Pose In-Betweening doesn't work correctly with animation layers #61099

New Issue

Ovionis · 2019-02-01T16:24:35+01:00

Ovionis commented

2019-02-01 16:24:35 +01:00

Windows 7 64bit

Blender 2.80 #552b2287db86
(2019-02-01 01:46)

When using any of the pose Tween operators (Shift/Ctrl/Alt + E) / (Pose > In-Betweens)
The operator only considers the final absolute keyframe, and not what type of action it is or the layered result.

_Combine_Animation_Tweening.blend
If you open this file and try to run any of the in-between operators, you will see the bone jump far from where it is, to the center, based on it's curves' exact keyframe values.

(Also, would it be difficult to also add object In-Betweening? Unlike bones, if you animate with objects, your only option ever is/was to manually insert keys in the graph editor)

Windows 7 64bit Blender 2.80 #552b2287db86 (2019-02-01 01:46) When using any of the pose Tween operators (Shift/Ctrl/Alt + E) / (Pose > In-Betweens) The operator only considers the final absolute keyframe, and not what type of action it is or the layered result. [_Combine_Animation_Tweening.blend](https://archive.blender.org/developer/F6476437/_Combine_Animation_Tweening.blend) If you open this file and try to run any of the in-between operators, you will see the bone jump far from where it is, to the center, based on it's curves' exact keyframe values. (Also, would it be difficult to also add object In-Betweening? Unlike bones, if you animate with objects, your only option ever is/was to manually insert keys in the graph editor)

Ovionis commented

2019-02-01 16:24:35 +01:00

Added subscriber: @Phigon

Sebastian Parborg commented

2019-02-01 17:33:08 +01:00

Added subscribers: @JoshuaLeung, @ZedDB

Joshua Leung was assigned by Sebastian Parborg

2019-02-01 17:33:08 +01:00

Sebastian Parborg commented

2019-02-01 17:33:08 +01:00

@JoshuaLeung unsure if this counts an actual bug or not. So I'll leave it up to you to decide.

Joshua Leung commented

2019-02-03 13:05:19 +01:00

Changed status from 'Open' to: 'Archived'

Joshua Leung closed this issue

2019-02-03 13:05:19 +01:00

Joshua Leung commented

2019-02-03 13:05:19 +01:00

Closing as "TODO" - a known limitation that the tools were not intended to support, but could in future when we have the time to do so.

These tools were not intended to take the NLA stack into account. Furthermore, layering/additive behaviour in the NLA has only been properly supported recently, is beyond the scope of the current intended uses of this tool.

Closing as "TODO" - a known limitation that the tools were not intended to support, but could in future when we have the time to do so. These tools were not intended to take the NLA stack into account. Furthermore, layering/additive behaviour in the NLA has only been properly supported recently, is beyond the scope of the current intended uses of this tool.

Ovionis commented

2019-02-03 17:57:10 +01:00

Added subscriber: @angavrilov

Ovionis commented

2019-02-03 17:57:10 +01:00

This comment was removed by @Phigon

*This comment was removed by @Phigon*

Brad Clark commented

2022-07-08 22:40:16 +02:00

Added subscriber: @BClark

Brad Clark commented

2022-07-08 22:40:16 +02:00

With the ability to evaluate the entire stack in tweak mode now, we need to retest this and see if it now works or can be made to work as it would be great to have the ability to use it with animation layer work.

Brad Clark commented

2022-07-21 05:47:25 +02:00

Yeah it would be nice if some of the tools could work or be updated to deal with the "visual location or blend to the final result pose and not just the key values but I am guessing it is much harder to do.

Ovionis commented

2022-07-21 18:44:08 +02:00

In #61099#1393465, @BClark wrote:
Yeah it would be nice if some of the tools could work or be updated to deal with the "visual location or blend to the final result pose and not just the key values but I am guessing it is much harder to do.

No, it is not.
I wanted to work with layers, so I went into the code and mimicked it in Python to create a breakdowner that does both and work with objects.
Took me around one or three days, and most of the time/effort was spent backtracking through the code functions, with me not knowing much of anything about C++. And also me trying to understand certain code that I actually didn't need and before I just started copying it verbatim. Like the code I was trying to copy was the specific math that the Combine layer uses with quaternions, and then I realized I mostly only needed the code from the tween operator(s).

Rather than keep trying to understand and mimic the layer math, I just go to the frames, resulting in my version being technically slower (not noticeably). Getting the layered value is functionally the same as getting the keyframe value. The only difference is there isn't any code available to do so in the Python API whereas there is such code internally (AFAIK). After getting the layer value, plugging it into the existing operator is simple.

Or another explanation:
The operators take 2-3 frame values (left, middle, right) and perform a generic slide between them.
All you have to do to make this work with layers is add their values in, instead of just using the current action.

Note: I believe my version was in sync with the internal one before the visual slider was added. I don't know if I could do that in Python as well and I haven't bothered to check, as it already does what I need it to do.....also I didn't release it. My point is that getting it done wouldn't be very difficult for someone that knew the language and wanted to get it done.

