Automatte is good

Flame Flame Questions
1

Really loving the automatte - well done autodesk. even realising that its perfect place is probably in G mask. Its also very fast. however really need to be able to define the active region with a mask rather than a box. the box is too limiting. also some controls to fine tune the matte would be good - maybe have three levels of speed - slowest being the most thoughtful and accurate. I think if its going to stay in g mask then you need to add the ability to Gaussian blur the mask within the node. nice one flame.

1 Like
2

when I say fine tune the matte I mean fine tune what its looking for

3

I think there’s a limitation with ML tools that they need a square area to analyze. So even if you could define the region of interest with a mask/spline it would still process behind the scenes with a box that covers that area.

Also in case you didn’t try yet… you can keyframe the region of interest.

2 Likes
4

I thought that might be the case. oh well!

5

If you already have a mask (for the region of interest), have you tried using that mask to cut out the image (i.e. comp on black), and then feed this into AutoMatte with a square ROI? I know it’s an extra step, but would remove all the clutter that may get in the way.

3 Likes
6

good call. I will try that

7

Well a hundred shots later - Automatte is my new favourite flame tool. something that really changes the way I work and speed things up. Thanks flame team. A standalone one would be preferred for me as I tend to prefer not doing too much in one node and keeping the mask separate etc would be more useful.

8 Likes
8

You can thumbsup here:

ML matte builder similar to modular keyer

2 Likes
9

Does Automatte run on old Intel based Mac’s?

Thank’s!

10

Unfortunately, it is not supported.

1 Like
11

After days and days of using automatte it needs to exist as a standalone node. Every time I want to use it I have to go into g mask, add source, add selective go to selective add ML. Too much. Just add automatte node to batch and bingo you’re in.

1 Like
12

The guy who fought against that is gone now, so maybe it will happen in the unicorn future.

1 Like
13

Here’s the issue with a standalone AutoMatte node: you are saying that you need a simple node that only runs the model, but then our guess is that you will say that it also needs multiple layers, an ROI, ways to visualize the matte, tracking, more ways to isolate/refine the result, controls to shrink/blur the matte, anti-aliasing, etc. And what you are describing is the Image node.

So there are only two extra steps to get what you want after you added an Image node:

  1. Select the Selective object (Press the Tab key once)
  2. Add an AutoMatte layer

Since we view it as two simple extra steps, we do not think it is worth it to spend development resources to basically duplicate the Image node.

Finally, you could consider either:

  • Saving an Image node with an AutoMatte layer already created to your User node bin.
  • Setting an Image node setup with an AutoMatte layer already created as the default by going to Node Prefs → Objects & Maps → Set Current as Default.

Using one of these will basically give you the same as an AutoMatte node with the only extra step being selecting the already existing Selective object.

Edited after kexianliang post: You can also set the pipeline to Matte by default if what you want to output a Matte.

1 Like
14

I think it is necessary to develop an standaloneAutoMatte node with the aim of creating Alpha Mattle （Mask） input channels for other nodes. For example, when we use the Skinworks toolbox to perform acne removal, whitening, and refinement on faces, we need precise mask input to achieve precise object isolation. So an standalone AutoMatte node is necessary， which functions like MLHumanFace Extraction; MLHumanBody Extraction； MLHumanHead Extraction； MLSky Extraction。

15

Why? What would it provide more than the Image node?

17

0633e629453f866c24f92b267f7bbb8c
0633e629453f866c24f92b267f7bbb8c1267×803 109 KB

So an standalone AutoMatte node is necessary， which functions like MLHumanFace Extraction; MLHumanBody Extraction； MLHumanHead Extraction； MLSky Extraction。

18

I think you are wrong. It’s like in the old days of the modular keyer. I never go anywhere near the modular Keyer anymore. I would have an automatte node hook it up and it provides an automatte. simple. very simple dont need the layers - but leave them in if you want, right now going into image / mask and all of that is a bit off a pain. trust me ive done this a thousand times in the last two weeks

19

I don’t get your points at all and would be OK chatting about it in a zoom meeting. But tell me what is the problem once you do:

  1. Add an Image node.
  2. Select the Selective object.
  3. Add an AutoMatte layer
  4. Go to the Controls submenu
  5. Set the Pipeline to Matte
  6. Go to Node Prefs → Objects & Maps
  7. Set Current as Default under Default Setup

Then:

  1. Connect a new Image node to a clip

That’s it. You have your AutoMatte node. If you do not want Image to default to this, simply drop the Image node in the User Bin at step 6 above and name it AutoMatte.

20

These do need an ROI with Tracking. These are Matchbox shaders with only the model in them. For AutoMatte, you need the ROI, the tracking, the post-process tools, etc.

21

Sorry Fred - thats sounded a bit harsh! I just think it would be a good idea