Edit

May 11, 2004

Videodrag proof-of-concept

It was live, and giggle-worthy, which may be a good thing... first cut of videodrag is in.

Click on the following image to play video

Lower quality video | Higher quality video

Notes:

schematic.jpg

USER EXPERIENCE

Choose a clip from jukebox
Stand in front of a bluescreen and be filmed miming along to clip
Watch preview for action, hard-of hearing subtitles, and guide cues (for Close up, Mid shot, Wide shot)
Move to marker A, B, C respectively depending on the cue
Pay to have clip sent to you

POST PRODUCTION PROCESS

Top and tail clip (longer the better) taking audio into consideration
Decide placement of punter as if they were one character (i.e. there may be more than one placement possibility per clip in which case go through whole process separately and add a number to clip name)
Create matte to separate foreground from background (optional but desireable)
Determine guide cues (based on particular clip context)
Record description of context and list of cues (e.g. C\t00:00:02:23:40 means close-up at 2mins:23)
Prepare cues using video compositor (or custom tool)
Export video as two files (_alpha.mov and .mov). The former is the matte, encoded with Animation codec (32bit Million+ colours) and contains RGB + an Alpha channel / and guides, no sound). This video determines when the blue-screened punter is visible or not. The second video is the background plate and contains the regular video and sound. This can be 24bit to save space.

ISSUES) Video dimensions/performance trade-off.
ISSUES) Choice of codec- PhotoJPEG/MPEG4/DV-PAL
ISSUE) Guide cues hard-coded. Can they be stored/generated separately so that lead-in timings can be altered according to difficulty setting?

ARKAOS NOTES

Create new synth (one per movie)
Create new patch (one per clip)
Map the alpha matte to C4 (latch, set layer priority to Always in front)
Map the background plate to D#4 (latch, set layer priority to Always in background)
Map the live camera to C#4 (adjust mask settings to key out background)
Import and rename videodrag_trigger.mid to clip name
Test clip using close-up, mid shot and wide-angle biped animations to test clip without live key
Test clip using keyed image

ISSUE) Latency of DV camera through Arkaos

MIDI NOTES

Prepare sequence with three keys to trigger alpha video, background video and live camera feed.
Give a few seconds of lead-in before notes

ISSUE) Hard-code live feed trigger or trigger manually

PERFORMANCE NOTES

Three markers - Close up, Mid shot, Wide shot
Treat the guide cues like a rhythm game (i.e. jump to the mark)

ISSUE) Show punter their image on preview?
Pro - punter can line up moves in line with background video, camera can be fixed
Cons - punter may be more comfortable not seeing own image, put faith in camera operator

NEXT STEPS

Sponsorship for Chromatte and Litering
Focus group (i.e. party) to test initial clips

Posted by .M. at May 11, 2004 08:24 PM
Comments

Looks excellent! Role on the blue screen

Posted by: damon at May 12, 2004 06:02 PM

Damon writes:
> This is probably a stupid idea..... But Im
> gonna tell it to ya anyway.
>Can we do it with flash? Combustion can export >stuff like matts to flash.

Not stupid. Maybe plausible in some cases to do alpha channels in flash BUT since we’ll normally need the foreground video as well in the file, not probably a good idea.

If all we were doing was switching the live feed on and off (so punter was always in foreground) then yes, this would make sense.

Maybe another idea to consider is to base the mask on the luma of the source image rather than an alpha channel. That's how I do much of my VJing.

In future put this kind of shit in blog comment so we don’t lose track of discussion for new-comers…

Posted by: .M. at May 12, 2004 11:34 PM

Looking nice.

My main gripes are with the overlaying of the live performer into an existing plate - sometimes what is directly behind them can be very distracting. Note in particular the close up on Hugo Weaving with the model overlaid - it looks as though Hugo is burying his mouth in the model's hair and singing into his scalp...

This is less of a problem for the wides and mediums, where depth perception can be a little looser, but on those close-ups, any chance to maybe matte out Hugo's face?

Other comment is placing - the current setup allows for only 3 placings, all dead centre. How about if the faders on the operator keyboard were setup for some control of lens length and lateral placement of the overlaid model?

Otherwise all sweet and good - Damon, that rotosoping must have taken you an aeon!

Posted by: ken at May 13, 2004 12:17 PM
Post a comment









Remember personal info?