Blog

DxO has now intro­duced the 5th ver­sion of its AI-based denois­ing tool Pur­eRAW. This fea­tures a fur­ther improved denois­ing AI called Deep­PRIME 3.

Pur­eRAW is, so to speak, the lit­tle sis­ter of DxO’s big RAW image proces­sor DxO Pho­to­lab. Pur­eRAW con­cen­trates ful­ly on AI-based denois­ing and the cor­rec­tion of lens dis­tor­tion. All oth­er cor­rec­tions, such as expo­sure, col­or or con­trast cor­rec­tions or crop­ping, must be car­ried out in oth­er pro­grams. Pur­eRAW 5 can be used either as a stand-alone pro­gram or as a plug-in for Adobe Lightroom.

As a long­time Adobe Light­room Clas­sic user, I use DxO’s Deep­PRIME denois­ing exclu­sive­ly via the inte­grat­ed Light­room plu­g­in, which is also includ­ed in DxO Pur­eRAW 5*. The fol­low­ing there­fore focus­es on this approach.

What’s new?

Accord­ing to DxO, Deep­PRIME ver­sion 3 includes a third com­po­nent com­pared to its pre­de­ces­sor: the simul­ta­ne­ous cor­rec­tion of the remain­ing lat­er­al and lon­gi­tu­di­nal chro­mat­ic aber­ra­tions at sub-pix­el lev­el. The AI mod­el was also trained with a ten times larg­er image data set than the pre­vi­ous mod­el. The list of sup­port­ed camera/lens com­bi­na­tions has also been extend­ed. Of par­tic­u­lar rel­e­vance to me is the sup­port of the dual cam­eras of the DJI Air 3s drone.

The new ver­sion also includes the option of AI-con­trolled denois­ing of images from Fuji cam­eras with X-Trans sen­sors as a “tech­nol­o­gy pre­view”. How­ev­er, this will ini­tial­ly only work with old­er X-Trans sen­sors. The cur­rent 40 megapix­el sen­sors are not yet sup­port­ed. How­ev­er, I can’t pro­vide more details here as I have no per­son­al expe­ri­ence with Fuji X-Trans.

As I already wrote in my review of DxO Pho­to­Lab 8 with Deep­PRIME XD2s, I now often use the quite good AI denois­ing already inte­grat­ed in Light­room Clas­sic. For “dif­fi­cult” cas­es and espe­cial­ly for the shots of my DJI drones, how­ev­er, the denois­ing in DxO Deep­PRIME is still absolute­ly unbeat­able in my opin­ion. I was curi­ous to see whether this could be improved even fur­ther in the new Deep­PRIME 3 version.

Installation

There is actu­al­ly not much to say about the instal­la­tion. After down­load­ing the instal­la­tion file from the DxO serv­er, you start it and, after accept­ing the oblig­a­tory legal terms and con­di­tions and select­ing the instal­la­tion direc­to­ry, it installs both the stand­alone Pur­eRAW 5 pro­gram (which I won’t go into any fur­ther here) and the plug-in for Adobe Light­room, which is the sub­ject of this article.

After installing DxO Pur­eRAW 5, the next time you start Light­room Clas­sic, it will noti­fy you that a new plu­g­in is available.

If you take a clos­er look in the add-on mod­ule man­ag­er of Light­room Clas­sic, there are even two plug-ins installed:

The first trans­fers the file select­ed in Light­room to DxO Pur­eRAW 5. The sec­ond then inte­grates the file edit­ed in Pur­eRAW back into Lightroom.

My workflow

I have been edit­ing and man­ag­ing my images in Adobe Light­room Clas­sic for years, as I have already described in detail here and here. I use sev­er­al steps to do this.

After a shoot, I first briefly view all the images in the library mod­ule and imme­di­ate­ly delete all obvi­ous­ly unus­able images (blurred, gross­ly under­ex­posed, out of focus, etc.).

I then usu­al­ly go through the remain­ing images again in chrono­log­i­cal order in a sec­ond run in the devel­op­ment mod­ule. I make the first short edit­ing steps (crop­ping, rough expo­sure cor­rec­tions, white balance).

