5.1k 12-bit RAW video captured with the M6 mark ii, using the 'RawBurstModeExtractor' tool

[TakahiroS]

Hello guys! This is my first post.

Recently, I came across a useful tool for automatically extracting individual RAW images from a roll of CR3 files taken by the raw burst feature of Canon cameras; the RawBurstModeExtractor tool (though it may be already commonly known).

Using aforementioned tool, I created a short video with my M6 mark ii and uploaded it to my YouTube channel. (Source footages are 5.1k30p 12bit raw, and the uploaded video is 4k24p encoded in H.265.) For those who are not familiar with how to make such videos, I've included some information in the description on the YouTube page.

-M6 mark ii 12bit raw video:

-The RawBurstModeExtractor tool (Thanks to Carson):

https://github.com/carsongmiller/RAWBurstModeExtractor

I hope you enjoy it!

 

[TakahiroS]

Is anyone familiar with recording raw video with M6 mark ii?

I believe that the combination of the RawBurstModeExtractor tool and the AI denoising software, such as the Neat video, makes the 5.1k 12bit raw video recording on M6 Mark II more practical. I feel it makes M6 Mark II a lot more attractive!

Many years back, I saw some people experimenting the raw video on M6 Mark II. However, they encountered two major issues. The first was that extracting individual raw images from the raw burst roll requires a lot of efforts due to limited functionality of Canon's DPP software. Fortunately, this problem has been resolved by the RawBurstModeExtractor, thanks to Carson. And the second issue was the short raw video duration (typically around 3 seconds). As Andi discovered (see the link below), underexposing each raw burst frame to around 8MByte can lead to 300 frames. From my experience, with a suitable fast SD card (see below), capturing 300 raw frames or more (e.g., over 10 seconds) seems possible even in bright sun light. These underexposed raw images usually include significant noise; however, the Neat Video can remove it without loosing image details, and the processed images still retain great grading capability since it is raw. The 10-second duration may not be long, but I think it is sufficient for many video works.

For reference, you can check out my YouTube video linked in my first post.

I’d appreciate any comments or additional insights. If anyone has information on extending raw video duration, please let me know!

Thank you!

-How to shoot many raw burst frames (Thanks to Andi):

-The best SD card I've tried so far (Does anyone know even better SD card?):

Nextrage NX-F2PRO series, SDXC UHS-II U3 V90, write 299MB/s read 300MB/s.

 

[PowerMike G5]

Interesting. Do you feel all the practical limitations of a workflow like this are worth dealing with in relation to whatever IQ advantages you get?

Regarding video on the M6 Mark II, I've created my own workflow to maximize quality as well. Essentially I shoot with a profile like Neutral and Fine Detail with adjusted contrast to tase, but also with all sharpening set to 0. Then use Topaz Video AI to not upscale the 4K video, but just sharpen it, and the results are really excellent, much more natural looking than the crude, jaggy sharpening in camera.

Its allowed me to have shots that compare to and match my Canon R5C, with its oversampled 4K.

I really suggest trying this out to see if you like it.

 

[TakahiroS]

Thank you for your reply!

I read your previous post about your workflow using Topaz AI — it sounds very appealing, and I’d love to give it a try! I think your workflow and the current RAW video capabilities could really complement each other, potentially turning the M6 Mark II into an excellent video camera in such a compact package.

The biggest drawback of the current RAW video is its short 10-second duration in bright sunlight. (In dimmer environments, the recording time can be significantly longer.) For certain applications where this limitation is acceptable, there may be some image quality advantages over the original M6 Mark II video. (I’m not an expert, so I could be wrong!)

The first advantage lies in color grading capability. As I mentioned in my previous post, shooting RAW video on the M6 Mark II requires underexposing, which reduces bit depth and usable dynamic range—perhaps to about one-third of the original 12-bit depth and 10 stops of the sensor’s dynamic range. Still, it offers significantly greater flexibility than the original M6 Mark II video, which is limited to 8-bit depth and about 6 stops of standard dynamic range.

The second advantage is resolution. While I’m not a pixel peeper and believe that FHD is sufficient in most situations, the 5.1K RAW video clearly shows finer detail—especially since the actual resolution of the original M6 Mark II video is reportedly around 3.4K. The extra pixels are also useful for cropping, panning, or tilting in post-production. (Note: the RAW video has a 4:3 aspect ratio, which is particularly advantageous for vertical adjustments like tilting.)

I expect that we will use your workflow in most situations, and choose the RAW burst video when we need extra detail and/or heavy grading.

Thank you!

 

[PowerMike G5]

Try using the Proteus AI model with "Add Noise" set to 5 and "Recover Detail" set to 20.