> In #61099#1393465, @BClark wrote: > Yeah it would be nice if some of the tools could work or be updated to deal with the "visual location or blend to the final result pose and not just the key values but I am guessing it is much harder to do. No, it is not. I wanted to work with layers, so I went into the code and mimicked it in Python to create a breakdowner that does both and work with objects. Took me around one or three days, and most of the time/effort was spent backtracking through the code functions, with me not knowing much of anything about C++. And also me trying to understand certain code that I actually didn't need and before I just started copying it verbatim. Like the code I was trying to copy was the specific math that the Combine layer uses with quaternions, and then I realized I mostly only needed the code from the tween operator(s). Rather than keep trying to understand and mimic the layer math, I just go to the frames, resulting in my version being technically slower (not noticeably). Getting the layered value is functionally the same as getting the keyframe value. The only difference is there isn't any code available to do so in the Python API whereas there is such code internally (AFAIK). After getting the layer value, plugging it into the existing operator is simple. **Or another explanation:** The operators take 2-3 frame values (left, middle, right) and perform a generic slide between them. All you have to do to make this work with layers is add their values in, instead of just using the current action. Note: I believe my version was in sync with the internal one before the visual slider was added. I don't know if I could do that in Python as well and I haven't bothered to check, as it already does what I need it to do.....also I didn't release it. My point is that getting it done wouldn't be very difficult for someone that knew the language and wanted to get it done.

Angus Stanton commented

2022-07-22 12:04:51 +02:00

Added subscriber: @ab_stanton

Angus Stanton commented

2022-07-22 12:04:51 +02:00

In #61099#1393742, @Phigon wrote:

In #61099#1393465, @BClark wrote:
Yeah it would be nice if some of the tools could work or be updated to deal with the "visual location or blend to the final result pose and not just the key values but I am guessing it is much harder to do.

No, it is not.
I wanted to work with layers, so I went into the code and mimicked it in Python to create a breakdowner that does both and work with objects.
Took me around one or three days, and most of the time/effort was spent backtracking through the code functions, with me not knowing much of anything about C++. And also me trying to understand certain code that I actually didn't need and before I just started copying it verbatim. Like the code I was trying to copy was the specific math that the Combine layer uses with quaternions, and then I realized I mostly only needed the code from the tween operator(s).

Rather than keep trying to understand and mimic the layer math, I just go to the frames, resulting in my version being technically slower (not noticeably). Getting the layered value is functionally the same as getting the keyframe value. The only difference is there isn't any code available to do so in the Python API whereas there is such code internally (AFAIK). After getting the layer value, plugging it into the existing operator is simple.

Or another explanation:
The operators take 2-3 frame values (left, middle, right) and perform a generic slide between them.
All you have to do to make this work with layers is add their values in, instead of just using the current action.

Note: I believe my version was in sync with the internal one before the visual slider was added. I don't know if I could do that in Python as well and I haven't bothered to check, as it already does what I need it to do.....also I didn't release it. My point is that getting it done wouldn't be very difficult for someone that knew the language and wanted to get it done.

So your operator still uses the keyframes to left and right, but adds in all the layers too? So this won't work if there are no keyframes? I wonder if it should work with no keyframes

> In #61099#1393742, @Phigon wrote: >> In #61099#1393465, @BClark wrote: >> Yeah it would be nice if some of the tools could work or be updated to deal with the "visual location or blend to the final result pose and not just the key values but I am guessing it is much harder to do. > > No, it is not. > I wanted to work with layers, so I went into the code and mimicked it in Python to create a breakdowner that does both and work with objects. > Took me around one or three days, and most of the time/effort was spent backtracking through the code functions, with me not knowing much of anything about C++. And also me trying to understand certain code that I actually didn't need and before I just started copying it verbatim. Like the code I was trying to copy was the specific math that the Combine layer uses with quaternions, and then I realized I mostly only needed the code from the tween operator(s). > > Rather than keep trying to understand and mimic the layer math, I just go to the frames, resulting in my version being technically slower (not noticeably). Getting the layered value is functionally the same as getting the keyframe value. The only difference is there isn't any code available to do so in the Python API whereas there is such code internally (AFAIK). After getting the layer value, plugging it into the existing operator is simple. > > **Or another explanation:** > The operators take 2-3 frame values (left, middle, right) and perform a generic slide between them. > All you have to do to make this work with layers is add their values in, instead of just using the current action. > > Note: I believe my version was in sync with the internal one before the visual slider was added. I don't know if I could do that in Python as well and I haven't bothered to check, as it already does what I need it to do.....also I didn't release it. My point is that getting it done wouldn't be very difficult for someone that knew the language and wanted to get it done. So your operator still uses the keyframes to left and right, but adds in all the layers too? So this won't work if there are no keyframes? I wonder if it should work with no keyframes

Ovionis commented

2022-07-22 17:43:24 +02:00

In #61099#1394009, @ab_stanton wrote:
So your operator still uses the keyframes to left and right, but adds in all the layers too? So this won't work if there are no keyframes? I wonder if it should work with no keyframes

The only limit to what goes into the 2-3 values is whatever the developer wants to go there. Where those values come from are wherever the developer wanted them to come from.
The internal developer wanted them to come from the current action's keyframes, and I wanted it to come from the current action's keyframes and the layers below.
If I (or a C++ developer internally) wanted, we could make the poses that are blended be completely random.

> In #61099#1394009, @ab_stanton wrote: > So your operator still uses the keyframes to left and right, but adds in all the layers too? So this won't work if there are no keyframes? I wonder if it should work with no keyframes The only limit to what goes into the 2-3 values is whatever the developer wants to go there. Where those values come from are wherever the developer wanted them to come from. The internal developer wanted them to come from the current action's keyframes, and I wanted it to come from the current action's keyframes and the layers below. If I (or a C++ developer internally) wanted, we could make the poses that are blended be completely random.

Brad Clark commented

2022-08-16 23:57:09 +02:00

" My point is that getting it done wouldn't be very difficult for someone that knew the language and wanted to get it done."
Great thanks for the update, good to hear.

" My point is that getting it done wouldn't be very difficult for someone that knew the language and wanted to get it done." Great thanks for the update, good to hear.

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Pose In-Betweening doesn't work correctly with animation layers #61099