It goes to the issue of cash, time, and personnel resources.
Choice A: spending any of the above on AI integration
Choice B: spending all of the above on non-AI-related features.
I prefer Choice B.
It is important to highlight that this is highly your opinion.
The approach and methodology employed by users when creating graphics is entirely subjective and at their discretion. No one can dictate what constitutes the “correct” way to create graphics, as each individual’s creative process and preferences are unique. Whether you chooses to use AI to craft a stunning, epic scene with minimal input or prefers to rely on traditional tools that require a lot more time and effort is entirely up to you. The key aspect is that users should be free to pursue their artistic vision in the manner most enjoyable and fulfilling for them.
With all due respect, I don’t understand this obsession with interfering with how others choose to create graphics. We are free individuals, we are free to use whatever tools available we find stimulating and joyful to use.
@joseph I propose a potential solution to address the recurring issue of “off-topic” discussions. Consider implementing a subtopics-feature to the forum, which would allow users to engage in parallel yet focused conversations that can be expanded upon by other participants. This could help mitigate the need for moderators to intervene and redirect the conversation back on track. Maybe it would be difficult / time-consuming to implement in terms of coding, I don’t know, but just an idea. 
2 Likes
To me, it sounds as if this is not about AI at all, but about features which you don’t think are useful.
The reason why I am getting this impression is because you did not reply to any of the actual points I made and why it would be more valuable to have them in Blender directly. You may not find those valuable, that’s fine.
To me, it makes no sense why you move the discussion to AI-based vs non-AI-based features.
Of course it is; that’s self-evident.
Because we’re competing for the same time, money, and resources. I want a country farmhouse, you want a city condo. There’s one builder, one stack of wood, and only 24 hours in the day.
Absolutely - people are free to create art with AI. That doesn’t mean I’m required to support improving their AI tools.
Yes, probably I wasn’t super clear …
If new AI tools emerge and become part of everyday 3D workflows, I’m pretty sure they’ll end up in blender as well. And it’s in fact already the case with AI denoisers.
Sure tools like :
Could be great to have, but they remain secondary tools for blender and not the main goal of the app, if we look at BF projects they tend to be more useful to many. Like asset browser can bring a lot of benefit for any users and in many places, an image upscaler while being great to have is much more a niche tool. Something like that is probably going to be a community contribution rather than a project paid by BF.
Finally these tools being AI based or not shouldn’t influence the decision of integrating them or not.
It’s just a more efficient algorithm to something that already exist.
But we get much more emotional about completely new stuff that might really change the way we interact or do CG. That’s probably these kind of stuff that @kkar had in mind when he raised concerns earlier. But at this stage it’s too early to discuss that , since (unless I’m mistaken) it’s nearly impossible to came up with a engineering plan for any AI based tool of that kind.
Something with a clear goal, and how much people are needed for how much time, and with of course a certainty of meeting at least an MVP of the goal at the end of the project.
Without these it’s clear that BF won’t invest in anything, to me for good reasons. Since they want to make the most results of their funding for the largest possible audience, which is IMO a good way to do for many reasons.
This is where the discussion starts to become speculative and probably not helpful here.
2 Likes
To understand each other better, perhaps we could make up a hypothetical situation and keep it simple.
Let’s say, hypothetically, that Blender 4.0 surprises us all with a full blown Text-to-3D-Object AI feature (developed in collaboration with Nvidia). You can now open a textbox inside of Blender, where you can for example type: “A donut with pink icing, high poly”, and a donut with pink icing pops up after a brief waiting period. It may not be perfectly generated, but good enough as a background filler or best viewed from certain angles.
I think that would be neat, and would probably play around a lot with the feature, make some cool stuff, and figure out ways to use it in workflows. Why would it, in your opinion, be bad?
Because I don’t have the skill, time or personal interest to model a high poly donut myself? If so, why do my personal goals or interests matter to Blender Foundation or anyone else?
Because I can model a donut with pink icing using the tools that already exist in the software. Thousands of artists can do the same. Artists such as myself would like BF to make the toolset more powerful in ways that benefit the actual craft of creating.
Your example benefits those who can’t do model at all, and despite all the resources available, have chosen to not learn how to do it.
SO, i want the BF to cater to people that actually want to create 3D, and not people who want a chatbot create it for them.
In a vague way, it makes me think of people who are really good at playing Guitar Hero on Xbox, and somehow a) believe they can basically play guitar and b) hang out with actual musicians, discussing the craft of writing music and/or c) trying to convince Stratocaster to create convert to building guitars with 4 buttons on the neck, or an auto-chord feature for the strings.
4 Likes
https://www.amazon.com/E-Z-Chord-10780307-for-Guitar/dp/B0002JKOZA … we’re almost there. roflmao
1 Like
Everyone will agree that having that for free in blender would be awesome. But nothing comes for free and having BF investing on these tools on behalf of something else is highly questionable.
Since the AI version doesn’t exist yet, it’s impossible to evaluate if it’s better than existing solutions, therefore if it’s a good idea that BF invest in these or not.
On the other hand, current projects have a clear goal and design, therfore they can be really discussed.
Do you mean something like an upscaler and the other examples I gave with that or are you referring to the text based functionality?
I don’t see why this should not be possible. Indeed, in my experience, it is even easier to do that with machine learning.
1 Like
I was thinking more about text based functionalities or some really innovative things AI could bring but are still in the research phase. Rather than an improvement over something that already exist, like the denoiser or an upscaler.
What you listed are great ideas that would make obviously useful tools, but still it’s questionable if BF should look into these compared to what they actually do, since it’s more secondary tools to me rather than big projects that have a greater impact on the software.
That said, if that’s something relatively simple to do I wouldn’t be surprised if someone from the community contributed such tools at some point.
