Blender 4.0 Released For This Incredible Open-Source 3D Modeling Software

[jacob]

Give me one blockbuster made with Blender since it’s used everywhere ?

I can cite hundreds made with Maya, Houdini, Katana. As of 2023, the biggest prod ever made with Blender was Netflix’s Next Gen (and the studio bankrupt since then).

So one again: "Blender is mainly designed for game development".

[DMJC]

Blender isn't the industry standard for film, and that's ok, it doesn't need to be. Blender is a great free 3D modelling/animation package. That's all it needs to be, it gets better every year and is becoming a useful tool for many people. Don't sweat the marketshare stuff. It really doesn't matter.

[nadirx]

Great release, but they still didn't clue in how a modern professional animation/vfx production works.

So many design decisions scream out loud the team never even spent 10 days at Pixar, Animal Logic, Weta or ILM to see what matters and what does not.

I'm curious about examples of this.

[kayosiii]

Blender is mainly designed for game development, where it has been wildly successful, but it *is* used in movies as well. AFAIK Netflix and Amazon use it for their own productions and Pixar has recently released Renderman for Blender, so there.

Blender is mainly designed as an everything in the one box solution for creatives on a budget. It's gained in the game space because this industry really expanded in the 2010s and the companies are generally younger and less entrenched in a particular way of doing things. VFX is more entrenched.

[kayosiii]

Great release, but they still didn't clue in how a modern professional animation/vfx production works.

So many design decisions scream out loud the team never even spent 10 days at Pixar, Animal Logic, Weta or ILM to see what matters and what does not.

I think that's an overstatement. Blender does have contact and contributors from within some of those companies. There are some well understood weak points for blender working in multi-artist environments and in multi-application pipelines, which is an area of ongoing work. On the other hand there are a lot of VFX artists have developed a workflow based on the quirks of their existing tools and call out Blender just because it doesn't replicate the workflow they expect exactly.

Ultimately, though the core blender team sees itself it's core audience as creatives on a tight budget looking for an everything in the one box solution rather than being a direct competitor to Maya/Max/Houdini/Nuke etc. It's also an open source project so the areas that get developed fastest has a lot more to do with developer individual enthusiasm for solving a particular problem than any long term strategic planning. The funding from the blender foundation does mitigate this to some extent but not completely.

Looking at the release notes for this release: The changes mostly look positive from an animation production perspective.
We get light linking.
We get node based tool and modifier design.
We get a lot of work on the pose library and asset sharing
We get improved view-port visualization quality.
and we get a major rewrite of the underlying data structures, making mesh construction much more flexible and performant.

I am curious as to why you think these features miss the mark and what you think they would be better of doing?

[rmfx]

I think that's an overstatement. Blender does have contact and contributors from within some of those companies. There are some well understood weak points for blender working in multi-artist environments and in multi-application pipelines, which is an area of ongoing work. On the other hand there are a lot of VFX artists have developed a workflow based on the quirks of their existing tools and call out Blender just because it doesn't replicate the workflow they expect exactly.

Ultimately, though the core blender team sees itself it's core audience as creatives on a tight budget looking for an everything in the one box solution rather than being a direct competitor to Maya/Max/Houdini/Nuke etc. It's also an open source project so the areas that get developed fastest has a lot more to do with developer individual enthusiasm for solving a particular problem than any long term strategic planning. The funding from the blender foundation does mitigate this to some extent but not completely.

Looking at the release notes for this release: The changes mostly look positive from an animation production perspective.
We get light linking.
We get node based tool and modifier design.
We get a lot of work on the pose library and asset sharing
We get improved view-port visualization quality.
and we get a major rewrite of the underlying data structures, making mesh construction much more flexible and performant.

I am curious as to why you think these features miss the mark and what you think they would be better of doing?

Some things that bother us and make it not suitable for highend prod are mainly:
no nodal render management, workflow is completely obsolete
no procedural material assignment neither and here too the workflow is crazy obsolete
no procedural object collections neither
fast performance hit (things are getting better though and direct competition is not much better)
global lack of advanced options everywhere, like anytime you try to do an advanced setup, you hit a wall
unconventional api decisions, that is an entire conversation by itself
unconventional colour management? Is it true they are dropping OCIO ? if yes wtf
completely wtf AOVs, no direct/indirect split, and no custom light path expressions neither to mitigate that issue ?
unconventional shortcuts for pro artists, they have an industry compatible set that it much better but left aside in docs
basic USD support, importing a crate is good, but USD is about composing USD files altogether, not just a file format
no Y up axis permutation possible, that's annoying for existing pipes, habits, and scripts.
no proper modelling history
inconvenient geo modeling hierarchy management, welcome in the 90s
not able to stack user interface editors with tabs, super annoying lack for work efficiency.

I criticize because I give them 5 euros tip a month since 2.80, and I never see these points tackled ever.
And doing a mini Houdini-like is to me, less fundamental than having a good modern rendering solution first.
That's for sure a good, even a great hobbyist software, but that list is mainly why no studios use it basically.

[kayosiii]

Some things that bother us and make it not suitable for highend prod are mainly:
no nodal render management,

as in render-farm management?

no procedural material assignment neither

are talking about assigning shaders to (parts of) objects right? what are the use cases for doing this as opposed to building the functionality you need at the shader level? what are you doing that can't be achieved with basic python / object nodes?

no procedural object collections neither

as in being able to select objects by a set of criteria? what's the use-case here?

no proper modelling history

This tells me that the problem here is that Blender doesn't behave like Maya. Blender doesn't have a modelling history at all, neither do most modelling applications. The tradeoff is a lot less underlying complexity(and fewer sources for bugs) and a lot less memory usage. Overall my experience with modelling in both applications is that Blender gives a much smoother overall experience with poly and subdiv modelling, but can't hold a candle to Maya for Nurbs. (I prefer both Blender and Maya to Max).

