Technique inspired by Masashi Wakui post

A couple of days ago my brother in law (using adobe products) and me tried to emulate the Masashi Wakui look following the excellent analysis from the post here in the forum. We got relatively close, but couldn’t get things to glow and there was still less color in our pictures.
Shortly after that I had the idea to use the residual of a wavelet decomposition to color areas. The coloring part worked nicely as expected but using the wavelet decompose brought another surprise that I hadn’t expected, “flooding” lights using a higher decomposition layer. Tried it on a picture from Tokyo as I had been there a couple of years ago and, sorry if I sound arrogant, but I am totally stoked about the result. :blush:

Tried it on a daytime scene with colored lights and it also works nicely, though the effect is different:

Tried it on another country, and believe it or not, it also worked! :scream:

I am really happy as things turned out and there are more possibilities to this technique. :kissing_closed_eyes::stuck_out_tongue::smiling_imp:


Wonderful results! Nicely atmospheric shots, too.

So what did you do exactly? Color adjustments on residual?

Thanks Pat.

I basically used 8 layers of wavelet decompose. This gives a residual that is just a blotch of colors. Use just theresidual in “color” mode over your picture and see how it colors the areas. Then go back on the opacity until you like it. The surprise was using level 8 of the decomposed layers on the original picture. This gives you basically a huge soft halo and gives the picture a blurry, light flooded feel.
(Ok, there was some color toning in the beginning also, but just very basic)


How did you apply the Level 8 layer?


Seems like we may need to write this into a formal tutorial :slight_smile: awesome results!


I think he means he used the level 8 detail scales over his original image.

Is level 8 the same as the “residual image” layer?

@Biff, @paperdigits:
I used the Wavelet scale 8 (my coarsest) in grain merge mode on the picture. The residual was used in color mode also on the picture.

Be aware that the number of wavelet detail scales depends on the size of your image. Mine was 4735x2911 pixels. Important is the size of the scale relative to the picture size.

I would gladly write a tutorial about this. Should I include the teal toning I did in the shadows before applying the wavelet decompose tool, or just focus on the wavelet fun and why it works in these kinds of pictures?


Ah, got it, sorry it wasn’t completely clear to me.

How you author the tutorial is up to you; if you were to use the images from this post as examples, I’d include everything you did to get this look, else you’ll have a bunch of comments to the effect of “it doesn’t look the same!”

Should you decide to author, I’d be happy to proof read, test instructions, and/or ask a bunch of annoying questions to flush out all the details :slight_smile: or whatever other help you need.

1 Like

Ok, I’ll probably go with the Akihabara picture (the first). I have already prepared all the pictures. As soon as I have written something I’ll send it to you (can you read word files?). Thanks for the offer. :slight_smile:

1 Like

As long as the word file can be read by LibreOffice, yes. But the site articles are authored in Markdown, which is a pretty simple plain text format. Are you familiar with markdown? The bonus of markdown is that it can easily be versioned with git.

How is that different from a normal bloom filter? The residual layer is basically a blurred version of the image, and merging that with grain merge should add some light spill around highlights and darken the shadows.

@houz I haven’t looked into the bloom filter yet. But I do not use the residual in grain merge mode I use it in color mode. I use the coarsest scale in grain merge mode.

@paperdigits I’ll have a look at Markdown.

For those interested I have created a Flickr album with pictures processed with this technique:
w F i l e s

Very interesting.
I tried this technique. I found that using the “residual” layer in Color mode (Color HSV) actually reduces saturation in the image, deafening the original colors. Using one of the higher numbered scales in Grain Merge mode scales is working.


This is a cool technique. Here is my attempt using a night-time photo I took of the Galata Tower in Istanbul last October. I used the G’MIC plugin in GIMP to split the wavelets (Details > Split Details [Layers]) with “9” levels (8 details and the residuals). I then loaded these on top of the original image as layers.

I used “Dodge” merge mode to merge the residual colors onto the original image, setting the residual layer’s opacity to 35. I thought “Dodge” merge mode worked better than “Color” mode, as it really made the more colorful parts of the image brighter.

I then used the “Grain Merge” mode to merge the two coarsest levels of details onto that. In G’MIC it labels them coasest to finest (1 - n), so that was levels 1 and 2. I left the opacity of level 1 at 100%, but reduced level 2 to 50%.

Here is the original image:

Galata Tower, original

Here’s the version after doing this procedure:

Galata Tower, 'Bloomed'

Pretty cool!

1 Like

Yes that may happen. Basically you are blurring colors, so if for instance you have a colored wall and a bright white light in front of it, the residual will be a mixture of the color and the white. So, naturally it will be paler, more desaturated. I have also experienced it and I actually like the look on some pictures. But if you don’t like it, just pull the saturation back up afterwards. I have done this on some occasions.

Let me see what you come up with, I’m curious. :slight_smile:

Here’s one with out the grain merges with coarse detail wavelets, but just using “Dodge” blending of the residuals back with the original image. As before, the opacity of the residuals is 35%.

Galata Tower, just residuals

Nice picture, Isaac. It’s fun how the lights just go pop!

Only thing is, something happened to your shadows, they are very noisy in the after picture.

Ok, here the noise is not that strong. That means that the coarse wavelet layers do that, which seems strange to me as they shouldn’t amplify things on such a small scale??

BTW, does anybody know what size the levels in Gimp’s wavelet decompose are?
And what size in the G’mic Plugin?