The third gain reduction howto, using the new TNT build! must see!

Started by 70MM13, December 15, 2019, 11:28:33 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

70MM13

tim's new Q menu system is working!

here's a new tutorial on using it, in which i recreate the opening shot from my recent music video, "biorobot's requiem", which i just redid using the new TNT build!

i did two videos (identically exposed) of the same scene, one at stock iso 200, and the other at TNT iso 120 (approx) and i am providing DNGs from both versions for everyone to play with.  I suggest loading them into rawtherapee, increasing the exposure by 1 stop to really see the shadows, and setting the histogram to log-log and increasing it's window vertically to your entire y resolution so you can really see the top spike and compare the two files.  you will see that i actually gave the 200 iso version a little MORE exposure to counteract the naysayers.

please grab the TNT build from tim's thread on "cleaner iso presets" and join in our quest!  this is only the beginning...

here's the file link:
https://bitbucket.org/rivertim/magic-lantern-danneclone/downloads/

and here's the video...  i recommend watching the second demo first so you can follow exactly what i am doing in this one!

https://youtu.be/kwxX8H7dTdM

here's another example, as posted later in the thread, included here also for your convenience:



https://drive.google.com/open?id=1-5ykG8f2Xsvnlnx8X-wBOD0t06Cjj4fL

edit:  fixed links.  Please don't link directly to builds as it results in outdated builds being left scattered through the forums causing issues in the future.  //Audionut

Danne

That´s better testing. You might want to consider producing a third dng file to complete the test. When completing iso 200 and iso 118 reset regs and then do nothing but pushing iso to 1600(ettr). Record that file and then use it to match brightness to the other two. Now push the shadows and tell me what file you rather work on for your next movie?

testfiles:
iso 200
https://bitbucket.org/Dannephoto/magic-lantern/downloads/M16-0538_frame_1.dng
iso 1600
https://bitbucket.org/Dannephoto/magic-lantern/downloads/M16-0539_frame_1.dng

Ettr:
iso 200, exposure unaltered


iso 1600, exposure matched to iso 200


iso 200 crop


iso 1600 crop exposure matched to iso 200


iso 200 push the shadows to see what´s going on in the darker areas


iso 1600 match the exposure to the iso 200 shot


If your iso regs can match the iso 1600 file i´d say we have a winner build. If not i´d say we still have iso 200 with some fine tuning creating iso 118 that still is useful even if it´s not the miracle of the century ;).

EDIT:
To get really clean shadows I would start out exposing dark as you show and then simply push iso to 800 or 1600 consistently. With those setting I would also record darkframes and apply them to the footage. This would be the unrivaled recipe for clean, deep blacks.

timbytheriver

@Danne

Thanks for the tests. Interesting. I may have misunderstood this:

Quotereset regs and then do nothing but pushing iso to 1600(ettr).
but I've included what I think you meant!

I'd say that the lower isos here have (as expected) much more dynamic range, and that all things considered where the reg tweak is applied, have a better quality of shadow noise – less 'glow', and more akin to film grain.

Iso 200 regs


Iso 200 orig


Regs reset iso 1600 (no exp comp)


Iso 1600 regs


Iso 1600 orig


I may have missed something important in the methodology, (in which case please point out this out!) but I would say that the audio analogy is an excellent one whereby trying to balance pre-amps to their optimum before they hit the master amps is good practice.

I think that's the aim with this regs work! :)
5D3 1.1.3
5D2 2.1.2

Danne

Who knows about methodology. Everyone is having their own agenda ;). Anyway. If regs apply on all isos it's good. Sorry but I should put more time into this but other stuff is taking my time.

ilia3101

I tried the -44 and -4 register values, canon menu at iso 100, looks like having tweaks reduced overall exposure, but couldn't tell a difference in shadow noise, so maybe the tweaked registers were better but I can't say. Don't want to watch 13 minute tutorial in full, so I don't know.

I was hoping to match both shots in camera so that equal amount of highlights would be clipped (by fine tuning shutter speed), then compare shadow noise, but unfortunately raw zebras do not work with the tweak so I couldn't do that.