If there are sev­er­al sim­i­lar images in a series, I syn­chro­nize the edit­ing to all images in a series. Dur­ing this first edit, the images are marked. The images that I want to keep and edit fur­ther are marked with a flag (but­ton ‘P’ for ‘Pick’). Images that I def­i­nite­ly don’t want to edit are marked with the ‘X’ but­ton for lat­er dele­tion, the rest remain unmarked.

In the third step, I then only take a clos­er look at the marked images by select­ing only these. Usu­al­ly, I then only have around 15 to 30 per­cent of the orig­i­nal images left. These are then processed in more detail using all the options of the Light­room devel­op­ment mod­ule. I now usu­al­ly rate images that I real­ly like with one star.

Since the inte­gra­tion of AI-based denois­ing in the Devel­op mod­ule (Detail / Denoise…) in Light­room, I usu­al­ly edit slight­ly noisy images direct­ly with it:

For spe­cial images that still have a lot of noise or for images from my DJI drones, which the Light­room algo­rithm has not yet been able to han­dle very well, I now use DxO’s Deep­PRIME plugin.

Integration von DxO DeepPRIME 3 in Lightroom

The image in ques­tion is now trans­ferred from Light­room Clas­sic to the DxO plug-in for pro­cess­ing with Deep­PRIME. The trans­fer is done unin­tu­itive­ly via the menu item Plug-in extras in the File menu. Three options are now available:

With “Pre­view and Process”, the image to be processed is dis­played as a pre­view in Pur­eRAW 5 after trans­fer. The result of the denois­ing can be checked very pre­cise­ly in a before/after com­par­i­son via a cen­tral mov­able bar. You can zoom far into the image with the mouse wheel or direct­ly by enter­ing the mag­ni­fi­ca­tion scale and also move the frame with the mouse.

The “Process instant­ly” option omits the pre­view and only dis­plays the pro­cess­ing options. This allows a faster work­flow if you already know which set­tings you need.

If sev­er­al sim­i­lar images are to be denoised, the menu item “Process direct­ly using last set­tings” car­ries out the denois­ing imme­di­ate­ly with­out fur­ther prompting.

After start­ing Pur­eRAW 5 with the first two options, which takes a moment, you can first select the rec­og­nized cam­era and optics mod­ules if they have not been loaded before. In my exam­ple below, my EOS R5 with the EF 400mm f/2.8L IS II USM:

If you do not want to car­ry out the opti­cal cor­rec­tions in Pur­eRAW and instead want to do this in Light­room after­wards, you can sim­ply select “None of the above” to skip these mod­ules. After­wards, the options win­dow will appear. Depend­ing on the select­ed menu item with (see above) or with­out image pre­view as shown here:

The pref­er­ence set­tings are divid­ed into two selec­table areas.

Corrections

When the “Cor­rec­tions” area is clicked, a new win­dow appears on the right which allows you to select the cor­rec­tion parameters:

The new and improved Deep­PRIME 3 process is pre­set. The Lumi­nance set­ting influ­ences the strength of the noise reduc­tion. The image is sharp­ened with “Force details” and “Lens Sharp­ness Opti­miza­tion”.

This is best checked direct­ly in the pre­view win­dow if the image was trans­ferred to Pur­eRAW 5 with the “Pre­view and pro­cess­ing” option. The set­tings here are a mat­ter of taste and depend on the indi­vid­ual motif. I rec­om­mend sim­ply try­ing out a few settings.

When choos­ing the opti­cal cor­rec­tions, every­one has to decide for them­selves whether they want to do this in Pur­eRAW or lat­er in Lightroom.

Output

After click­ing on the low­er “Out­put” area or the sym­bol in the pre­view win­dow, the out­put for­mat of the edit­ed image can be selected.

As I still want to edit the images in Light­room, I always choose DNG as the out­put for­mat, as this pro­duces the high­est quality.

How­ev­er, as this is a lin­ear DNG for­mat, you have to accept that the result­ing file is a good three times as large as the orig­i­nal RAW file from the camera.

This is because the lin­ear DNG for­mat con­tains the com­put­ed col­or val­ues red, green and blue (i.e. three val­ues) for each image pix­el, while the cam­er­a’s orig­i­nal RAW for­mat only stores one col­or val­ue that cor­re­sponds to the Bay­er col­or fil­ter in front of that par­tic­u­lar pix­el (i.e. red or green or blue).