I criticize because I give them 5 euros tip a month since 2.80, and I never see these points tackled ever.
And doing a mini Houdini-like is to me, less fundamental than having a good modern rendering solution first.
That's for sure a good, even a great hobbyist software, but that list is mainly why no studios use it basically.

Being a mini Houdini-like is not the end goal of the new nodal interface. Several of the oldest subsystems in the application got to the point where they needed replacing, a node based system looked like a good way of providing the missing functionality and would allow a lot of the low level code to be shared between those subsystems, reducing the amount of work that had to be done. What's interesting about this release is that it allows node systems to be seamlessly integrated into the interface as a way of building 3rd party tools. Another point to make here is that developers aren't completely interchangeable, two subsystems in particular require specialized knowledge, the renderer(cycles) and the view-port (EEVEE), you can't just throw the developers working on other areas of Blender into working on these subsystems to speed up development time.

I would say that multi-artist workflow is more fundamentally important than advanced rendering options for now.

[rmfx]

as in render-farm management?


are talking about assigning shaders to (parts of) objects right? what are the use cases for doing this as opposed to building the functionality you need at the shader level? what are you doing that can't be achieved with basic python / object nodes?


as in being able to select objects by a set of criteria? what's the use-case here?


This tells me that the problem here is that Blender doesn't behave like Maya. Blender doesn't have a modelling history at all, neither do most modelling applications. The tradeoff is a lot less underlying complexity(and fewer sources for bugs) and a lot less memory usage. Overall my experience with modelling in both applications is that Blender gives a much smoother overall experience with poly and subdiv modelling, but can't hold a candle to Maya for Nurbs. (I prefer both Blender and Maya to Max).




Being a mini Houdini-like is not the end goal of the new nodal interface. Several of the oldest subsystems in the application got to the point where they needed replacing, a node based system looked like a good way of providing the missing functionality and would allow a lot of the low level code to be shared between those subsystems, reducing the amount of work that had to be done. What's interesting about this release is that it allows node systems to be seamlessly integrated into the interface as a way of building 3rd party tools. Another point to make here is that developers aren't completely interchangeable, two subsystems in particular require specialized knowledge, the renderer(cycles) and the view-port (EEVEE), you can't just throw the developers working on other areas of Blender into working on these subsystems to speed up development time.

I would say that multi-artist workflow is more fundamentally important than advanced rendering options for now.

You don't know what is nodal render management ??
Go check what is Katana, Solaris, or Gaffer...

The goal of procedural material and collections ??
Automated updates of scene content and operation through predefined naming convention... the base of a professional workflow.
If your have a shader that is assigned proceduraly to "/root/assets/vehicles/A*/door*" you don't need to reassign manually the shaders if the asset is updated with new meshes. Same goes with rule based collection that can grab content automatically without doing manual edits all the time.
In Blender you just can't... good luck with huge scenes with lots of updates but since nobody used it in big prods, I guess nobody knows what they're missing.

The rest: I don't have time.

[kayosiii]

You don't know what is nodal render management ??
Go check what is Katana, Solaris, or Gaffer...

I have been out of the industry for about 5 years and node is one of the most overused terms in computer science, even in the context of rendering it can refer to several different things.


It sounds like you are dealing with a workflow that has to compensate for data loss from importing files using bridging formats.
Is that what is going on here?

If your have a shader that is assigned proceduraly to "/root/assets/vehicles/A*/door*" you don't need to reassign manually the shaders if the asset is updated with new meshes.

Part of my old job was doing this sort of stuff in 3dsmax (specifically translating materials between different renderers), I have had a quick look and test of the current python API everything needed to do this sort of thing is there. If I were trying to set up something like this in my old studio, I would have a script read overrides from a text file with a naming convention and use that to swap out the materials.

Same goes with rule based collection that can grab content automatically without doing manual edits all the time.

Are we trying to import data from non Blend based sources here?

If you are saying that blender doesn't cut it as the last point in a multi-application vfx pipeline then I tend to agree with you. The much more likely place to see it in the context of a movie or TV production is as one of the apps creating assets that end up somewhere else.

[Panix]

Hey, anyone here with an AMD RX 7900 series card test their gpu in Blender 3.6 and 4.0 - either in Windows or Linux (I don't care which distro - although, it's probably easier with Fedora, Tumbleweed or something with most recent software versions?)? You need to use HIP-RT to get any worthwhile performance - so, that means using it in Windows for the moment....

I compared scores of AMD gpus - and Nvidia gpus - with Blender 4.0 - Nvidia cards have major regressions with the current version of Blender 4.0 (vs 3.6) and AMD cards have very slight regressions (although, the 7900 XT has a strange improvement in performance). Linux performance is generally better - by a decent (approx.) 3% at least - although, there's still regression with the latest version of Blender.

There's some 50+% increases going from using CUDA to Optix. I doubt AMD gpus will achieve anything like that - I'm just curious if a 7900 XT or 7900 XTX can match 3090 optix performance, at least, with working HIP-RT. Any chance? That's still not impressive but anything less is a major disappointment. HIP performance for the 7900 series is very close to CUDA performance (to the 3090 and 4070) according to the Open Blender data website.... I dunno if that will translate to HIP-RT (incorporating ray tracing) - I read claims of 25% improvement but that isn't much. It's nothing to boast about.