@onionbrain I haven't started using it in comparisons yet but I plan to shortly. How big a step forward would you say it is from Cluster 7 and where do you notice the improvements the most?

I remember at high ISOs with Cluster 7 AN Sharp 2, the 60P modes yielded I-frames similar in size to CM Night but much lower than Crossfire 3 CM Night. If you get a chance to test 60P before I do, could you take a look at whether there's been much change in that area? I'll probably compare 24H modes first. :)

Shot using Moon Trial 5 and various lenses.

@joofmagoof, excellent test! Could you upload some original files on the dropbox? Specially first shot (focus on the river) and last one would be very informative.

@onionbrain Thank you for your comment! I appreciate it! I like your vlog on youtube!

@onionbrain this is so wonderful to hear that you used hacked GH2 for such high profile work. I'ts really inspiring. I'm starting to work on a documentary for an European TV station, but filmed in the US. Would you be so kind to share what kind of set up you used and gear/lenses? These maybe too many questions, but I'd really appreciate whatever info you can spare:)

1. Would be very grateful to know what kind of post-production route you took - e.g. transcoding, editing software.

2. Also, did you have any problems with low light situations?

3. So there is no noticable difference between long GOP and Intra (like Moon)? I'd like to get maximum quality but preserve card space as well so Long GOP would be a smart choice.

@driftwood driftwood I have two 64Gb 45mb/s cards. Since Drewnet latest patch doesn't use really high bitrates, do you think it's safe to use these cards?

Thank you in advance

@mastroiani Thanks.

1) Copy the cards to a disk drive, import in to Premiere Pro CS6, edit, and export. That's it. No transcoding, and Premiere can deliver to almost anything.

The GH2 is a 24p (h) camera. That's what it does on a level that competes with cameras way out of its league. If you're delivering to a PAL market -- the big idea is that you film in 24p, produce in Premiere on a 24p sequence, and at the very end when it's time for final export -- you encode it out on a 1080/50i stream. Premiere Pro will handle the rest to make it work. For the hacked GH2 -- the notion that you need to film in the consumer implementation of "25p" in order to deliver to a PAL client -- is a very mediocre option that somewhat defeats the whole strength of the hacked GH2.

2) For low light -- ADD LIGHT. The whole "low light" thing is one of the most misunderstood concepts in the amateur/aspiring production community. Stuff that looks like low light from Hollywood -- wasn't actually filmed in low light. What you see on a screen may look dark with deep shadows and all -- but -- there was enough light on the production end of that scene to blow all the fuses in Afghanistan.

All cameras live on light -- they eat it. The Sony F65 would SUCK without enough light.

You can add light cheaply with work lights or whatever (always custom white balance), you can explore that whole new wondrous world of LED lights from China that span $20 to $2000 for almost the same product, I've successfully used high CRI, CFL bulbs in those multiple unit soft box type things from B&H, or you can spend a fortune on those obsolete Arri lights. Whatever you do -- you add light -- and learn how to shape it and use it. If you develop this skill -- it doesn’t matter if your GH2 isn't as good as a Sony F3.

Beyond that you can raise the levels in post on hacked GH2 footage a stop or two with good results.

3) I'd agree with that entirely. The qualification being "high bitrate long-gop," like Cluster X DREWnet trial 8 (not standard long-gop like the AVCHD variety). I tried to break it -- I aimed it at television static and at fast moving water with wide angle lenses. I couldn't do it. So, yes, I'd be entirely comfortable taking a high bitrate long-gop into any fast motion situation.

Enjoy the journey!

@driftwood I don't understand the "static" thing. I feel like people are missing something here. As mentioned above -- I threw everything at it and couldn't "break it" visually. It absolutely TROUNCES that Canon XF codec (50 Mbps, mpeg-2) that people rave about.

@peternap Thanks, old friend. I'm certain you're busy as hell these days!

@onionbrain The wonderful thing about Cluster X DREWnet Trials 7 & 8 is that they comfortably get down to incredibly low quantisation values of between Q8 to 16 - this is on a par with the best QP from the new GH3 BUT its doing it in smaller subblocks of 4x4 throughout the whole picture! (thats stunning providing tigher quanted detail) Its not suprising you love it :-)

I'll be doing some water/movement comparisons between moon and DREWnet to determine how each get on and the amount of bitrate used between them shortly.

@dado023 Thats the whole point about prediction - predictive frames such as p/b DON'T require as much bitrate as i frames.

The noise issue comes down to more than lighting it would seem. As an experiment, I filmed a few dark objects in a well lit room at 0 EV (the incident light reading from a meter) and progressively stopped down the lens -6 EV's, from f/2.8 to f/22. And what I discovered is that it's not so much low light but dark objects that make the noise apparrent. At proper exposures, noise is evident on a 'well lit' black object even though every light-colored object appears entirely noise free. It would seem as though the noise is always there but just gets masked by light. So, from my understanding, the ideal when shooting on the GH2 would be to avoid dark objects altogether - which is both impractical and aesthetically undesirable - or find a way to reduce the noise that exists in any given file.

Question (and I don't know whether it's even answerable or a proper question to ask): where on the luminance scale does the noise exist, or what is the average IRE of the noise put out by the GH2 sensor?

@thepalalias Thanks for confirming and re-enforcing. I think that's an important distinction from what I previously understood to be the case. This concept represents a paradigm shift for me.

Has anyone done an exhaustive test of Nebula? If Drewnet is good with motion, why would Nebula be needed?

@thepalalias In my film days, I worked out using a spot meter that the maximum difference in luminance of the lightest and darkest objects in shadowless light is about 3.2 stops. (I used a white car and black tires as markers on a cloudy day.)

So, if I have this right, a lighting ratio could be worked out that would ensure that dark objects recorded noise-free, to be pushed back down in post as you set out. Then, you could set up the lighting using an incident meter, greatly simplifying the problem.

For example, if the GH2's DR for noise-free blacks is 5.2 stops, You could set the base exposure with an incident reading at the camera, then take an incident reading in shadow. If the delta is less than 2 stops, you could drive on. If not, add fill light.

@CurtisMack Nebula is v good too. A lot of people prefer a shorter GOP (myself included with moon) where the i frames are closer together and there is less prediction. However its scalular - the shorter the GOP - the more bitrate is required to handle the higher avg of i frames in a second. With Long GOP, prediction frees up the memory hogging that short GOP requires by sampling sub blocks from the previous or future (as in B frames) i/P source frames used for prediction.

DREWnet is a very powerful version of traditional Long GOP as the i frames and low quantisation are running at maximum ensuring superior looking Intra frames and so forth sampled blocks - all running at around 80-88Mbps.

Nebula or other short GOPs arent quite as powerful on the QP /DC returned quantisation values but motion for me always looks better with fewer motion vector errors. All Intra is traditionally easier to edit but as onionbrain rightfully states in his video - with today's powerful computers - modern day cpus can easily handle the extra mathmatics to process Hi bitrate Long GOP. And prediction works very well in DREWnet - almost 99% free of artefacts. DREWnet users are happy to report that motion vector errors are few, if any.