# Before Geometry Nodes and the Beginning

* Was hired by Blender with the idea in mind that I'll be working on the everything nodes project.
* Started out doing bug fixing for 2.8 for a couple of months.
* Beginning of 2019 started thinking and writing about the Everything Nodes project. (https://wiki.blender.org/wiki/Source/Nodes/EverythingNodes)
* At that point the idea was to build a general "functions" system that allowed users to build node systems with can be used/executed in different places in Blender (modifier, drivers).
* Wrote more wiki posts about:
* Splitting up the modifier stack.
* Mesh type requirements.
* At the time my work was still more theorical, the overall design for how to integrate this in Blender was not there yet.
* At some point we wanted to start an actual project and had to decide between node based procedural modelling or a new particle system.
* After some back and forth we ended up going for a new particle system because:
* Movie production really struggled with the limitations and oddities of the old particle system.
* Procedural modelling was nice, but was not as immeditaly necessary for the studio.
* Also, we really wanted to get rid of the old particle system (in fact, during the development of Blender 2.8, it has been removed completely at some point, but has been brought back).
* So I set out to design and implement a node based particle system, which was also fairly new territory for me.
* During the development it was mostly just me trying to come up with a particle system design.
* In the process I iterated on the design a couple of times.
* In each iteration there was a wiki post and some discussion in a devtalk thread.
* In September 2019 I presented the system to the Blender Studio team.
* They were able to create some cool stuff: https://photos.app.goo.gl/PgCVGGaUqMuMJmKC6
* While fun to use and powerful, over time it became more clear that the system does not provide solutions for the needs of the team for better object scattering (which turned out was what they mainly needed a replacement for).
* Interestingly, that part does not actually need simulation generally.
* So that was actually out of scope of the particle *simulation* project I have been working on.
* Development went on for a couple of months.
* Some of the work was actually reviewed and merged to master already.
* We still use parts of that today (c++ stuff, tbb wrappers, multi function evaluation, generic float2/float3 attribute types).
* In September 2020 we had a workshop at the studio in Amsterdam.
* Particle nodes project was put on hold for now until the design is approved by more people.
* Decision was made to change the original decision and to work on procedural geometry first.
* This has a more obvious integration path into Blender and can serve as a basis for future particle nodes.
* We noticed that the problem here was really a fundamental miscommunication about what people meant by "particle system".
* To avoid running into the same problem again we assembled a team containing developers, artists and a product owner.
* This is when the geometry nodes project started.

-------------

**Title slide**

Hello, welcome to our presentation about the development process in the geometry nodes project.

**RRR slide**

Since the beginning of the project 2 years ago, we've come a long way. We've seen lots of adoption in all kinds of projects around the world. One of the most recent highlights was the use of geometry nodes for set dressing in RRR, a recently released epic action drama film made in India, which is actually the most expensive Indian production to date.

**Team photo slide**

Getting to this state however, was not a straight forward process. Today we want to tell the story of how geometry nodes came to be. Throughout the presentation you will see multiple cases where we had to pivot and reevaluate old assumptions. On the photo you see the core team behind geometry nodes during one of our design sessions when we all met in Amsterdam. From right to left there is Simon, who is our main artist stakeholder. He is usually the first actual user of many of the things we develop and provides valuable insights into the problems that arise when using geometry nodes. Simon also gave a presentation earlier today, which will be available online. Then there is me and Hans who are the main developers on the project. We are mainly responsible for the technical design and implementation into Blender. On the left side there is Dalai who is our product manager. His main task includes making sure that our priorities and designs are agreed upon and in line with actual use cases and with what other stake holders like Ton have in mind.

**Outline slide**

Before we even got to that team there were the dark ages, the time before geometry nodes. I'll start with talking about the times that led up to the current team. Afterwards Dalai and Hans will talk about the time from there on until today and into the future.

**Where does it begin slide**

It's hard to find a good time where to begin the story. The earliest mentions of the everything nodes project in Blender that I could find are about 10 years old. The idea of the everything nodes project was that many more parts of Blender would become node based, besides just the compositor and materials. There has been work on particle nodes as early as 2012 by Lukas Tönne. However, that never reached a releasable state and work on it was eventually stopped. During that time I happened to start developing the Animation Nodes add-on for Blender, however that add-on was never supposed to be included by default. A new hope for a native everything nodes project was born when I was hired by Blender by the end of 2018. For the first few months I only worked on bug fixing and helping to get things ready for the 2.8 release.

Eventually the time had come that I gathered enough experience in the Blender source code that I could start working on the everything nodes project. There was no concrete plan what to work on exactly yet. So I tried addressing various general design questions like what is the future of the modifier stack and what features does a mesh data structure need to make it usable in a procedural context. I wrote various posts in the Blender wiki about these topics but that in itself didn't really lead to usable Blender features.