(I have no agenda but to get less shadow noise :D)

ilia3101

Quote from: timbytheriver on December 16, 2019, 12:41:34 PM
I'd say that the lower isos here have (as expected) much more dynamic range, and that all things considered where the reg tweak is applied, have a better quality of shadow noise – less 'glow', and more akin to film grain.

Iso 200 regs


Iso 200 orig


Shadows are better in the "Iso 200 regs", but highlights are more clipped, so it just looks like 1 stop higher exposure to me (or 0.99 as the sliders say). I would like to see it so that the lamp is clipped to the exact same degree in both shots, then compare shadows.

timbytheriver

@Ilia3101 Thanks for testing! There's actually some very useful info in that 13min vid – I'd persevere to the end! :)

Highlights are very slightly clipped, but so they are on the iso-200-orig shot! In fact I left that Over a tad just to address this point.

Also, it seems we are up against the limit of the accuracy of the exposure tools in ML. In the 'clipped' version the histo was telling me -0.2EV to clip! So it gets really hard to judge absolute clipping.

Maybe if we cold harness the power of the raw_diag.mo tool – we could get ultra-granular with our exposure: which is what is needed to make this work best.  :)

Anyone up for it? :)
5D3 1.1.3
5D2 2.1.2

timbytheriver

@Danne

It would ultimately be very satisfying to understand what we're seeing with numbers; a mathematical formula. I can't supply that I'm afraid.  :(

In the meantime, the testing will go on! And the reg tweaks: like a TV, if we don't like what we're seeing – we can always hit the 'Off' switch.  ;)
5D3 1.1.3
5D2 2.1.2

ilia3101

Quote from: timbytheriver on December 16, 2019, 01:43:30 PM
@Ilia3101 Thanks for testing! There's actually some very useful info in that 13min vid – I'd persevere to the end! :)

Ok I'll watcgh it later.

Quote from: timbytheriver on December 16, 2019, 01:43:30 PM
Also, it seems we are up against the limit of the accuracy of the exposure tools in ML. In the 'clipped' version the histo was telling me -0.2EV to clip! So it gets really hard to judge absolute clipping.

"Clipping" in my opinion is when a channel reaches the raw white level, which with your tweaks is lower, so the zebras think it hasn't clipped, when it actually has.

Quote from: timbytheriver on December 16, 2019, 01:43:30 PM
Highlights are very slightly clipped, but so they are on the iso-200-orig shot! In fact I left that Over a tad just to address this point.

Maybe they are clipped in iso-200-orig, but much less than iso-200-regs, when you compare them, iso-200-orig clearly has more of the original bulb colour still left.

70MM13

a few words of advice, and another comparison file for you to check out...

when using this mode, be very careful while calibrating the preamp gain to match the spot meter while pointing at a clipping bright light source.  make sure that you actually reach the level where the "OVER" indicator comes on!  if you do not, you won't be getting full range and it will never show clipped highlights on your display.

here is another comparison, this time as extreme as it gets!

i'm pointing at an exposed filament of a 4W transparent bulb in a dark room.

as in the first comparison file, i intentionally exposed the stock iso 200 higher than the TNT120.  but look at which one shows more of the ruler!

hmmm... :)

enjoy and please continue playing!

PS: is everyone aware that you can use this with ANY ISO?

remember, OVER is your friend while calibrating, just don't go any further ;)





https://drive.google.com/open?id=1-5ykG8f2Xsvnlnx8X-wBOD0t06Cjj4fL

DeafEyeJedi

One of the greatest 13 minute Blog if not the greatest, to be frank! (Thank goodness for the CC Subtitles)  8)

Funny thing is over the weekend I was wondering why my Q button wasn't responding to the shortcut keys.

Turns out I was using the build that I had downloaded right before Tim posted another build w the fix. D'oh!

Now that I've watched Tommy's vid ... I'll give this one a ride and will report my experience. Thank you all!
5D3.113 | 5D3.123 | EOSM.203 | 7D.203 | 70D.112 | 100D.101 | EOSM2.* | 50D.109

70MM13

jedi master!  right on, thanks!

i'm glad you're going to try it out... make sure you install and enable the RAW_diag module and watch this first, you'll be glad you did:

https://www.magiclantern.fm/forum/index.php?topic=24720.msg223628;topicseen#msg223628