Blender 3.0

[niklebedenko]

First of all, I want to say a big thank you for spending the time to make this library - Blender is such an immensely powerful tool for making animations that it only makes sense to have a Manim-Blender fusion.

I've successfully installed this on Windows 10, Blender 3.0 (I did not replace userpref.blend and startup.blend, see below), and rendered the tutorial animation you put in the README - I've got a few questions that I was wondering if you could help me with:

1. After rendering the animation of text spinning, I had a look around at what Peeps had put into the Blender scene, and all that was present was the Bezier curves from the text - no keyframes for the animation. Is there any chance to have the keyframes added to the timeline, so that a more visual editing is possible? Or possibly using Peeps to generate the objects & keyframes, then using Blender's rendering GUI to do the rest?

2. I tried to re-render the animation after changing the text (with render=True as before), but no video was rendered, although the objects in the scene were updated. Is this intended behaviour / am I missing something? Saving the script / reloading a couple times didn't help.

3. I noticed in one of the GIFs on the README, you showed some sort of real-time updating of the script resulting in the scene changing. Was this achieved by immediately re-running the script with script_init(... , render=False) to only create the objects?

I would really appreciate some clarification - even better would be some sort of tutorial video showing you using the library to produce something simple.
Note: userpref.blend contains some of your user-specific output paths e.g. C:\Users\momin\AppData\Local\Temp\ so I didn't use it on my system. Also, I prefer to keep the GUI for Blender as vanilla as possible so I also didn't use startup.blend, since all it seems to do is change the GUI up a bit.

[niklebedenko]

I've just found out what I was missing RE. (2) - it seems that Peeps will only render the video again if there is not an existing video with the same name. I believe this is due to a number of lines in frame.py - specifically, in Frame.r() and Frame.stop(). Their implementations are designed to be robust against the same frame / video being rendered multiple times by accident.

However, this requires any existing video files to be renamed before another render is possible. I believe this could be fixed by checking for existing video files inside Frame.__init__(), then adding a numerical suffix to the file names of the new videos / frames if an existing video file is found.

Does this change sound good? Can submit a PR if you'd like.

[hmomin]

This actually sounds like a pretty decent idea, however, the reason I had implemented it the current way is because often, there'd be a bug somewhere in the code that didn't show up until the animations actually started rendering. Then, they'd break the script while rendering and I wouldn't find out until I woke up to check on everything. If I then fixed the bug and restarted the script, I wouldn't want it to spend all that time re-rendering all the videos it's already made - it would just use up unnecessary time and computational power.

Another reason is I'd often find that a certain animation didn't play out the way I expected it to. One might call these "silent bugs", in other words, bugs in the code that wouldn't break the scripts, but would render the animations incorrectly. Then, I could fix those bugs to render things properly and just delete the video(s) I didn't like and re-run the script -> it would then only try to re-render the videos I deleted, saving a lot of time.

[niklebedenko]

Thank you, that makes sense. As a new user, I found this was a little confusing - would there be scope for adding my suggestion as a default setting to prevent confusion, then have the option to disable it to have an accelerated workflow as you describe?

I suppose the most robust solution would be a caching system similar to ManimCE, which might both remove new-user confusion and enable the accelerated workflow.

[hmomin]

Yes, we could definitely add your suggestion, since it has its own usefulness in certain situations, while keeping the current setting as an option. I'm not too familiar with the caching system that ManimCE implements, but I'm also open to that as well.

[hmomin]

Glad you’re enjoying the library 🙂.

Thanks a bunch for testing it out on Blender 3.0; someone else recommended that I ensure it works on 3.0, so it’s great to know that it already does - I’ll have to add a note somewhere in the README.

I’ll have time to review any pull requests over the weekend, but I can address your questions now:

1. After rendering the animation of text spinning, I had a look around at what Peeps had put into the Blender scene, and all that was present was the Bezier curves from the text - no keyframes for the animation. Is there any chance to have the keyframes added to the timeline, so that a more visual editing is possible? Or possibly using Peeps to generate the objects & keyframes, then using Blender's rendering GUI to do the rest?

Yes, peeps just takes the Bézier curves that make up the Tex/Text and treats them like objects in Blender. I don’t remember why, but when I was first building out the library, I tried to use programmatic keyframes, but found them to be lackluster. It might have been that the output videos were not high quality or maybe something else. If you’re interested in more granular control over a rendered animation, I’d recommend you take a look at the advanced sample script where we use the f.video() context manager. In principle, you can use that to make any kind of animation you like involving the letters in the Text object, including morphing, rotating, etc.

