images.
If you want to try it on your own files, DxO offers a free trial of PureRAW. Or, if you want a full RAW editor alternative to Lightroom, the same DeepPRIME engine is built into DxO PhotoLab.
How these examples were made
- Camera: Nikon D3500 (firmware 1.00b)
- Input image format: Nikon NEF
- Lenses in this batch: AF-P DX Nikkor 18-55mm f/3.5-5.6G
- ISO range: 100 – 25600
- DxO PureRAW (v. 6.5.1) — DeepPRIME XD3 (Luminance 40, Force Details 0) · Lens softness: ON · Chromatic aberration, vignetting, distortion: ON · Dust removal: OFF · DNG output (uncompressed)
- Rendering: Adobe Lightroom Classic 15.5, identical develop settings for each before/after pair, with the lens corrections handled by each side’s own engine: Lightroom’s lens profile (distortion + vignetting at default) and chromatic-aberration removal on the Lightroom side, DxO’s on the PureRAW DNG (Lightroom’s lens corrections switched off there so nothing is corrected twice); Lightroom’s sharpening and noise reduction switched off on the DNG (PureRAW already did those); no output sharpening
- Lightroom lens profile: Adobe (Nikon AF-P DX NIKKOR 18-55mm f/3.5-5.6G VR) — used on the Lightroom side of every comparison (it lifts the corners by roughly 31 – 298% at these apertures)
These are all photos I shot myself, in ordinary shooting conditions, not test-chart frames.
Every before/after pair was rendered from the same file with the same settings; the only variable is whether the file went through PureRAW first.
PureRAW lets you dial each tool up or down. For the purposes here I’m using my standardized settings that amount to sensible defaults rather than chasing extremes: DeepPRIME XD3 noise reduction at its default strength, lens softness, chromatic aberration, vignetting and distortion corrections on, and dust removal off.
I’ve used the same settings for every frame, so the comparisons are apples-to-apples. And to keep it fair on the optics, the Lightroom side has Lightroom’s own lens profile switched on (distortion and vignetting corrections at their defaults) and chromatic aberration removal on in every comparison, so each side has its optics corrected by its own tools. I’ve also tried to include a variety of different ISO settings and scenes.
Why I’ve chosen these settings here
PureRAW does more than noise reduction. Its tools fall into two groups.
The sensor corrections — DeepPRIME’s noise reduction and demosaicing — work on what the sensor recorded.
The optical corrections — lens softness, chromatic aberration, vignetting, and distortion — fix what the lens did to the light on its way in. The sensor corrections are the main event in this post, but I’ve left DxO’s optical corrections on as well. I can’t really think of a good reason not to fix chromatic aberration in every photo (we’re not doing lomography here!), and it’s a setting I always enable by default on my own photos. For my own photos, I am, however, more choosy about when to apply vignetting and distortion fixes, because to my own eye, those optical “flaws” are sometimes positives and something that I want to retain. But for the purposes here of using a standard set of settings to make comparison easier, and because these types of optical profiles and corrections are defining strengths of DxO’s software, I’m applying those fixes in these side-by-side examples.
Lens softness correction is such an integral part of how PureRAW’s output looks that turning it off would misrepresent the tool. And DxO’s optical corrections — built on the enormous body of lens and camera test data that was their original claim to fame — are a genuine strength that separates PureRAW from Lightroom’s built-in tools.
The images above
In the examples above, I’ve included two types of comparisons:
- My edit, upgraded: the before is my finished Lightroom edit, plus Lightroom’s own lens profile (distortion + vignetting at default) and chromatic-aberration removal, switched on for every comparison; the after is the same edit applied to the PureRAW DNG, with Lightroom’s sharpening and noise reduction — and its lens profile and chromatic-aberration removal — switched off on the DNG, because PureRAW has already done those jobs (leaving them on double-processes the file — the most common reason PureRAW output looks over-sharpened). So each side has its optics corrected by its own tools, and everything else is identical on both sides.
- Untouched: both files with no editing at all beyond those lens corrections, at Lightroom’s default settings, so you can see what PureRAW does on its own.
With distortion correction on both sides, each frame is shown as its engine renders it corrected — geometrically the same on both sides, so it doesn’t skew the comparison.
There’s a technical note worth knowing: at identical settings, an unadjusted PureRAW DNG renders about 0.1 stop darker than the original in Lightroom. That’s because PureRAW writes its own baseline exposure into the DNG (here PureRAW’s DNG carries +0.20 EV; the camera’s file has no such tag, and Lightroom applies its own hidden baseline for this camera). So rather than trust the tags, I measured the difference for each pair (central area, linear light) and offset the DNG’s exposure by exactly that amount (-0.06 to +0.13 EV) — you’re seeing PureRAW’s processing, not an exposure difference.
