Fuji X-H1: new APSC video oriented camera

[Django]

11 hours ago, Andrew Reid said:

It seems Fuji are relying on Sony's hardware, with their own spin on ergonomics and colour. This is a good thing... Even if the H1 will be basically the innards of an A6500 in a Fuji body, it will be nicer to use than the Sony.

1

Sony may be building Fuji sensors (as they do for Panasonic/Nikon).. but claiming the H1 will be a A6500 in a Fuji body is a bit of reach!

For one the X-trans sensor design is quite different from standard Sony Bayer. It reduces moiré & gives a more organic/film noise at high iso.

AF systems are different. Rolling shutter is also nowhere near as bad as A6300/6500.

As stated above Fuji definitely have beta testers for their hardware "outside of the lab" but they usually go to.. Fuji users/advocates (like does every other brand!).

I'd say you'd probably have better luck with beta testing Panasonic bodies, if not already the case (GH5S?)

[Don Kotlos]

I have yet to see any advantage of the X-Trans sensor with video. The X-trans design might work alright for photographs, but for video it is much worse than a typical Bayer design. I would be glad if Fuji dropped it from their X-H1. 

That is due to two main factors :

1. Demosaicing x-trans design is much more computationally intensive

2. It is much more sensitive to undersampling

For optimal performance they would have to sample the whole sensor and use their fancy algorithm which I don't think they are doing with the X-T2. Look at how A6500 or the A7rII/III can give better results than the X-T2 with similar sampling. If you undersample differences become even larger. See the terrible aliasing with X-T20 compared to a similar sampling with A7rIII.

 

The difference in rolling shutter between X-T2 and A6300 is not that big and is quite terrible for both really. I consider anything above ~20ms to be not great. For example GH5 is pretty good at this:

GH5 --------------- 15.0 ms 
a7R II 4K FF ------ 19.9 ms 
XT2 4k ------------ 30.7 ms 
a7R II 4K S35 ----- 33.3 ms 
a6300 4K 24fps ---- 39.0 ms 

XT20 --> Doesn't look great

[Mattias Burling]

For stills I really dig x-trans. Pushing ISO feels almost endless.

[Andrew Reid]

X-trans also has a unique film like noise grain, which is noticeable in raw stills. Very very good for that, and fine detail.

Apart from the different bayer pattern, the rest of the sensor is manufactured and designed by Sony, based on what is in the A6300.

I don't find moire in 4K a particular problem on the X-T2. Only on a chart. Even the X-T20 is pretty good in the REAL world, outside of the chart tests.

Sure, it is not A6500 6K oversampled good... but you need to pixel peep to see this.

[PannySVHS]

4 minutes ago, Andrew Reid said:

 

Sure, it is not A6500 6K oversampled good... but you need to pixel peep to see this.

That is a good point. I have not found any true to the term shorfilm done with this wonderful XT2 on the web. Neither with cameras like the G7, GX85, G85, GH5 barely. I would love to see a Fuiji color shortfilm some time. But won´t happen soon, I´m sure.

[Don Kotlos]

I have the X-E3 and the stills are amazing. Fuji is the only company that I trust shooting jpegs, not only because of color but because the have this very organic noise with no compression artifacts. 

But in practical everyday use for video it does not look good. Believe me that you will notice the aliasing. Any type of fine pattern like hair for example does not look great. Forget about deep DoF outside shots. 

I would be crazy to not shoot video with the A7rII even in FF mode, but yeah for stills the Fuji is a joy to use. 

[IronFilm]

On 12/3/2017 at 5:09 PM, jonpais said:

I think most here would balk at purchasing $4,000 cinema lenses, but rental's always an option. While it's comparing apples to oranges, I know, these make the Veydra Mini Primes look like an even bigger bargain than they already are at $1,000 each, and X-mount is available as well. The Fujinon lenses are plastic and unlike the similarly priced all metal Sigma Cinema zooms, they don't cover a full frame sensor; though it's true that the Sigma 50-100mm suffers from breathing and isn't parfocal afaik.

I'd rather have one Fujinon zoom than 4x Veydra

Also Sigma's 18-35 & 50-100 don't cover full frame. 

[Django]

15 hours ago, Don Kotlos said:

I have the X-E3 and the stills are amazing. Fuji is the only company that I trust shooting jpegs, not only because of color but because the have this very organic noise with no compression artifacts. 

But in practical everyday use for video it does not look good. Believe me that you will notice the aliasing. Any type of fine pattern like hair for example does not look great. Forget about deep DoF outside shots. 

I would be crazy to not shoot video with the A7rII even in FF mode, but yeah for stills the Fuji is a joy to use. 

Please don't put the XE3 (or XT20) in the same basket as the XT2 or upcoming H1..

That camera does pixel binning with it's 4K image whereas the XT2 oversamples so you don't get any aliasing and have a much cleaner/detailed image.

Here is an XT2 video promo.. all footage ungraded and mainly shot handheld.. as you can see zero problems with hair or other fine texture:

 

[Don Kotlos]

26 minutes ago, Django said:

Please don't put the XE3 (or XT20) in the same basket as the XT2 or upcoming H1..

