I have spent many hours adding colour to black and white photographs, originally by using specialist oil paints directly onto prints and now, using Photoshop. It is a passion of mine.
Hand colouring or colourisation as it is also referred, is very time consuming but well worth the effort when the addition of colour enhances an image. One might say, adding colour can bring a photo to life.
Whilst I will undoubtedly continue to use Photoshop for all of my restoration and retouching needs, I though I would try out some of the automated apps on offer for adding colour to black and white images to see how well they perform.
The image used for this comparison is a lobby card featuring Eva Bartok in the 1952 American Technicolor film “The Crimson Pirate”, from Warner Bros. I found this image at my local trash and treasure market. It required a small amount of restoration and was then ready to colour.
My fist colourisation attempt was performed using the Algorithmia API. You simply upload your image or point to the image URL and all the work is done for you. I uploaded a high resolution image however the finished image I downloaded once the process was complete was reduced in size by over 500% leaving me with a .png file that looks fine on my computer screen but at just 800 pixels on the longest edge is too small to consider printing.
My next colourisation attempt was made using using the My Heritage colorisation online tool. Their website claims that this tool is the“world’s best deep learning technology”. It works by simply uploading your image to the site and the process is very quick. The file may be downloaded as a jpeg and whilst there is some loss in overall image size, you are provided with a print quality file. This online app is available for free for a limited number of images. You then need to be a ‘My Heritage’ subscriber to continue to use the app.
The following image was colourised using Colorise SG. This free online tool is described on their modest web site as a “deep learning colouriser prototype specifically for old Singaporean photos. ” They also acknowledge that “The purpose of colourisation is to generate an image with colours that are plausible. It by no means guarantees that the colourised image is an accurate representation of the actual snapshot in time.” Of course, this is true with all methods of colourisation.
Next I used ColorSurprise AI Pixbim. You will need to download this app to use it. Initially it works as a trial for 7 days, after which you must purchase a licence should you wish to continue use. All outputted images from the trial app are heavily watermarked.
When using this app you have a choice of ‘low’ or ‘high’ settings. The slider does not allow any setting in between these two options. It appears that the low setting is a ‘quick’ process. The ‘high’ setting takes longer to process. See results below
These automated apps may be fine for adding a touch of colour but the overall results, with this particular image were not fantastic. They all seemed to cope quite well with the skin tones but struggled in areas where there was more contrast. Samples on the web pages for each app displayed much better results so I was disappointed with my image. Perhaps they will work better on other images? I will let you be the judge
In the meantime, I will be sticking with Adobe Photoshop. Of course this is not an automated app and it can be a very time consuming option but there really is no comparison for quality with full control over colour selection, opacity, saturation, blending modes and the ability to precisely paint colours and adjust as needed. Plus there is no loss of quality with the output image. If a job is worth doing, it’s worth doing well.
For this image, I used footage from the movie as a reference for colours in an effort to maintain authenticity, something that cannot be done using an automated tool.
There are other automated colourisation apps available for mobile devices from the app store or from the Google Play store. If you have tried any of these, I would love to hear your feedback.