Morphing between multiple cameras
One of our clients is looking to create a bullet time effect for an upcoming spot involving some sports action shot on green screen. If we set up an arc of 6 cameras separated 30 degrees from each other to shoot the action, would re-flex be able to create the in-betweens to create a fluid looking move between a series of 6 stills? In reflex we've really only done A to B morphs that only dealt with 2 shots - nothing that involved many images stitched together. Can re-flex accomplish smooth movement for a shot like this? I don't want the action to appear steppy as the morph keyframes move from image A to B to C to D. We'll want the most fluid move possible. Back in my elastic reality days I think this was called a jigsaw morph, but my old copy of elastic is long dead. Any suggested tips to achieve the same in re-flex? Thanks very much.
RE:Flex in which application? :)
If stills in Combustion and AE we have RE:Flex Morph (still morpher you locate on timeline) and usually such you can copy the shapes directly in RSMB to add motion blur as well
If 6 moving image sequences, using RE:Flex Motion Morph:
Maybe Preedit your 6 camera tracks into 2 sequences like this
(sequence 1,3,5 in trackA and 2,4,6 on trackB)
and go from Warp 0 to 100% to 0%, -- and 100% is where your cut to the next sequence you will morph to will be in your preedit.
to have shapes with alpha (no background) will help.
beware of too much rotation between views with a lot of interframe occlusions (eg a karate guy throwing his arm in front of his body).
Not knowing you or the material and how fast the transition is in the end, I would be more comfortable if you said 15 degrees :) Also in case you can also warp under 0% and over 100% -- get like maybe 7 degrees of extra interpolation and do only 90 degrees
Another approach when possible is to use a large turntable with a single HD camera and simply retime that.
Sorry about not specifying the host software....This will be in AE. Also, it will not be a motion morph. It will be morphing between the same still frame captured from 6 cameras. One of the scenes involves a dancer who will leap up into the air. The action will come still then the cameras will sweep around to front as the dancer stays still in mid air and then come back live when the move ends.
I could probably convince the art director to place the cameras closer together if you feel it will help us with the potential crossover problems you are talking about. Knowing this info, do you think re-flex is up to the task to create a fluid morph that feels continuous? My main concern is that it will feel choppy from position to position. Can you offer any advice how we might be best to attack this or should I do my setups the way I normally would when doing an A-B morph and just overlap them over the six shots?
The simple thing first:
1) RE:Flex Morph allows you to KeyFrame where your still frames are. So in this case it would be a single set of splines you animate in time.
2) You would save yourself a lot of trouble if the dancer was jumping say with his arm by his body as not to create "holes"/layers -- otherwise I am afraid you will have to break the limbs in separate layers and "fabricate" something to temporarily fill holes (parts of a picture not visible from next camera). To visualize maybe raise your hands facing front facing level of your head in front of the mirror and turn 30 degrees increment, next frame your hand is now in front of the head, this is the really hard part with rotations (the part of your head that is now behind the hand on one view).
3) RE:Flex has a magic button called Auto Align which computes optical flow between your stills. However the magic has limits. I remember at Mass Illusion (prior to this company) when we did the inital tests for the Matrix (with Nick Brooks as main morphing artist, I think it's on the first DVD at the end) with a lot less cameras then they used in the movie, we ended up having to break down the guy into per limb layers, stabilize the per limb layers, morph the stabilized views, then unstabilize them back... very painful
1) I suggest you get your still camera and shoot someone from two positions and evaluate how much work it is. I have seen good results (not without work) with about 15 degrees with careful thinking, not trying to discourage you but I am skeptical with 30 degrees view differences unless you patch it with tons of effects like motion blur streaks.
2) Stiching a live-action sequence to a set of stills is also not without problems either if the camera are at different res... if this is what you are thinking
3) Note FYI you can also do this in Twixtor at that point by working with a 6 frames input sequence and splines as it has built-in support for mattes so you can get 3 layers over BG. You drop a twixtor sequence into another AE comp, setting speed in Twixtor to 100% and then you play with the timeremapping curve in the other comp.
4) I still think this sort of effect when possible is easier with a large motorized turntable and a lot of FPS to start with and then you retime it to your taste. There are some commercials out there which do such effects but they typically use much more then six frames.
5) Another out of context idea, if your still camera can do 6 FPS, you could also play with acting the action sort of in slowmo and if you have 6 cameras, you place the cameras real close to each others (so the change of POV is easier) and you in post move most left camera to most right camera back and forth. If you can somehow find a way to rig the six cameras to something someone can hold while sitting on a chair that can rotate you could probably do some real creative perspective change (however without strings it's hard to jump in slowmo)
Pierre thank you so much for your thorough responses. We definitely have a lot to consider and if we can up the quantity of camera and shrink the distance between them, I will definitely recommend that to the production company. As always budget is an issue and it falls on the post house to work miracles.
Pete just sent me this link
It would be another approach :)