A few months into that we decided that it was time to build something more concrete, but we didn't know what yet. The two main topics we considered were either procedural modelling or a new node based particle system. The decision certainly was not easy. Procedural modelling would have been great and was definitily more within my comfort zone. However, the Blender Studio had no immediate need for it. On the contrary, they did struggle a lot with the old particle system and were really looking for a replacement. At the same time, other developers were also very keen on removing the old particle system for a long time. During the development of Blender 2.8 the entire particle system was actually removed at some point, but was brought back when it became apparent that no proper replacement will be available soon. Based on these arguments it seemed very logical to start working on a new particle system.

**Particle Simulation project slide**

So I set out to design and implement a new node based particle system, which was fairly new territory for me. Coming up with a design was obviously the tricky part. The process was fairly iterative. I started out writing a wiki page describing the design and then asking for feedback on devtalk. Based on feedback and my own experience with the early prototype I updated the design a couple of times, each time in the form of a new wiki page. The process appeared to be working well, people had fun using the system and were creating many different effects that were impossible to do with the old particle system.

**Particle system examples slide**

Shortly before I was going back to Berlin to work remotely, I presented the current state of the project to interested artists and developers at the studio. That was about one year after I started working for Blender and four months after starting to work on particle nodes. The team was able to create some cool looking effects in the 90 minute session.

Work continued as before for a couple more months. In February of 2020 we had a few meetings with Brecht and Dalai and created a plan for how to get particle nodes into the master branch. In the months following that more and more parts of the particle nodes were merged into master as an experimental feature. It all looked like we were going to have particle nodes soon.

Everything changed in September of 2020 when I got to Amsterdam again to continue design discussions. When I was there the project was somewhat surprisingly put on hold. That mainly happened because we could not agree on a high level design anymore and, more importantly, the implementation did not address the actual needs from the studio.

**Planning slide**

Besides that, Dalai and others started to notice that this is a more general problem with Blender development, so it's not unique to the particle nodes project. Looking at the planning for 2020 we can see that there are many different things going on at the same time. Especially note that each of these boxes has one developer assigned to it who is supposed to work on that project. This leads to developers working in isolation leading to features that are disconnected from actual use cases. For example, note how the particle nodes and hair project were both planned for 2020 but never really happened.

**Recap slide**

Knowing all of that now makes it easy to pinpoint the issues of the development process, but at the time it wasn't all that obvious. The interesting aspect is that communication between me, the community and artists at the studio did happen but not in a way that results in features that help Blender Studio projects.

**Pivot or Die slide**

Now we were in a somewhat difficult position, since we stopped the particle nodes project but we also didn't really have any idea for what to do next. We could either come up with a design for particle nodes that everyone is happy with or we could do another project instead. This is our first big turning point where we reevaluated our assumption from the beginning about whether a new particle system or procedural geometry would be a better first goal. The particle nodes project is on hold ever since.

**Andy interview slide**

In order to figure out how to do things better, Dalai interviewed Andy for what he needed most in past open movie projects. Surprisingly, that showed that many of the issues came from a fundamental misunderstanding for what different people meant by "particle system". To me, that always referred to particle simulation, so particles moving through space, reacting to events etc. However, Andy was mostly looking forward to a better scattering system which does not really need simulation at all and therefore is a very different project.

With all those learnings in mind we wanted to do it better the next time. Now I invite Dalai to the stage to talk about how things continued from here.

**Assemble team slide**

- title:
- welcome
- rrr:
- many users
- highlight rrr, most expensive indian movie
- team:
- simon: art, hans and me: devs, dalai: product owner
- outline:
- time before geometry nodes
- until now + future
- where it begins:
- everything nodes
- 2012 particle nodes
- animation nodes
- hiring me
- 2.8
- everything nodes
- particle system:
- particles vs procedural modelling
- iterative workflow, proposals with feedback from community
- examples:
- presented to team
- fun to use
- created things impossible before
- things continued for some months
- started merging to master as experimental
- couldn't agree on design anymore
- put on hold
- planning:
- similar issues in other projects
- many different projects at the same time
- usually one developer assigned per project
- leads to isolated work
- recap:
- in hindsight it's easy
- during the process it felt alright, because we did have a couple of workshops, there was community involvement etc.
- somehow the type of communication was wrong and did not lead to features the studio was actually looking for
- pivot or die:
- difficult situation
- no project
- first pivot: reevaluate initial assumption about particles vs procedural geometry
- andy interview:
- find actual present and past use cases of the particle system
- most use cases where related to scattering objects
- surprisingly revealed a fundamental misunderstanding
- with all of that in mind we wanted to do things better the next time
- now dalai takes over