Compositor category

Compositor nodes appear in Sequencer pane and allow you to slide input/output nodes in time. As a side effect, IPOs for other nodes could be available for time and other adjustments.

Similar to Shake's Time tab.

Summary: Blender offers a wonderfully integrated package for potentially powerful non linear editing capabilities. To enhance this, the sequencer should be more tightly integrated in an seamless and transparent manner to the nodal compositor. This would both double the value of the sequencer and provide greater flexibility and features offered via the nodal compositor.

Example #1 CROSS FADING: As with the current implementation, the sequencer has a "Cross" effect used to achieve cross fading. This seems duplicated with the nodal compositor's "Mix" node. In an ideal world, the cross fading would be achieved through a pipeline into the nodal compositor. This would group allow for greater tweaking of the cross fade as well as providing a higher level of abstraction for quick application of a cross fade via the sequencer.

Example #2 BLUR: In a similar fashion to the above example, the "Blur" effect is also duplicated via the nodal compositor's tool base. Unifying the two aspects of the application would be useful for final post production polishing.

Example #3 TIME LINE: In some instances, harnessing the curve adjustment on a time node is more complicated than adjusting the time line in the sequencer. If the two were bound together, either adjustment would suffice and provide a much enhanced experience for the end-user.

In closing, more tightly integrating the sequencer with the nodal compositor would allow for a great number of positive outcomes for the Blender project, not limited to:

1) Stability. Developing only _one_ effect / node code set is far more effective than having to maintain two or more. Consider the examples above.
2) Consistency. Harnessing the same algorithms for a desired effect would increase the consistency obtained through Blender usage in output product.
3) Simplicity. Developing only one set of algorithms would reduce program size when it comes to the availability of plugins and effects.
4) Customized Noodle Effects. A group of nodes would easily be reused within the sequencer to achieve standard canned effects. This might open up avenues to build a noodle library for compositors / editors.

[....]

The Bezier spline has one major limit for doing proper rotoscoping. The spline's points are not animateable except for if you add hooks to the points.. The problem with this is that the handles also follow the empty, thus cancelling the auto feature on the points.. It would be extremely handy if it was possible to either keep the "auto" feature on even if the points are hooked. Or maybe an entirely new spline-object, like "roto-spline" with the possebility of animating the points without using emties and the "hook feature"..

hope this was understandable... :p

In addition to the current node system it might be useful if one there to have a set of nodes which were aware of the x-axis location and have time_variable settings like one have for effects in the sequencer window, in fact on could completely replace the sequencer with time based nodes.
see http://matthaus.woolard.googlepages.com/blender_node
for more details

There have been at least 2 ideas submitted for dealing with the compositor and sequencer. This is just an entry for my 2 cents. I mentioned this out of frustration in the bf-blender mailing list a while back and then thought, it would be better to actually post it on blenderstorm. So here's ideas (pasted from my email) that I think would allow the compositor and sequencer to be more integrated and useful.

1) Compositor should have sequencer input and sequencer output nodes.

2) We need to be able to have any arbitrary number of sequencer timelines and compositor networks (correct term?) as desired. These should be data blocks contained within the scene.

3) Timelines need to be able to have the output from another timeline as an input.

4) Compositor networks need to be able to have the output from another compositor network as an input. (Yes you could have just one big ass compositor network, but this allows for breaking it up into separate data blocks.)

5) The data of the timelines / networks need to be abstracted similar to how we abstract objects and object data.

6) Get rid of the compositor / sequencer rendering options. Both are on by default. If you don't need any compositing or sequencing done, you don't create any compositor node networks or sequencer timelines. (Besides you can't really enable just one if they are as integrated with each other as this proposal is suggesting)

6b) If your scene has compositing networks and/or sequencer timelines, but you still want to do a render without those, there should be toggle button (similar to the current "compositor" and "sequencer" buttons) to enable a "direct render".

I have some ideas for COMPOSITE Nodes:

1) Input->Value is limited to 0..1 - why? Sometimes i need higher values - my workaround is today to use Convertor->Math->Add (value,0)

2) Input numbers are not precise
they are stored precise but can not edited again:
enter 1.0/17 [enter]
and then add *17 [enter]

3) New Node: Input->Real Time
with switch frame/seconds
to get real frame number or
to get real time in seconds (input->time return only value 0..1)

4) New Node: Output->Value
to display current value if possible. This feature could make new blender calculator :-) or you can check values

5) New Node: Distort->Align
Could be useful for watermark-image-effect. Today you can only move images with distort->displace. But this is depend on your output resolution. This new node could align image left/middle/right and top/middle/bottom with additional x/y-space distance

6) New Node: Input->Pick
get color from x/y point from input image
or some extra options (get darkest, get highest color from image without x/y-option)

7) New Node: Color->Warnings
Colorize pixels with a warning color that are over a range 0..1 to show over-highlighted pixels in image

[....]

Not sure if this is a bug or a feature...

Many of the compositor nodes modify the image, however they pay no attention to the rendering option "key" or "premultiplied" when dealing with alpha. Generally they process image data using a premultiplied approach.

This can be a real PITA when using blender in combination with 3rd party apps. Especially when you have the "key" option selected for the rendering, and send it to nodes that change things using premultiplied compositing. Then you get this ugly mess that is a mix of both.

Try using a blur node on output rendered with the "key" option set. It totally screws everything up.

Have a node in compositor which would take your picture and give you black/white "mask" of color you would pick from it...
After that you could tweak the RGB curves node and the factor value would be this "mask"....
This could help for example if you render your image and want to turn color of this blue ball to the red... So you don't have to rerender the whole picture

Often one wants to combine 2 or more (bordering or overlapping) ID Masks.
The problem is that they often are anitalised and mixing them with nodes (i.e. "Mix" node) often results in mask-"halos" around some objects.
(EDIT: [Unusable] example of node-mixing http://www.flickr.com/photos/63231715@N00/2494419158)

So a way to combine them before they are converted to a bitmap (if that is even possible) would be quite handy here.

Of course a way to get a non antialised output of the "ID Mask" node might work as well and it may even be more flexible ("Mix" nodes will work correctly on purely BW images)

Example:
Combining mask 1 and mask 2 in this image:
http://wiki.blender.org/index.php/Image:Manual-Compositing-Node-IDMask_ex.jpg
Note: No, one can't simple take an inverted 0 mask here, because this is a simplified example and it will break if there are more masks.
The second example (with "halo" around borders") was already linked above.

Disclaimer: I'm still pretty much a noob when it comes to the compositor, so there may already be a way to do this cleanly. C&C welcome.

Werner

it would be a good idea to allow creating multiple compositing node setups, this would allow for example to add different compositing options for different strips in sequence editor (requiring probably to create an vse strip input node)

A new node that allows for keyframe input. It would have it's own little part in the IPO curve window so that the value can be edited and keyframed, then input into another node or multiple nodes that need keyframed values. This would be helpful to change the blending factor or the translation of an object for example.

A Random Value node could also be useful to allow for changing values in other nodes. This could be used to help create randomly flickering lights or overall flickering of the scene for example. This could also benefit from the keyframe node, allowing for the user to choose when the flickering occurs and to turn off the random node at certain parts in a sequence.