Portrait Virus Mutations

[John Dalton Synthetik Software]

The results from Michael Wight's Portrait Virus project at this year's Siggraph 2010 Studio are up at portraitvirussiggraph2010.blogspot.com/ . The Portrait Virus is a long term project Michael has been carrying out over the years where he uses Studio Artist to generate sets of digital portraits.

Michael's Portrait Virus project inspired me to start my own Portrait Virus Mutations project. The idea is to delve into the notion of stack filtering and what you can do with it.  Stack filter involves working with a stack of images as an input to a process that generates a single output image. As opposed to a normal image processing operation that takes a single source image and renders an effected output based off of the single source image.  Think of the stack of images as being a folder of images if that makes conceptualizing things easier.

For now i'm restricting myself to using the sets of painted portraits from Michael's portrait virus project as the source imagery for my portrait virus mutations.  

Brian Eno makes a big point of utilizing the creative power inherent in setting up artistic situations with a limited set of options and then forcing yourself to live within those limited options. I think this can be an interesting study in personal creativity.  It's easy to get distracted by infinite choices these days in any digital art form or software program, certainly its the case with Studio Artist where there's really an infinite range of potential effects you could generate. Forcing yourself to live with limitations or restrictions in a project can open up new avenues of thought and perhaps allow exploration of techniques you may otherwise never have tried.

I also think there's real power in getting involved with projects that force you to generate and post a new art image every day. I always get excited when i see Studio Artist users taking on these kinds of daily post projects, and would urge you to consider you own variation on it if you've never tried it.

You are welcome to check out the Portrait Virus Mutations site at portraitvirusmutations.blogspot.com

[John Dalton Synthetik Software]

What i love about starting projects like this is that you never quite know where they are going to lead you. In the case of Portrait Virus Mutations, there's been several interesting things that have popped up.

One slightly embarrassing thing i discovered was some weird behavior associated with the inverse source warp feature in version 4. What we in the software industry refer to as a bug. So that is now corrected and we're pushing back the 4.03 update a little to incorporate those changes.

The thing about visual art and the underlying algorithmic processing involved in creating artistic imagery is that 'mistakes' might often look quite interesting. So to the casual eye you might think, hey that looks pretty cool, and never catch on to the fact that technically the processing is incorrect in some way. Now from the standpoint of aesthetics, i don't think you can say there is such a thing as an incorrect process. You might like the way something looks, or not. And someone else might have the exact opposite opinion. And even your opinion might change from day to day.

One of the main Studio Artist design decisions is that when there are various processing options that lead to different looking results we tend to give you access to all of those options rather than restricting your choices. It's a very different philosophy from a lot of other companies, who would focus on limiting your ability to adjust options. There are advantages and disadvantages to both approaches. They certainly lead to very different artistic output capabilities for the users of said programs.

Now a big part of the portrait virus mutation project involves creating a spatially normalized stack of imagery. Again, when i say stack think of a folder of images if that makes it easier to conceptualize. Part of the reason i like the stack terminology is that there is something inherent in the ordering of the images in the stack that is going to influence the outcome of many of the processing techniques applied to the stack, especially ones based on temporal processing effects. Temporal means time based if the terminology is confusing.

The way to create a spatially normalized stack of portraits is to the use Inverse Warp PASeq Context feature available in Studio Artist 4. What this lets you do is draw a set of bezier curves to define anchor points on a standardized image. You can then edit those bezier curves for all of the other images in the stack, so that after being processed by the inverse warp every image in the stack will be warped to the spatial positioning of the initial standard model image. So if you're dealing with portraits, you are interested in things like the facial features. So the positions of the eyes, mouth, overall head shape, etc should all match up after they have been spatially normalized.

I've set up the Portrait Virus Mutations blog to auto-post new imagery every day. So there's a lot of different effect variations that are explored over a series of posts that will be unfolding over the next month. And the results look pretty cool in many cases, but many of them are actually based on using the old inverse warp code that was not working in what i would consider the technically correct way it should work. So this gets into the interesting conceptual area i discussed above, which is technical 'mistakes' that might be visually interesting even though technically the processing used to generate them was incorrect in some way.

After running a lot of different stack filtering variations on Michael's Siggraph 2010 Portrait Virus paintings, i decided to try a series based on the painted imagery in his daily personal portrait blog. There was something that was bugging me about the imagery i was generating using Michael's self portraits that ultimately lead me to discovering the problems with the inverse warp code and of course we fixed that as soon as we could quantify what was going on. And of course after making those changes the processing based on the spatially self-aligned portraits took on a whole new level of interest. At least for me, because i was looking for certain effects to really pop out that weren't in the original portrait series that now are with the second personal portrait series.

I thought about going through the queued daily post images for portrait virus mutations and redoing everything, but i think it's more interesting to leave things the way they are and let the whole process of self discovery unfold over time on the blog. So at some point i'll revisit the original siggraph 2010 portraits and check out the visual differences between the technically incorrect and technically correct processing. Of course some people might like the technically incorrect processing the best, which is part of the point of this discussion.

On the other hand, the whole inverse warp concept is a really powerful one that i feel is very under utilized in Studio Artist 4. So i'm glad we were able to improve it's performance. And this whole process of putting together this particular little art project has me thinking about a lot of different variations that could be generated using this kind of creative process. One based on generating an image from processing multiple images in a stack of images.

[John Dalton Synthetik Software]

There's a new Studio Artist News post that discusses how to work with the inverse warp features to build spatially normalized image stacks. And the post provides some more information on the whole concept of stack filtering in general as well.

Again, thanks to Michael Wright for letting me use his Studio Artist generated portrait virus imagery as source input into my own Studio Artist experiments with stack filtering.