Don't blow out the highlights

I think he was just being 'poetic' there :-)

I guess my objection to the blanket statement about light-starving CMOS sensors for the best IQ lies in the differences between the kind of HD cameras guys like Hurlbut and Rodriquez shoot with, and consumer cameras like the GH2. Maybe you can starve an Alexa or a RED and get great results. Certainly you can surprisingly accetable low-light footage if you go with one of the extreme high-bitrate GH2 patches like Driftwood's. But clearly, the GH2/5D2 sensors are happier with a lot more light than I'm seeing in Hurlbut's clips.

If I'm missing something here, please pile on. But when I shoot my hacked GH2 with CBrandin's 44mb patch, things look best with a well-lit subject and a healthy histogram. As the light dims, things go progressively wobblier.

I think Hurlbut's great but I don't think the starved CMOS recommendation works for those of us in the real world shooting consumer DSLRs, even hacked GH2s.

I don't know if you guys realize these DSLRs see into the light a lot better that film cameras do. Since that is the case I would expect one to need a lot less lighting equipment on set than they would with film cameras. Maybe Hurlburt is referring to this.

@Ian_T

My sentiments exactly - full frame DSLR's basically devour light - hence 'starve the CMOS' - I wasn't being cheeky up above about Shane being poetic :-)

At the dark end of the GH2's dynamic range, the noise level of the image sensor decreases by less than 2/3rds stop as you increase the ISO over the camera's 4-stop video range from ISO 160-3200. At the same time, however, the dynamic range at the bright end decreases from nearly 11-stops at ISO 160, dropping down to about 7-stops at ISO 3200:

http://www.sensorgen.info/PanasonicDMC_GH2.html

What this means in practice is that the GH2 has a fairly constant level of shadow noise, regardless of the ISO setting. If you don't expose the sensor to enough light through the lens, image contrast will not be much stronger than the background noise. In addition, the granularity of the 8-bit compressed color data is very coarse at the dark end of the scale, and that tends to exaggerate the local contrast among shadow noise pixels and image details. This is where the GH2's low-light performance falls short of high-end Nikon DSLRs:

http://www.sensorgen.info/NikonD3X.html
http://www.sensorgen.info/NikonD7000.html

As ISO is increased, the camera amplifies the GH2's relatively constant shadow noise, making both noise and image details brighter and more visible. At the same time, the highlight clipping limit drops proportional to the ISO setting, reducing the overall dynamic range. The GH2 maintains a dynamic range of more than 9-stops up through ISO 800, which is close to the optimal ISO for low-light conditions. If you can't get adequate exposure levels at ISO 800, you need to either lower the shutter speed or get a faster lens.

The GH2 works great in low light, but there's no room to bring out detail in the blacks if you desire later. (Maybe this is improved with the latest ridiculously high datarate patches. Can't wait to test.) That's fine if you want to crush all your blacks and don't care about detail. It's fine for video and web where the blacks are going to get destroyed in compression anyway. Not so great for projection or broadcast, though.

When I saw the Shane Hurlbut quote, I too was perplexed. I've been experimenting with raising by blacks above the noise floor--either with lighting or Tiffen UltraCon filters. It makes for a very rich look when you crush the blacks back down and still retain detail without the noise.

I think the semantic split here has to do with what each end-user's trying to achieve. My error was in applying Hurlbut's starve-the-CMOS advice globally, like it was something every DSLR video shooter should do for the best results. But what I think he really meant was that we can shoot much darker scenes with modern CMOS cameras and things won't fall apart like they used to...*IF* your desired end result is a dark scene. His example clips are all very dark, and they look great. I've shot very dark scenes with a GH2 using what's now probably considered a middle-of-the-road patch, CBrandin's 44mb, all the way up to ISO1250 and they look fine. This linked example was shot with a Canon FD 35mm f2, no post processing:

http://dl.dropbox.com/u/20129/gh2lowlight.zip

So if you're shooting a movie and you want a scene to look really dark and noir, starve away.

But if, say, you're shooting something indoors with available light and you're stopping down to keep everyone in focus, etc. You're starving your CMOS but Unkie Shane sez thumbs up, so you shoot away, thinking you can kick up levels in post. Except you can't, because when you go to grade it brighter you're going to see a lot of nasty artifacts and noise even Neat Video can't really cope with transparently. Waaaa! Unkie Shane lied to us!

No, he didn't. Unkie Shane's right, you can starve your CMOS if you want a dark scene. But if you want to grade later and bring levels up so the scene looks like it had more light to begin with, then, no, you should never starve your CMOS. You should feed it plenty of light and keep the nasties down at the barely perceptible noise floor where they belong.

I saw Hurlbut say the "starve the sensor" thing at the end of a product video for his new line of rigs. I think it was on FilmRiot. (guilty pleasure.) So it sounds like he's saying it a lot. I'd love more clarity on this. The guy is a talent, no question. If he knows something, I'd love to benefit from his insight.