I find those settings to nicely sharpen the footage in a natural way without smoothing out any details. Let me know what you think compared to the raw footage you've shot.

 

[ThrillaMozilla]

Sorry, without checking the links, I have to wonder why exposure would have anything to do with the raw burst duration. This is just speculation, but I would think image size would be the operative parameter. Perhaps reducing the image size to 8MB or whatever is the way to do it. ???

 

[nnowak]

Sorry, without checking the links, I have to wonder why exposure would have anything to do with the raw burst duration.

When you drastically underexpose, you basically only record from middle gray to black. Since you are missing all file information for middle gray to white, the file size gets significantly smaller. This is really easy to see in JPEG images if you shoot two images back to back, but one with the lens cap on. The all black lens cap image will have a tiny file size.

The downside is that this massive underexposure dramatically reduces dynamic range while also significantly increasing noise levels. Underexposing by 4 stops would have the same effect on image quality as using an ISO 4 stops higher.

 

[ThrillaMozilla]

Sorry, without checking the links, I have to wonder why exposure would have anything to do with the raw burst duration.

When you drastically underexpose, you basically only record from middle gray to black. Since you are missing all file information for middle gray to white, the file size gets significantly smaller. This is really easy to see in JPEG images if you shoot two images back to back, but one with the lens cap on. The all black lens cap image will have a tiny file size.

The downside is that this massive underexposure dramatically reduces dynamic range while also significantly increasing noise levels. Underexposing by 4 stops would have the same effect on image quality as using an ISO 4 stops higher.

Interesting hack, although if you have to underexpose by 4 stops, and you're starting with 12 bits of linear (raw) data, the remaining 8 bits must be hard pressed to cover the range. The noise must be pretty bad, and it must lose significant detail.

I haven't messed with video. I wonder how it compares to just shooting video without the hack.

 

[TakahiroS]

Thank you for the comments, ThrillaMozilla and nnowak!

Yes, as nnowak explained, underexposing each frame is a way to capture more raw burst frames. Here are some additional information about the M6 Mark II's raw burst video. (Since I'm not a specialist, there may be some misunderstandings—please let me know if you spot any!)


Additional information on the raw burst video

The images above correspond to the raw burst footage shown around the 0:39 mark in the video linked in my first post. This footage consists of 235 frames—approximately 8 seconds at 30 fps, or 10 seconds at 24 fps—with each CR3 frame being about 8.6 MB in size.

The upper graph in Fig. 1 shows the waveform scope of the original raw burst video in DaVinci Resolve. Assuming the vertical axis is linear (is this correct?), it suggests that 10 bits of depth (out of 12 bits) and around 8 stops of dynamic range (out of the sensor's full 10-stop range) are still usable.

Fig. 2 shows the result after adjusting gain and some grading, where we can see some significant noise. However, as shown in Fig. 3, the noise is effectively removed by AI-based denoising (using the Neat Video plugin) without losing much detail. This may extend the usable dynamic range. The result is clearly better than the FHD image shown in Fig. 4.


The raw burst video and the original M6 Mark II's 4K video

Finally, the comparison above shows the raw burst video (originally 5.1K, downsampled to 4K and encoded in H.265) alongside the original M6 Mark II's 4K video (4K H.264, without any hacks). The former appears enlarged due to a 1.33x crop, whereas the latter uses the full sensor width. Ignoring the color mismatch, the raw burst video retains more fine detail. (As PowerMikeG5 noted, AI-based tools like Topaz AI can help minimize this difference.)

Thank you very much for reading. I hope this helps!

 

[TakahiroS]

Thank you for the information.

I will report how it goes when I try it.

 

[TakahiroS]

Hello!

I created another video using the Raw Burst mode of the M6 Mark II. This time, the footage was captured around dusk, so heavy underexposure was not necessary. (From my previous posts, underexposing is necessary when shooting Raw Burst video in bright sunlight to extend the recording time.) I observed that the underexposure can be relaxed or avoided in dim scenes and in scenes shot with a fast lens that produces a lot of bokeh.

If you're interested in the Raw Burst video or beautiful scenes from Hiroshima, Japan, please check it out!

Thank you!

 

[KEG]

Looks great, I have to try this at some time

 

[KEG]

What if you just put a really strong nd filter on the lens?

 

[TakahiroS]

Thank you for the comment, KEG!

Yes, we should give it a try. Nothing is too difficult as long as we have the M6 Mark II.

 

[TakahiroS]

Yes, we do want an ND filter on the lens, because the Raw Burst video requires　underexposure, as mentioned in previous posts. Under bright sunlight, achieving the best results may require a combination of fixed and variable ND filters. (I'm currently using NiSi’s TRUE COLOR ND VARIO.) However, the video posted on August 10 didn’t require heavy underexposure, as it was recorded in a dim environment around dusk. Choosing the optimal ND setting can be tricky — it often takes some trial and error to find the right balance between image quality and video duration. When you try the Raw Burst video, please let me know how it goes!

