adamwilt.com > HDV > Cineframe mode copyright © 2005 Adam J. Wilt  
 
CF24, CF25, CF30: How the Sony HDV cameras fake 24, 25, and 30 fps search

created 2005.01.23

The Cineframe shooting modes in the Sony HDV camcorders let you simulate the look and feel of progressive-scanned images, but the camera section itself is still running at its normal, interlaced field rate.

Cineframe 30 and Cineframe 25

Cineframe 30 on the 60i FX1 (Cineframe 25 on the 50i FX1E, or either one on the Z1 depending on its field rate) works just like the slow shutter on the FX1/Z1 and on Sony's SDTV DV cameras: it throws away one field and doubles the other. The resulting image has half the motion resolution of the normal interlaced mode and appears like a progressive frame, but it only has a single field's worth of information, so its vertical resolution is somewhat degraded, and diagonal lines may look "steppy" or "jaggy".

Because the field-tossing happens in HD, the downconverted SDTV output looks considerably better than field-doubled standard definition. Instead of being half as good as normal interlaced mode, Cineframe looks only slightly worse than interlaced. I definitely prefer the FX1 to any of my SD DV cameras if I have a need to shoot slow shutters or "the cineframe look" for standard-definition release.

How about in HD? If you have a true 1080-line display, Cineframe looks just like field-doubling (as you might expect), with half the vertical resolution of normal interlace shooting. However, if your HD display downsamples to 720 lines, 768 lines, or the like (as of early 2005, most of them do), the resolution loss and jaggy diagonals may be somewhat hidden by the rescaling, just as when it's downsampled to SDTV.

Cineframe 24

Cineframe 24 (CF24) is the really interesting one: it implies the look of 24fps motion picture film. For casual uses, it works very well: the images appear to have 3:2 pulldown, with two "interlaced" frames and three "progressive" frames in every five, and the motion is pleasingly stuttery. However, the camera itself is still running at 60i, so there's definitely some funny business 'twixt the chip and the clip.

Cutting to the chase: CF24 cherry-picks fields from the camera's 60i video stream to simulate the juddery motion of 24p. However, the motion is worse (more uneven, more syncopated) than true 24p imagery, and it cannot be reverse-telecined to get smooth 24fps imagery. CF24 is fine as a special effect if you're staying on video at 60i, but it's not suitable for a true 24fps feel, nor will it work for film-outs: the motion is too unsmooth, and the vertical resolution of the image is damaged. For film-outs, shoot at 60i and use a program like DVFilm's Maker or your film-out facility's own in-house process to convert 60i to 24p.

If you're still with us (good on yer!), Cineframe 24 appears to be mixing and matching a witches' brew of field-doubled, frame-based, and synthesized images in its five frame cycle:

  1. CF24 uses a five-frame cycle with apparent 3:2 pulldown. If we adopt 3:2 terminology, video frame 1 contains what would correspond to the A frame of a 24fps transfer; frames 1, 2, & 5 are progressive; and frames 3 & 4 show interlacing.

  2. In frames 1 & 2, the bottom active scanline is blanked (black), and vertical resolution is comparable to field-doubling. In frames 4 & 5, the top scanline is blanked, and vertical resolution is also decreased, but it appears that the other field is being doubled as there is a slight vertical shift in the detail compared to frame 1 & 2 (see the test-pattern details below). Only frame 3 shows all scanlines intact, with full vertical resolution.
frames 1 and 2
frame 3
frames 4 and 5
Frames 1 and 2
Frame 3
Frames 4 and 5

Hmm, we've got a problem here... item 2 implies field-doubling one field for frames 1 & 2, field-doubling the other field for frames 4 & 5, and leaving frame 3 interlaced. But in item 1, I state that frame 4 contains two distinct fields of video. What gives?

I built a full-frame red DV clip in Final Cut Pro with burned-in timecode. I then used my "Field Balance" FXScript filter to desaturate the first field of the video entirely, so for 1/60 sec the screen is gray, and the next 1/60 sec it's red (that way we have visible differences at the field level to make things easier to see). I put this clip up on an NTSC monitor and shot it with an FX1, panning a bit to create in-frame motion.

When shooting at 60i, each frame looks like the one shown here: two interlaced fields, one gray and one red.

sample 60i frame from FX1

In CF30, I'll get either an all-gray or an all-red image, or one split between the two the same way in every frame, depending on the phase difference between the camera and the monitor.

When shooting at CF24, things are a lot more interesting, as the following six-frame sequence shows. I'm showing downconverted DV captures, but the HD material has the same pattern to it. Commentary below the corresponding frame...