The image can be writ­ten to the same direc­to­ry as the orig­i­nal image. This is the default option that I use.

To dis­tin­guish it from the orig­i­nal, the suf­fix “DxO_DeepPRIME 3” is added to the file name by default. I have left it like this, but the renam­ing can be exten­sive­ly cus­tomized to your own taste using the “Edit” button.

If desired, the edit­ed image can also be saved in a sub­fold­er of the orig­i­nal fold­er or in a com­plete­ly sep­a­rate directory.

Final­ly, you can spec­i­fy where the edit­ed file should be export­ed to. This is only nec­es­sary if DxO Pur­eRAW is used as a stand-alone pro­gram. If you use the Light­room plug-in as described here, the edit­ed file is auto­mat­i­cal­ly trans­ferred back to Light­room as spec­i­fied above. If desired, it can also be added to a Light­room collection.

And off we go…

After select­ing the options, the actu­al pro­cess­ing can now be start­ed. Depend­ing on the hard­ware installed in your com­put­er, more or less patience is required here.

The graph­ics card installed in the com­put­er is of par­tic­u­lar impor­tance. On my lap­top with Intel Core i9-11980HK and an Nvidia RTX 3080, pro­cess­ing a 45 megapix­el file from the Canon EOS R5 with the RTX 3080 only takes around 10 sec­onds. How­ev­er, with the UHD Graph­ics built into the CPU, Pur­eRAW 5 needs approx. 100 sec­onds for the same image. With­out the sup­port of the graph­ics card, the 8 cores of the i9-11980HK already need 2 min­utes. On old­er hard­ware, this can also take sev­er­al min­utes. Over­all, Deep­PRIME 3 is for­tu­nate­ly sig­nif­i­cant­ly faster than the AI denois­ing inte­grat­ed in Light­room and also than the pre­vi­ous Deep­PRIME versions.

After edit­ing, DxO Pur­eRAW 5 clos­es auto­mat­i­cal­ly and Light­room Clas­sic returns to the fore­ground. Re-import­ing the edit­ed file takes a while and it is import­ed into a new­ly cre­at­ed Col­lec­tion in the ‘DxO Pur­eRAW 5′ fold­er. This is named with the date and time of editing.

Results

So far - so good. We have now suc­cess­ful­ly used Pur­eRAW 5. But what improve­ment does it bring com­pared to con­ven­tion­al noise reduc­tion in Adobe Lightroom?

A lot. Here is an example:

We vis­it­ed the famous Fushi­mi Inari-Taisha shrine in Kyō­to on a dark and rainy day in Japan. This world-famous shrine con­sists of thou­sands of orange-red torii stand­ing in a row.

With­out a tri­pod and with the nec­es­sary aper­ture (f/13) due to the required depth of field, I had to take the pho­to at ISO 12,800 despite a rel­a­tive­ly long expo­sure time of 1/25 second.

On the left you can see the image processed with Light­room, on the right the image denoised with DxO Pur­eRAW 5. Admit­ted­ly, the dif­fer­ence is not very big when scaled down. But at 300% mag­ni­fi­ca­tion you can see what Deep­PRIME XD can get out of the noisy original:

I still find the result very impressive.

DeepPRIME 3 vs Lightroom AI denoising

But Light­room also offers the option of denois­ing with its own AI sup­port. How does it com­pare with DxO’s Deep­PRIME 3? Here is the exam­ple above in comparison:

On the left you can see the result of the Light­room AI denois­ing (lev­el 80), on the right the result of DxO. The dif­fer­ence is much small­er here. Which you like bet­ter is a mat­ter of taste. Light­room shows less detail, DxO more arti­facts. How­ev­er, both can cer­tain­ly be changed by adjust­ing the options. What you like bet­ter is cer­tain­ly a mat­ter of taste, per­son­al­ly I like the result with Deep­PRIME 3 a lit­tle bet­ter. But I would like to inves­ti­gate this in more detail.

Is Deep­PRIME 3 even bet­ter than the AI denois­ing in Light­room Classic?

So it’s time once again for a com­par­i­son test.