2. I tried to re-render the animation after changing the text (with render=True as before), but no video was rendered, although the objects in the scene were updated. Is this intended behaviour / am I missing something? Saving the script / reloading a couple times didn't help.

You’re right, it’s intended behavior - usually, I use render=True whenever I’m done writing all the scripts and want to render out the videos, as in I just set render=False at the top of the script to render=True and sleep while the computer gets chugging. When working with peeps, I would rarely render out videos as I was writing scripts, since the rendering takes too long.

3. I noticed in one of the GIFs on the README, you showed some sort of real-time updating of the script resulting in the scene changing. Was this achieved by immediately re-running the script with script_init(... , render=False) to only create the objects?

Yes, it's just reloading the newly edited script (Blender doesn't automatically update the script if you're not editing it in its native text editor) and running it to show the end state of the objects after the animation takes place.

I would really appreciate some clarification - even better would be some sort of tutorial video showing you using the library to produce something simple.

A tutorial video sounds like a great idea, I was actually planning on making one a while back to help promote the library and clarify some things that aren't so obvious over text - just been quite pressed for time lately (aren't we all? 😂)

Note: userpref.blend contains some of your user-specific output paths e.g. C:\Users\momin\AppData\Local\Temp\ so I didn't use it on my system. Also, I prefer to keep the GUI for Blender as vanilla as possible so I also didn't use startup.blend, since all it seems to do is change the GUI up a bit.

The main reason userpref.blend is included is to provide keyboard shortcuts for automatically reloading/rerunning the script. With it, you can use any text-editor/IDE to write scripts (I use VSCode, myself) and then hover over the scripting window in Blender and use Alt+R to reload the script and Alt+F to run it. You're right that startup.blend just changes the GUI layout to one that I settled on myself after I had been using peeps for some time.

[niklebedenko]

Thanks for your reply, very helpful. A couple of follow-up questions:

1. The workflow for checking the final state of a scene is to re-run the script with render=False, and the GUI will be updated with a static image. But how do you check the animations? Do you have to create a render and then watch the video?
2. The first example in advanced sample script ball_shift_by_path() seems like it could be massively simplified with the use of Blender's built-in functionality - i.e. Keyframes + Move along path constraint. Plus, using Blender's built-in functionality for animations would remove the need to wait for a render to check animations, since the viewport renderer could show you them in real time. I've been experimenting with this myself, and I've found programmatic keyframes to be a viable method for several simpler animations (translate / rotate / scale, fade, morph) - is there any scope for this being supported, or is there a fundamental architectural difference which prevents this?

[hmomin]

1. The workflow for checking the final state of a scene is to re-run the script with render=False, and the GUI will be updated with a static image. But how do you check the animations? Do you have to create a render and then watch the video?

Yes, unfortunately, you'd have to create a render and then watch the video to know for sure how it's going to be animated. The GUI will show you the end state of the animation and the starting state is whatever was in the GUI before. Over time, I gained a lot of trust in the way the animations were being made, so I would perform hard renders less and less to save time. You can force any f.play, f.multiplay, f.superplay call to render by setting render=True just within that specific call instead of the whole script.

2. The first example in advanced sample script ball_shift_by_path() seems like it could be massively simplified with the use of Blender's built-in functionality - i.e. Keyframes + Move along path constraint. Plus, using Blender's built-in functionality for animations would remove the need to wait for a render to check animations, since the viewport renderer could show you them in real time. I've been experimenting with this myself, and I've found programmatic keyframes to be a viable method for several simpler animations (translate / rotate / scale, fade, morph) - is there any scope for this being supported, or is there a fundamental architectural difference which prevents this?

Wow, I had no idea it could be simplified that much 🤦‍♂️. I'm not a Blender expert, just familiar with it at a basic level, so I wasn't even aware of the "move along path constraint." This might also simplify a lot of the lambda function manipulations used to shift stuff along weird trajectories... Also, you're right, using keyframes would allow you to check the animations in real time.

There is probably some opportunity to incorporate keyframe-based animations somehow. I'm not entirely sure what that would entail and can't promise that I will have time to test/code this, but I can offer any guidance if anything existing in the library is confusing you. If you'd like to contribute code that works with keyframes, it may be useful to do it in an additive sense: in other words, create new methods that generate the corresponding animations using keyframes and eventually we can delete the old methods...

[niklebedenko]

Yeah it's super simple - Blender is awesome. I made this little video to showcase a couple of those animations. https://youtu.be/CmyS3OJT0yw
I'm in the process of making a little toolkit plugin which makes a lot of the Manim-style animations for you with just one click. Will let you know how that's going once I've made a bit more progress on it.

[hmomin]

Sounds good, that video is awesome - you've definitely got a lot of experience with Blender!