FX1 CF24 frame 1

Frame 1: A field-doubled image. I think this first frame of the five-frame "3:2 pulldown" cycle doubles the upper video field. In terms of the captured clip material, it's the first field of frame 00:00:39;09. It's the only time we see frame ;09 of the source; this corresponds to the "A" frame of the traditional pulldown with a "2" capture--two fields recorded.

FX1 CF24 frame 2

Frame 2: Field-doubled; same source field captured, but one frame (;10) later, a difference of 1/30 or 2/60 second.

FX1 CF24 frame 3

Frame 3: full frame, but a full frame of what? We have the same gray (first) field of ;10 in our upper field, making ;10 correspond to the "B" frame with a three-field capture in the pulldown sequence. The lower field is the red, second field of source frame ;11, captured a whole 3/60 second later! The camcorder has repeated the first (upper) field from its previous frame and only replaced the lower field with new info.

FX1 CF24 frame 4

Frame 4: This is another curious one. We clearly preserve the red, second field of ;11 in the upper (?) field, making ;11 the two-field C frame of the simulated pulldown sequence, split across two separate frames. But the other field? It's the second (red!) field of ;12, shot 1/30 sec after the red field of ;11--our frame 4 is made from two separate lower-field captures!

My best guess is that the lower field from our frame 3 was held over, essentially field-doubled into the upper field, and then overwritten with a new lower field. Now, I may be wrong; if so, please let me know (and kindly explain what's really going on!). But I think I'm right...

FX1 CF24 frame 5

Frame 5: We preserve that lower field capture of ;12's second field, field-doubling it, making ;12 the final, three-field D frame of the simulated pulldown.

FX1 CF24 frame 6

Frame 6: the start of the next five-frame cycle reverts to the upper field capture of source frame ;14's first field. In the process, source frame ;13 has been entirely skipped; we went from the second field of ;12 to the first field of ;14, a jump of 3/60 sec or 1/20 sec.

Thus, mentally reverse-telecine-ing the clip, we have:

The resulting time syncopation shows the 2:3:2:3 alternation of the simulated 3:2 pulldown cadence.

But what price does this clever simulation exact from your images?

First, of course, the 2:3 pulldown simulation is derived from 60i material, not 24p material, so the resulting "reverse-TKed" material shows that 2/60, 3/60, 2/60, 3/60 inter-frame syncopation, not the even 1/24, 1/24, 1/24, 1/24 temporal rhythm of true 24p material. You can see it below: I tried some simple motions with both the Panasonic DVX100 in 24P (3:2 pulldown) mode as well as the FX1 in CF24. I reverse-TKed both clips in Apple Cinema Tools 2.2, supered a series of frames, and highlighted the centerlines of the moving object (a pen I waved in front of the lens):

DVX100 true 24p cadenceFX1 CF24 cadence

The DVX100 footage shows a relatively smooth progression from frame to frame, while the FX1 footage shows the characteristic cadence of CF24 simulation: some interframe motions are 150% of the others.

The other issue is the mix of full-resolution and field-doubled images in the five-frame sequence. Although the differences are small, they do show up on relatively static images as a periodic vertical "breathing" as the image jiggles up and down very slightly and fine horizontal details pulsate in sharpness. The result is very much like standards-converted video going from PAL to NTSC or vice versa. You may wish to exploit it as an in-camera "film-look" effect simulating film's gate weave or jitter, but if you want rock-solid 24p emulation, this isn't it. The results are less pronounced in SD, due to the smoothing effects of the downsampling, but they're still there.

If I were going out to film, I'd stay with 60i or 50i in the camera, preserving as much resolution as possible and capturing the most temporal information I can. I'd leave it up to the film-out process to intelligently resample field or frame data as it saw fit to get the best possible image with the fewest possible motion artifacts.

But if I'm just looking to do a quick, in-camera bit of "film look" fakery, and CF30 (or CF25) still looks a bit too smooth to hammer my point home, and I'm not too concerned that the motion isn't true 24p, CF24 is a nice tool to have in the toolkit.



All materials on this page copyright (c) 2005 by Adam J. Wilt.

You are granted a nonexclusive right to reprint, link to, or frame this material for educational purposes, as
long as all authorship, ownership and copyright information is preserved and a link to this site is retained.
Copying this material for a publicly-accessible website (instead of linking to it) is expressly forbidden except
for archival purposes: this page is dynamic and will change as time goes by, and experience has proven that sites
copying my material do not keep their copies up-to-date, thereby doing their visitors, themselves, and me a disservice.

Contact info

Created 2005.01.23