That camera does pixel binning with it's 4K image whereas the XT2 oversamples so you don't get any aliasing and have a much cleaner/detailed image.

I did not. My point was that the advantages that X-trans design might hold in stills it works against it in video. 

In the dpreview examples you can see than aliasing is still a problem even after oversampling. Usable definitely yes, but perfect as with a bayer oversampled image no.  For example at 1:16 in video that you attached the bright part of the hair looks like mush because of aliasing:

You wouldn't see this with an A6500. 

Now if they somehow manage to do the whole sensor readout and apply the same algorithm as in stills it could be better, but they would still have to improve the rolling shutter cause it is bad. 

Now don't get me wrong, I have two Fuji cameras and I would love to get an X-H1. 

 

[Django]

XT2 already does full sensor readout with no line skipping or binning and oversamples from 6K.

As Andrew said, only on those (dpreview) charts might you find "issues".. in the real world that doesn't apply as you can see for yourself in the numerous XT2 vids out there..

Also, again as others have pointed out, there are advantages to X-trans design in video such as the mainly luma noise at higher iso which gives a much more filmic noise than Bayer sensors where random/fixed noise patterns or banding may be encountered.. 

[Attila Bakos]

It actually downsamples from a 5120 x 2880 area, it has a 1.17 crop.

[Mattias Burling]

That amount of pixel peeping shows nothing imo. Who would do that in real life.

[timmyturntable]

1 hour ago, Mattias Burling said:

That amount of pixel peeping shows nothing imo. Who would do that in real life.

I agree.  If I had a client that wanted to pixel peep to that level, I'm probably working with the wrong clients.  As a non-Fuji user, the appeal for me has always been the colors and profiles...

[Marcio Kabke Pinheiro]

On 03/12/2017 at 6:44 PM, Andrew Reid said:

I may be biased, but I think Fuji should get me to try out the upcoming X-H1 before it's released. I asked to try the 4K X-Pro 2 firmware and they told me it isn't something they 'do' ahead of release... I am sure they do in fact have beta testers of their hardware, outside of the lab. I would love to give them my feedback.

It seems Fuji are relying on Sony's hardware, with their own spin on ergonomics and colour. This is a good thing... Even if the H1 will be basically the innards of an A6500 in a Fuji body, it will be nicer to use than the Sony.

To really compete with Panasonic though, they need to go far beyond what Sony are doing.

No 8bit-only codec for starters.

Well, if Fujirumors is right, will not be this time: 4K, 1.17x crop, 30p, 4:2:0 8 bit 200 Mbps

https://www.fujirumors.com/fujifilm-x-h1-4k-1-17x-crop-video-specs/

[no_connection]

7 hours ago, Don Kotlos said:

That does not look like aliasing, you can see strands being just fine right next to the other strands. You can also see that those "aliases" have motion blur in them. Hair can look that way when there is a very small light source. You can also see some "glitter" through the hair. Granted it could be a bit of both, I did not check the video to verify

[Andrew Reid]

7 hours ago, Don Kotlos said:

I did not. My point was that the advantages that X-trans design might hold in stills it works against it in video. 

In the dpreview examples you can see than aliasing is still a problem even after oversampling. Usable definitely yes, but perfect as with a bayer oversampled image no.  For example at 1:16 in video that you attached the bright part of the hair looks like mush because of aliasing:

You wouldn't see this with an A6500. 

Now if they somehow manage to do the whole sensor readout and apply the same algorithm as in stills it could be better, but they would still have to improve the rolling shutter cause it is bad. 

Now don't get me wrong, I have two Fuji cameras and I would love to get an X-H1. 

 

Viewing distances are an anti-aliasing filter.

4K -> 1080p is also an AA filter. Display format as well.

Not to make excuses for a camera's technical shortcoming but it helps to get it into perspective.

Real world is everything.

[liork]

Well, no big news here: https://www.fujirumors.com/fujifilm-x-h1-4k-1-17x-crop-video-specs/

[Andrew Reid]

So they added F-LOG and IBIS to an X-T2 basically (if the rumours are to be believed)

I think I'm looking forward to the GH5S a lot more.

Also rumours Canon will get in the game with a GH5 competitor... XC10 sales better than expected, EF mount version, blah blah blah.

GH4 and Samsung NX1 were 2014/15 cameras, so Canon and Fuji have a lot of catching up to do.

[Attila Bakos]

17 minutes ago, Andrew Reid said:

So they added F-LOG and IBIS to an X-T2 basically (if the rumours are to be believed)

Exactly, and you don't really want to use F-Log with 8bit 4:2:0, I have found it to be unreliable even in 8bit 4:2:2 Prores. Sometimes even a simple S curve and a gamut conversion is enough to break it, I think it really needs 10bit, or some tweaking. Let's wait and see if the X-H1 gets it externally. My hopes are pretty low though.

[IronFilm]

4 hours ago, Andrew Reid said:

Also rumours Canon will get in the game with a GH5 competitor... XC10 sales better than expected, EF mount version, blah blah blah.

 

Wouldn't a XC10 with an EF mount (& thus S35 sensor) be simply a C100mk3?