Revisiting a Certain Compositing Test
So a little while back, I tried out Blender's Node Editor for the first time, and I realized I was able to composite footage of myself onto an image of the same shot's background (which, as it turns out, would be the same base technique at the heart of the superhuman leaping/falling distances in Clash of the Jedi and ABYDOS). I decided to try it out using the Keying node, and while it looked pretty cool, it wasn't quite what I had in mind…
So, over half a decade later (during my hiatus from the blog in 2019, in fact), I realized I could probably do a better job by now, so I decided to redo the same test. No problem: just fire up After Effects, drop Keylight onto the footage and throw on an adjustment layer to color correct the remaining yellow-green tint… but then I realized an issue. I didn't even have After Effects in 2013; of course I'd be able to do a better job in a more compositing-oriented program! Instead, I decided to go back to Blender.
First off, a warning about this. I tend to enjoy going back and updating older projects (as you may have figured out from my Yoda posts). The big problem with this strategy is that while it can document growth over time (or even hopefully encourage people new at it), it's easily possible to get too wrapped up in proving something about your progress to yourself. If you're particularly like me and find yourself wondering “Well, should I have gone back to Blender ~2.61 to make sure I was only using the features available to me in 2013?”, that's probably a bit much. At the end of the day, growth-based updates can be fun, but they can also be an easy ego trap to fall into. And, keeping that in mind, it's time to get to the more technical part: the next three paragraphs!
For the core keying, what mainly occurred to me was that I had no idea what I was doing and needed to poke around YouTube. Welp. At least that helped with the ego-proving risk. Anyway, I still used the Keying node, which is apparently best used as a factor input for a Mix node that combines the foreground and background elements.
The core issue with the 2013 video was actually simpler than I'd thought: the still photo I'd taken as a “background” was shot at a 4:3 aspect ratio, while the video itself was 16:9. Keying the blanket's orange color out of the video wasn't a huge issue for the center, as it had the yellowy background photo to fall back on, but the edges of the video were stuck at an orange-less green. I'd solved this a bit later the same year by scaling the image up, but the slight mismatches still gave everything a rather unpleasant yellow-and-green “duplicate” effect. To compensate for this in my update, I decided to match them further using a Transform node on the image, rotating it a mere .6 degrees to make the match as close to seamless as I could. (Of course, I could have just used a screenshot from the original video before I walked into the room, but that hadn't occurred to me then.)
The final step was the color grading: the colors were pretty decent this time around, lacking anycool ugly edgy? distinctive green tones, although both the original footage and the photo still carried an unfortunate yellow tint, likely due to the lighting in the room (while I shot ABYDOS on a different camera, one may still notice that the footage from that room has the same issue; thankfully the camera I used in ABYDOS II has a white balance adjustment for video). It took a Color Balance node (shifting the midtones more Magenta to dump the green), a Hue/Sat/Value node and two blending nodes (a teal Color and a blue Overlay, respectively), but the room finally looked appropriately colored--and, better yet, the use of the higher-resolution photo as the background actually brought out a massive amount of detail (such as the words on the “Virtues” poster) that the video footage had totally lost! Without any further ado, here's my updated compositing test:
I'm still transparent in the videos, although some of that's to be expected when one uses orange or red for a matte; I could use a mask on myself if it were an important scene, but I don't think it'll be needed here.
So, over half a decade later (during my hiatus from the blog in 2019, in fact), I realized I could probably do a better job by now, so I decided to redo the same test. No problem: just fire up After Effects, drop Keylight onto the footage and throw on an adjustment layer to color correct the remaining yellow-green tint… but then I realized an issue. I didn't even have After Effects in 2013; of course I'd be able to do a better job in a more compositing-oriented program! Instead, I decided to go back to Blender.
First off, a warning about this. I tend to enjoy going back and updating older projects (as you may have figured out from my Yoda posts). The big problem with this strategy is that while it can document growth over time (or even hopefully encourage people new at it), it's easily possible to get too wrapped up in proving something about your progress to yourself. If you're particularly like me and find yourself wondering “Well, should I have gone back to Blender ~2.61 to make sure I was only using the features available to me in 2013?”, that's probably a bit much. At the end of the day, growth-based updates can be fun, but they can also be an easy ego trap to fall into. And, keeping that in mind, it's time to get to the more technical part: the next three paragraphs!
For the core keying, what mainly occurred to me was that I had no idea what I was doing and needed to poke around YouTube. Welp. At least that helped with the ego-proving risk. Anyway, I still used the Keying node, which is apparently best used as a factor input for a Mix node that combines the foreground and background elements.
The core issue with the 2013 video was actually simpler than I'd thought: the still photo I'd taken as a “background” was shot at a 4:3 aspect ratio, while the video itself was 16:9. Keying the blanket's orange color out of the video wasn't a huge issue for the center, as it had the yellowy background photo to fall back on, but the edges of the video were stuck at an orange-less green. I'd solved this a bit later the same year by scaling the image up, but the slight mismatches still gave everything a rather unpleasant yellow-and-green “duplicate” effect. To compensate for this in my update, I decided to match them further using a Transform node on the image, rotating it a mere .6 degrees to make the match as close to seamless as I could. (Of course, I could have just used a screenshot from the original video before I walked into the room, but that hadn't occurred to me then.)
The final step was the color grading: the colors were pretty decent this time around, lacking any
If you're curious how it stacks up, here it is compared to the raw footage, my After Effects test, and the more mild 2013 update I'd mentioned earlier:
I'm still transparent in the videos, although some of that's to be expected when one uses orange or red for a matte; I could use a mask on myself if it were an important scene, but I don't think it'll be needed here.
Comments
Post a Comment