Comparative test

Once again, I used the noisy RAW files from my pre­vi­ous com­par­i­son tests as test sub­jects. The first test object was again the shot of our cat Tom tak­en at ISO 12,800 with my Canon EOS R5.

First of all, here is the orig­i­nal image again, which has not been denoised. As it was still too dark despite ISO 12,800, I bright­ened it up by anoth­er f-stop in Light­room. It actu­al­ly cor­re­sponds to a shot with ISO 25,600!

I’m sure we all agree that the pic­ture is so unus­able. So get rid of it?

The AI can do it!

Since the avail­abil­i­ty of AI-based denois­ing meth­ods, I no longer have any prob­lems work­ing with 5-dig­it ISO val­ues. How­ev­er, the results that can be achieved in this way still amaze me. Here is an exam­ple in direct com­par­i­son. On the left is the orig­i­nal and on the right the ver­sion processed with DxO’s Deep­PRIME 3 (crop at 200% magnification):

That’s real­ly very impres­sive, isn’t it?

But now to the inter­est­ing com­par­i­son with the AI denois­ing inte­grat­ed in Light­room Clas­sic since ver­sion 12.3. In Deep­Prime 3, I set the lumi­nance val­ue to 75 and the lens sharp­ness opti­miza­tion to Stan­dard / 100:

In Light­room Clas­sic, I have also set the AI denois­ing strength to 75. Both val­ues have worked well for me so far. Then I start­ed the denois­ing with the “Process imme­di­ate­ly” option:

In Light­room, I then sim­ply set the sharp­en­ing to the default val­ue of 40 for all images:

The oth­er set­tings in Light­room were not changed. Here are the images in overview (this is already a por­trait crop from a land­scape image from my Canon EOS R5).

In the image scaled down to 2048 pix­els, there are no sig­nif­i­cant dif­fer­ences between Deep­PRIME and the Light­room AI. Even the orig­i­nal, which has not been denoised, looks scaled down quite acceptable.

In prin­ci­ple, down­scal­ing in itself already rep­re­sents effi­cient noise reduc­tion. In order to be able to eval­u­ate the dif­fer­ences that still exist, you have to take a clos­er look. In the fol­low­ing I there­fore show crops of the above images enlarged to 200%:

When enlarged, you can now clear­ly see the suc­cess­ful noise reduc­tion with both Deep­PRIME 3 and Light­room. The direct com­par­i­son of the two AI denois­ing process­es is now interesting:

There are hard­ly any dif­fer­ences in the orig­i­nal res­o­lu­tion. That’s why I’m show­ing anoth­er crop here at approx. 400% magnification:

At this mas­sive mag­ni­fi­ca­tion, Deep­PRIME actu­al­ly shows even more detail, the image appears sharp­er, but there is also a lit­tle more noise. All in all, I find it a lit­tle more appeal­ing and nat­ur­al, but giv­en the min­i­mal dif­fer­ences this is cer­tain­ly a mat­ter of taste. Whether this jus­ti­fies the addi­tion­al invest­ment in Pur­eRAW 5 if you already have an Adobe Light­room sub­scrip­tion is up to you to decide.

What about DJI drones?

Unlike con­ven­tion­al sys­tem cam­eras from e.g. Canon, Nikon or Sony, the cur­rent DJI drones use a quad-bay­er design for their high-res­o­lu­tion sen­sors. This means that the col­or fil­ters are arranged in a much coars­er grid than in the usu­al clas­sic Bay­er grid, which makes it more dif­fi­cult to assign col­ors to the indi­vid­ual image pix­els. I have already writ­ten more about this in my review of the DJI Mini 3 Pro.

What was the problem?

When edit­ing the DNG files of my new­er DJI drones, I was very dis­ap­point­ed with the results of the AI denois­ing in Light­room. Light­room pro­duced ugly line arti­facts in the AI denoised images, which made them unus­able for me in this respect.

