[Bf-vfx] VFX related stuff that I think could be improved

Bartek Skorupa (priv) bartekskorupa at bartekskorupa.com
Sat Sep 7 13:08:56 CEST 2013


Hey,

Here go my two cents:

1. Speed:
I'd only like to say exactly what Sebastian said: Let's make it faster :-)
Cache, render only what necessary, this kind of things.

2. Canvas:
Some users love backdrop. I hate it. For me canvas is a separate window. Let it be the UV image editor, I don't care. I believe I'll hear different opinions here.
Anyway:
a) Ctrl-Shift-click should automatically switch UV image editor to "Viewer Node".
b) Transform controls:
This should simply be easy. I personally don't see the need for all those "Translate", "Rotate" etc nodes. IMHO Anchor Point, Position, Rotation and Scale should just be the properties of every node. "Opacity" would also be nice.
I make the node active, hover the mouse over "canvas", hit "R" and rotate. just like objects in 3D View. Hit "S" and scale. Just imagine: Select, hover, "SXX", move a bit, click, done. Image gets resized along "local" X just a bit. How much? As much as needed. You simply watch what you are doing and do it.
Additional gizmos in canvas would also be nice. Just click the gizmo and move the image, rotate or scale, do it with Ctrl or whatever other key to constrain proportions… I'm in heaven :-)
Those gizmos could control the active node's output. Just that simple.
I should also be able to enter numeric values of those properties in Properties panel of the node.
… and pleeeeeeeeease give us the Anchor Point. It's important to be able to rotate or scale around different point than the center of the image.

3. Masks:
Well… I simply don't know how on earth one can effectively use those crazy controls. It takes ages to draw simple shapes. Bezier curves with easy control of handles right after placing a point is the only option IMHO. Sorry for mentioning After Effects once again, but there I simply click and the point gets added. Then without releasing the mouse I drag and the handle gets defined. It's intuitive and easy. You don't think, you just draw.
In my opinion the right place for adding masks should just be the compositor window.
Shift-A, add mask, move the mouse over canvas and do the magic.
Select more than one mask nodes and you can control them simultaneously. How about that?

4. Color display when alpha different than 1.0:
I don't know why it's done this way. Every pixel has R, G, B and A values. Why when I disable alpha display I can't see the colors as they are? In order to see the real values (and look) of the image in semi-transparent or transparent areas I have to do all the premultiply, unpremultiply, unprepremutipremultidivide-setalpha-unsetalpha magic just to see the colors. It drives me crazy. I'd simply want to see the RGB values of transparent pixels. It used to be this way some ages ago. Why did it change?

5. Several new "distortion" nodes would be a great help. Something like: "Corner Pin", "Bezier Warp", "Mesh Warp", "Spherize" etc. plus "Movie Distorsion" where I could simply enter the values without having any movie clip. Why do I need this? I would simply like to distort some image using predefined distortion values.

6. Overscan: Well… What more can I say. It's crutial to be able to render larger image just to crop it after distortion. I have my workaround for this, but having a simple option in render settings would be great.


Those are my two cents.
Thank you.

Bartek Skorupa

www.bartekskorupa.com

On 14 sie 2013, at 10:44, François T. <francoistarlier at gmail.com> wrote:

> Compositor cache and overscan (I think that's what you meant by overshoot) are, in my case, the two most wanted things. 
> Yet Canvas comes really really close after. 
> 
> About having some primitive shape for masking, I guess a simple python addon could do. 
> 
> 
> 2013/8/13 Sebastian König <koenig.sebastian at gmx.net>
> Hey all!
> 
> Ton recently asked for feedback about things that could be improved in terms of workflows and tools for VFX work in Blender. 
> So I started collecting some things.
> And now I will just dump that list here! 
> Maybe it helps to start to discuss some things, at least to collect feedback.
> There is probably much more to it, but anyway, 
> here it is:
> 
> 
> 1. Compositor speed
> The speed is not optimal, and especially with huge nodetrees it can become quite slow. I wont focus on the speed issue though in this post. Just make it faster! And give us cache! :)
> 2. Compositor Canvas
> Currently it is a bit strange how Blender deals with dimensions. 
> The node tree determines the current „working“ size from the first node in the chain. So if you have an image in put that is 1024*768 for example and you connect a viewer to it, that image is displayed in that size, which is okay. But if you put that image through a translate node, and shift it 200 pixels to the left, it looks like these pixels would be cut off, the backdrop nor the image editor (when set to viewer node) doesn't show them. But they are still there. If you alpha over that image over another, larger image, you can see that they are still there. The entire system is kind of weird and hard to predict. 
> It would be nice to have some kind of visual feedback of the canvas size and dimensions. There could be a dashed line in the backdrop or viewer node that indicates the output render dimensions, and it should be possible to move, scale, rotate, transform any image or render input freely on the canvas.
> And it would be nice to have some visual controls for image transformations like rotate, scale and move other than just numeric input. I wouldn't to have a UI for that like in Gimp, Photoshop or AfterEffects. These controls would need to be displayed only when the transform node is active. 
> 
> 3. "Overshoot" : This is mainly related to tracking. Basically it would make it possible to distort CG images. At the moment, when we have tracked a footage that has lens-distortion, we can either undistort the footage to match the CG, or we distort the CG images. But that will only work if CG is always in the center of the frame and never goes over the image borders. Because if it did, the movie-distort node would bend the pixels inward to match the barrel distortion of the footage. As it is now that would leave an empty border around the image. That's why we would need to be able to render larger images so that when we apply the movie distortion the CG pixel match the footage.
> 
> 4. Lens Distortion Node should support alpha channel
> 
> 5.  Masking
> Masking feels a bit unfinished still. After Tears of Steel almost nothing really happened to improve masking.
> Some issues:
> - You cannot duplicate masks. That’s really annoying and makes it a bit tedious to mask multiple small tracking markers. You have to paint a circle for every single one of them.
> - Easy shape presets would be nice. There could be a circle and a square as shape preset. Well, that would be a bonus, but at least being able to duplicate masks is really essential.
> - I suppose if it is possible to duplicate a maskshape it will probably also be possible to copy and paste them. Sometimes you want to be able to move a maskshape to a different mask layer, especially if you have tediously animated it, but need to combine it in a different way.
> 
> 6. Visual Feedback for masking:
> It should be possible to see the black and white mask output right there in the MCE, including the feathered output. It’s not even super important to have realtime playback for that, even though of course that would be nice. But if I have to constantly switch to the compositor just to see the result it gets a bit annoying. 
> - And it should not only be possible to see the mask output, but also the composite result, specifically the viewer node output. 
> Currently you can use the image editor for that, but it doesn't give you the timeline display of animated mask keyframes, nor does it give you the ability to use tracks for that.
> That’s why I would really like to be able to pipe the viewer node output to the movieclipeditor, because there I already have all the functionality that I need to for rotoscoping and tracking. 
> 
> 7. Drawing masks: 
> Bartek Skorupa once already mentioned it, but in fact sometimes it would be nice to have bezier curves for masking as an option. Sometimes these would make it easier to control round shapes next to sharp corners. 
> - But even nicer it would be if you could control the length of a curve handle point by the duration of clicking. That’s how it works in After Effects and in fact that’s quite nice. If you ctrl+click shortly it would make a rather sharp curve point, but if you keep it clicked it would keep growing or shrinking the curve point. A bit hard to write how that works, but I hope it’s clear. That way drawing masks could be a lot faster, because you could already define shape and curve falloffs while drawing.
> - The direction of masks
> Currently you have to draw the mask counterclockwise so that the curve handles point to the outside. If you draw the mask clockwise the control points point to the inside. That’s not really a problem, but as soon as you try to feather it or enable AA for the rasterization it just fails. 
> So for example you have the mask with handles pointing inwards. Rendering that mask in the compositor with Antialiasing enabled will still create a mask with sharp jagged edges.
> Even worse with feathering. Even though it is possible to create an inner feather control curve, it simples doesn't compute.
> It would be nice if it works in both ways. At least it should be possible to have an inner feather control, that just feathers as  you would expect. Which means that the main outer control curve defines the end of feather where it is black, and the green feather line defines the falloff to the center of mask where it is white. 
> Even better if direction wouldn't matter and you could drag the feather control curve to both sides of the curve. 
> 
> 8. Feathering.
> Draw the feathered curve works kind of okay as long as you keep it super simple. But as soon as you have large and small control points close to each other, or have corners, it gets really crazy. There are a few workarounds with different feather offset settings and self intersection checks, but overall the whole way the feather curve is calculated could be smarter
> 9. Mask Interface
> The whole concept of Mask Layers is actually quite nice, but the interface is a bit crowded. What I find unnecessarily crowded is the opacity display in the mask layer line. Maybe it’s just a personal thing, but I almost never ever use it. And if I did, there is already a big slider for me to use, no need to display the value in the masklayer line.
> 
> There is probably a lot more, but this would be the first batch of things I would like to see improved.
> 
> 
> Cheers!
> 
> Seb
> 
> 
> 
> 
> _______________________________________________
> Bf-vfx mailing list
> Bf-vfx at blender.org
> http://lists.blender.org/mailman/listinfo/bf-vfx
> 
> 
> 
> 
> -- 
> ____________________
> François Tarlier
> www.francois-tarlier.com
> www.linkedin.com/in/francoistarlier
> _______________________________________________
> Bf-vfx mailing list
> Bf-vfx at blender.org
> http://lists.blender.org/mailman/listinfo/bf-vfx

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.blender.org/pipermail/bf-vfx/attachments/20130907/766d17d5/attachment.htm 


More information about the Bf-vfx mailing list