Colour shift: none worth mentioning. I also check whether PureRAW’s files come back with a different colour balance than the originals (some cameras do). Measured over the central 60% of each pair in linear light at identical Lightroom settings, the Nikon D3500 DNGs stayed within 7% on the red-to-green and blue-to-green ratios (my threshold for “significant” is 4%), so what you see in the sliders is processing, not a colour cast.
How PureRAW fits into a real workflow
I’m using these with Lightroom Classic, but PureRAW isn’t limited to only Lightroom-centric workflows. You can use it as a standalone app, and the underlying DeepPRIME technology is also built into DxO’s own full-suite editor app PhotoLab (which is itself well worth a look if you’re after a Lightroom alternative).
In practice, PureRAW is a pre-processing step, not an editor. You can run it as a standalone app (just drop RAW files on it, pick a preset or adjust the settings, and run process) or straight from Lightroom Classic via its plug-in — select the photos, send them to PureRAW, and it hands back the processed versions reimported into Lightroom.
Either way, the output is a new DNG file for each RAW: PureRAW never touches your original. If you’re using Lightroom, the DNG is created next to the original with your existing edits carried across, so you’re not starting over — you just switch to the new file. (TIP: make sure to save any Lightroom develop settings to the file before sending it to PureRAW or the settings won’t be applied to the PureRAW version automatically.)
With this workflow, there are two key considerations to factor in:
- File storage: those DNGs are linear (already demosaiced) files, and they can be considerably larger than the RAWs they came from — the ones in this post average about 3.5× the size of the originals. If you run it across a whole archive, that adds up quickly. So I don’t run it across everything.
- Processing time. DeepPRIME is computationally heavy. So most of the time I’ll only apply it to selected images rather than everything. Running it across dozens or hundreds of images is entirely possible, and I’ll sometimes do that if I have the time, but it takes a while (depending on the speed of your computer) and eats storage space.
There is a way to take the edge off that. PureRAW 6 added a High-Fidelity DNG compression option, which DxO says gives files up to about four times smaller than the uncompressed DNGs, with no loss of perceptual quality. The catch is that it only applies when you process an original RAW — it won’t shrink DNGs you’ve already exported. The files in this post were written uncompressed.
What PureRAW doesn’t do
It’s also worth being clear about what this isn’t, because the marketing can blur the lines:
- It doesn’t add resolution. The DNG has exactly the same pixel dimensions as the RAW. It cleans up and sharpens what’s there; it doesn’t upscale. (That’s a separate job for tools like Lightroom’s Super Resolution or Topaz Gigapixel.)
- It doesn’t fix blur. Camera shake, motion blur, and missed focus stay exactly as they were. The “lens softness” correction is a profile-based sharpening for the lens’s known optical softness, not a deblurring tool.
- It doesn’t recover what was never captured. Blown highlights are still blown, and a badly underexposed frame gets cleaner, not correctly exposed — that part is still done in your editor.
A note on the “AI” label
PureRAW and its rivals describe themselves as AI-powered. Which is the marketing buzzword of the day. But it’s worth being precise about what that means here, because “AI” can be a double-edged sword, and the kind of AI we’re talking about here is a different thing from the AI most people now have in mind.
This is more accurately referred to as machine learning. What that means is that DxO trained a model on a very large set of paired examples — noisy sensor data and what the clean version should look like — and the software uses that training to make a better-informed guess about what each pixel really was.
It works from the data in your file and nothing else. It doesn’t invent content, it doesn’t pull in imagery from anywhere else, and it can’t invent a detail out of nothing — where the noise has genuinely destroyed information, you get softness or waxy texture, not made-up content (a good reason to not push those sliders all the way to the right). Push it hard enough on a truly ruined frame and you get mush or waxy textures, not fabrication.
That’s different to how generative AI works. Think ChatGPT or Claude or SeaDance or Nano Banana. That’s a fundamentally different type of technology, one that synthesizes new pixels. And that isn’t what PureRAW uses and isn’t what’s happening in any of the comparisons above. Your photo is still your photo.
Download the RAW and DNG files
If you want to take an even closer look at the results here, I’ve posted the full-size original RAW files and PureRAW’s DNG output for the examples above for subscribers to my newsletter. The download index is here: RAW file downloads for subscribers. The page is password-protected; the password is in the newsletter’s welcome email.