Here is an exam­ple from my Pho­to­Lab 8 test report: I took the pic­ture of the Are­nal vol­cano in Cos­ta Rica with my DJI Mini 4 Pro at ISO 400:

Here you can see the results of the AI denois­ing from Light­room and DxO’s Deep­PRIME. On the left you find the result from Light­room in 400% mag­ni­fi­ca­tion, on the right the result of the denois­ing with Deep­PRIME XD2s:

In my opin­ion, the denois­ing in Light­room was com­plete­ly unus­able for DJI drone images with the 48 megapix­el Quad Bay­er sen­sors. DxO deliv­ered much bet­ter results with Deep­PRIME XD2s, so up until now I’ve denoised all my DJI drone images with it.

And what about now?

To test the new Deep­PRIME 3, I have now denoised the above image again with both the cur­rent Light­room Clas­sic (V 14.3) and with DxO Deep­PRIME 3. And the result sur­prised me twice: on the left you can see the results from Light­room at 400% mag­ni­fi­ca­tion, on the right the results of the denois­ing with Deep­PRIME 3:

For­tu­nate­ly, the result of the denois­ing is much bet­ter with the cur­rent Light­room Clas­sic ver­sion 14.3 as well as with Deep­PRIME 3. The pre­vi­ous­ly annoy­ing vis­i­ble arti­facts in Light­room are gone. Nev­er­the­less, the result with DxO Deep­PRIME is still much bet­ter, which can be seen par­tic­u­lar­ly well in the details on the slope of the vol­cano and in the foliage of the trees. Here is a direct com­par­i­son of the two images:

Espe­cial­ly on the slopes of the vol­cano you can clear­ly see the dif­fer­ences. In Light­room, these appear com­plete­ly smooth and with­out detail, where­as DxO Deep­PRIME 3 still brings out the details of the veg­e­ta­tion there. I like the DxO result much bet­ter here.

Resumée

With the lat­est AI-sup­port­ed denois­ing func­tion Deep­PRIME 3 includ­ed in DxO Pur­eRAW 5, DxO has achieved a fur­ther increase in qual­i­ty. The pro­gram has also become faster - at least on my hard­ware. How­ev­er, the improve­ments since the pre­vi­ous ver­sions have been small­er and users of Adobe Light­room since ver­sion 12.3 already have an alter­na­tive AI-sup­port­ed denois­ing option that also deliv­ers very good results - at least with my Canon sys­tem cameras.

How­ev­er, the sit­u­a­tion is dif­fer­ent with some new­er DJI drones. Although the cur­rent Light­room ver­sion 14.3 now also copes bet­ter with the Quad Bay­er sen­sors of the new drones, Deep­PRIME denois­ing is still clear­ly superior.

I will there­fore prob­a­bly con­tin­ue to pro­ceed as before:

Depend­ing on the sub­ject, my very noisy shots are processed with both Light­room and Deep­PRIME 3, with Light­room being more stream­lined and quick­er to use. Deep­PRIME 3 is still reserved for the “hard cas­es” and all my DJI drone shots.

How­ev­er, if you do not use Light­room, DxO Pur­eRAW is cer­tain­ly a good choice for denois­ing high-ISO images. How­ev­er, an addi­tion­al pro­gram such as Cap­ture One or the full ver­sion of DxO Pho­to­Lab is then required for fur­ther pro­cess­ing of the images.

My recommendation:

• If you don’t yet have any expe­ri­ence with Deep­PRIME 3, I strong­ly rec­om­mend down­load­ing the free tri­al ver­sion of Rur­eRAW 5 (direct­ly here at DxO*) and test­ing it inten­sive­ly with your own images. DxO is extreme­ly gen­er­ous and allows you to test its pro­grams for 30 days with­out any restrictions.
  
• How­ev­er, if you have not yet built up a large image data­base and are still com­plete­ly open in this respect, I rec­om­mend that you also take a clos­er look at DxO Pho­to­Lab 8* as an alter­na­tive. Although Pho­to­Lab 8 does not yet sup­port the Deep­PRIME 3 process pre­sent­ed here, I believe that this will be added soon. Per­haps this solu­tion will give you every­thing you need in one program.

Prices and availability

DxO Pur­eRAW 5 is avail­able on the DxO web­site for down­load* at the fol­low­ing prices:

• DxO Pur­eRAW 5: 119,99 €
• DxO Pur­eRAW Upgrade from Ver­sion 3 oder 4: 79,99 €

*= Affil­i­ateLink