[Bf-committers] render percentage and compositor behavior?

[David Jeske]

On Thu, Jun 27, 2013 at 1:44 AM, Sergey Sharybin <sergey.vfx at gmail.com>wrote:

> > From your reply, it sounds like the preview behavior is correct, and the
> > render behavior is wrong. Aka, that Mix and Alpha Over nodes should
> output
> > in the resolution of their first input. Does that sound right? If so I'll
> > file a simple bug/blend.
>
> Well, yes and no :) Just preview outputs in the actual resolution of a
> buffer in compositor. Composite output crops stuff to a render resolution.
> And this behaves as it is intended and used to behave since forever i
> guess.
>

Can we agree that the preview appearance should always match the final
appearance? This is not currently true.

Technically speaking it's possible to make render result be the same as
> composite, but that'd be a breakage of current things and here i could not
> make any decisions. Would rather leave it to Ton to make such a decisions.


I'm not sure what you mean by "make the render result match the composite". Let
me rephrase the problem.....

 Currently there are many situations where composite "preview" and
composite "final" do not look the same. This is because in final, nodes
individually crop to "3d render resolution", but in preview they ignore
render-resolution and "appear" to crop to the size of their first input.
 (see my blend file)

As for how to fix this... Two options are:

(a) Change the preview behavior to match today's "fixed resolution" final
behavior..... Individual node previews will appear as if they are cropped
to the current scene "render resolution", so they will appear the same as
the "final".

(b) Change the final behavior to match the "flexible resolution" preview
behavior.... Individual node "final renders" will be done using strictly
their input resolutions, ignoring render-resolution. "render resolution"
will only affect the size of 3d RenderLayers sent into the composite, and
(possibly) the final composite output size.

Shot term, "a" might be a good option, because it will not change final
render behavior, it'll just fix the preview to match.

Long term, I think "b" is the right choice, for several reasons.

1. As far as I can see, B is strictly more capable than A. I believe
everything users can do today in the compositor with "A" can be done with
"B". I am certain there are several things that can be done with "B" that
can *not* be done with "A". For example, today's "A" model can not handle
composite situations where any intermediate nodes have outputs larger than
the 3d render size.

2. Blender is an excellent 2d image and video compositor even if you don't
need 3d rendering, arguably the best open-source compositor! I argue it is
confusing for compositing only users to put in 2d sources, wire them
together, and then have crop behavior affected by a 3d render-resolution
settings for a 3d renderer the user is not even using. There are other UI
issues here, like "Render Percentage" and the "Post Processing" section of
the render panel, but I'll save that for a separate discussion.

3. A primarily composite user might want to add just a small amount of 3d
to a video, such as a small rendered 3d titling + logo. For speed reasons
it is beneficial to render only the small title, and translate it into
place. This can't be done "interactively" today, because

Well, [forcing all render-layers to the composite-scene-resolution] is how
> render pipeline is designed to work atm, and it completely makes sense. I
> could see how having different resolution could help in some cases, but in
> general it'll be just asking for problems. Namely it'll be just PITA to
> keep resolutions in sync in general
>

I think this could be easily accommodated by a (default set) option in the
RenderLayer node to force the render in the current scene's resolution.

VFX: for example you'll need to go over all the scenes o change render
> percentage.
> That's not something artists will expect to do.
>

Currently render-percentage and the compositor is very broken. It only
"works" if you are only compositing 3d renders. If you add any video or
image sources, the output is very very broken and requires manually
adjusted scale nodes after every image/video source.

Long term, I think "render percentage" should really be "proofing / proxy
percentage" and should "magically" downsize all rendering, compositing,
matting, video sequence editing, etc.

I also think there is a growing UI issue regarding the view-specific
Property-sidebars and the global Properties-panel. Because the Properties
Panel is dedicated to 3d operations, the different view Property-sidebars
(node-editor, movie-clip-editor) have started to effectively become "view
specific" property panels. Which is sort-of okay, except that it's not
possible for non-3d users (such as compositor or VSE only users) to use
only a part of Blender without digging into the complexity of the global 3d
Properties Panel.  I'm going to make a proposal about this, but it's a
separate discussion.