 

[kyserM6]

I finally jumped is this post !
I made a raw burst test



with a sd card Nextorage nx-f2pro256g r/w 300mb/s. the video was created from raw burst sequences. the first sequence CSI_0311.CR3 is made from 122 dng images of around 20mb shown on the video until 5.9s [the rest after are the succession of other sequences, CSI_0312.CR3, CSI_0313.CR3…].

I tried to made several burst sequences successively by pressing the shutter button for long times creating CSI_0311.CR3, CSI_0312.CR3, CSI_0313.CR3, etc. But as you can see , the video is desynchronized. It is like I lost frames between burst sequences. I used DaVinci to line up the sequences and export in video

I used a nextstorage nx-f2pro256g sd card under your recommendation. the exposure was dark to try to get the maximum of frame. but I could only get 122 images for the first sequence [CSI_0311.CR3 size:1.156GB].

How could you get 300 frames in one sequence? I was limited to 122 in the first sequence …If I understood well, your dng files size was around 8mb, mimes are around 20mb for each dng images despite the dark exposure…

With your Nextorage sd card how many frames is displayed in raw burst mode on your M6 m2 screen before to shoot ? [mine is 143]

thanks

 

[TakahiroS]

Thank you for the comment (and the comments in my YouTube post), kyserM6!

I used a nextstorage nx-f2pro256g sd card under your recommendation. the exposure was dark to try to get the maximum of frame. but I could only get 122 images for the first sequence [CSI_0311.CR3 size:1.156GB].

How could you get 300 frames in one sequence? I was limited to 122 in the first sequence …If I understood well, your dng files size was around 8mb, mimes are around 20mb for each dng images despite the dark exposure…

I think it might simply be because your underexposure wasn't sufficient. Based on my experience (and Andi’s YouTube post), each CR3 RAW image in the burst roll should be around 8MB in size (or roughly 15MB or less if converted to a DNG RAW file) in order to achieve 200 to 300 frames.

In the example video showing a cityscape under bright sunlight (upper figures 1 to 4 in this post ), the single RAW burst roll contains 234 CR3 frames. Each CR3 image is 8.6MB in size, and the corresponding DNG RAW file is 16.3MB.

In another example video showing trains in a dimmer environment (in this post ), I got around 250 frames in many of RAW burst rolls.

I think the required underexposure setting may vary greatly depending on the type of scene being recorded. Your example showing the LCD monitor seems to be one of the most challenging cases, as each image frame contains a countless number of fine LCD pixel structures across the entire frame. This can lead to larger file sizes and require extremely heavy underexposure. However, we don't usually encounter such extreme conditions in real-world scenes. In my experience, it's easier to capture more frames in dim environments (e.g., around dusk or in a forest) or in scenes shot with a fast lens that produces significant bokeh.

Could you try the Raw Burst video to record something more realistic?

With your Nextorage sd card how many frames is displayed in raw burst mode on your M6 m2 screen before to shoot ? [mine is 143]

Mine shows 147. We are the same in this aspect.

Thank you!

 

[kyserM6]

Now I have also 147 after formatting the sd card. this number go down after a bunch of shoots. this main number depends on the sd card performance I guess...

the previous test video was made with a sigma 30mm lens.

thanks for your replies, will make a test in real world next time, however is there a recommended setting to get the maximum of bursting time before shooting in burst mode apart from playing with the light exposure ?

thanks for your sharing!

 

[TakahiroS]

Now I have also 147 after formatting the sd card. this number go down after a bunch of shoots. this main number depends on the sd card performance I guess...

Thank you for the useful information! It's interesting that this number represents the conditions for continuous shooting, but it doesn't limit the number of recordable RAW Burst frames. (There's no cap on the number of RAW Burst frames?)

So far, I've only created two videos using the RAW Burst mode on the M6 Mark II, and I'm still experimenting to find the best settings. Aside from extending video duration, I've noticed a few things. When shooting RAW Burst video, the ND filter plays a more important role than it does in regular video recording. So it's worth investing in a high-quality filter. (I'm currently using the reasonably priced Nisi True Color Vario filter.) Moreover, an AI-based denoising tool is especially important. As is commonly known, RAW videos tend to contain noise, and the current RAW Burst video is even worse due to the necessary underexposure. To address this issue, the default denoising functions in video editing software may not be sufficient, and an advanced denoising tool, such as Neat Video, is essential.

I hope you have a successful RAW Burst video recording!