Research is something BF shouldn’t do because of it’s uncertainty, and they also need to cherry pick what to work on to make the most impact out of their funding.
Whether what they implement should be AI based or not, I don’t see the point of debating that, whatever is simpler or better should be the answer there.
This, exactly. The Blender development team is small. I moved from Daz to Blender to make things. I don’t need a tool for lazy cheapskates who want pre-made assets without having to pay for them, commission them, or putting in the time to find Creative Commons/freebie options to modify.
And that’s what a lot of this AI discussion ends up being about, right? People don’t just want AI tools to make things easier like denoise or better auto-weights/quad-remesher or whatever: they want the program to make them free assets so they don’t have to learn to make it themselves, nor pay another artist who has.
Any time the Blender devs spend training a neural network to analyze and regurgitate re-mixes of other artists’ models is time not being spent on adding sculpt layers, improving high poly performance, adding realtime GI or anisotropic shading to Eevee, improving auto-weights, or UV unwrapping tools, or the texturing workflow, or rigging and animation and cloth tools and physics and nodes.
And that’s all those AI text prompt outputs are: remixes of other artists’ work. Because people want custom art, and they want it free and low effort.
People who just want a “make art button” can go use another software. I want sculpt layers.
2 Likes
What about speed / time? You could work faster if everything don’t have to be modelled by hand. This approach could also result in cost savings by avoiding the purchase of pre-made models from websites such as TurboSquid and other similar platforms.
Furthermore, even if you have the skill to model a donut with ease, it’s possible that your skill set does not extend to other types of 3D-objects, such as horses, dragons, cars, buildings, terrain, whatever else. Wouldn’t this hypothetical tool then come in handy?
In addition, for example, an artist may have a strong passion for modeling buildings and vehicles, but find vegetation modeling to be tedious and boring. In such cases, this hypothetical tool could prove highly beneficial. By leveraging AI to generate the “boring” elements, artists can focus on their area of expertise and creative expression, while still achieving high-quality results in their work.
Furthermore, Blender clearly aims to be a versatile and comprehensive tool, ranging from 3D-modeling, 2D animation, sculpting, simulation, video editing, etc. Why should AI generation capabilities be treated as an exception? Given the diversity of creative uses for Blender, it’s likely that other types of users may also benefit from AI features.
In my hypothetical scenario, the AI would be “good enough”, at least for background fillers or best viewed from certain angels.
And in reality, yes, we can not know for certain how powerful / useful such a tool would be. Is that the main reason why some people are against the idea of spending resources on generative AI for Blender?
I agree with you on that for sure.
1 Like
So… It’s about not paying artists? Turbosquid is the platform, it’s the ARTIST you’re not paying.
That’s the issue a lot of 3D artists have with neural network based generative art. It’s about undercutting artists who trained to make a living on their art, by instead training a neural network to study and then regurgitate re-mixes of their art to avoid paying the artist.
2 Likes
I have no debt to any artist, it is my hard earned money and I spend it on what I want (or not on anything at all if I so wish). Of course I would use a tool that does the job instead of paying someone else if it were possible.
However, I agree that it’s ugly and morally questionable to train AI on other people’s work without asking for permission first. But hypothetically, if no AI system did that, everything was trained with clear permission, would AI be fine in your opinion?
Yes, I think given the state of things it’s too risky for blender foundation to start working on such tools.
If such tools existed like many open libraries does (Alembic, USD, OCIO, OpenVDB…) , I don’t see the issue of incorporating that into blender at least as an addon. But again, it’s questionable if that should be done by BF or rather a community driven project…
Anyway, you made fair points about use of external resources, we don’t always have the time to model everything, and it’s ok to focus on other things than modeling, even being lazy can be a strength as a professional.
But to me there is already so much to do with all the free resources available. You don’t need an AI to generate ok-ish models to put in the background :
Free online ressources
https://polyhaven.com/models
https://archive3d.net/
https://3dwarehouse.sketchup.com/search/collections?order=popularity&direction=desc
So AI generated 3D models, since it’s hypothetical we can argue that produced models would be as good a the currently available one, but truth is, it’s hard to tell…
We can argue that AI generated model would be unique, they are also very likely to fail at reading our minds. I’d bet much more on procedural models to be tweaked to our linking.
If you are up to the challenge, let’s pretend I’m working on a damaged car model and I need texture for that, can you produce a few AI generated textures with actual tools to replace these ones ? :
This took me 10mn browsing my texture folder to find these, and in reality I’d spend probably 1 or two hours and look in the web too to get a lot more…
Thing is, the few time I tried I never managed to make something convincing and/or useful with an AI generator. Can you demonstrate that it’s a more efficient way of working beside the fact that it’s a very cool tool ? If it doesn’t work well with image how would it be for 3D models ?
Traditionally, this has been handled by paying artists who specialized in those boring areas for their assets. For a one time fee that ensures the developers Scatter/Botaniq/Grasswald/whatever can feed their families and continue producing art, you too can have a library of assets to keep your scenes lush and overgrown without having to model boring foliage yourself.
Or the Blender Foundation could train a neural network on those assets to regurgitate re-mixed assets off them so no one will have to pay human artists for that work ever again.
1 Like
why not go a step further … have the AI write blender so we no longer have to wait on human developers who need money to feed their familys. and then … AI’s to run AIBlender to create all the artwork… and then …
Where do I get tickets for the Axiom’s voyage? >grins<
(wall-e ref, for those that didn’t remember the ship name)
3 Likes
Well, it’s a never ending debate…
While what you say seems fair to artists, 15 year before it might sound the opposite : at some point there was only few external resource, and great chances that artists were payed each time there was a need for boring stuff.
But some people came in and made assets you buy one time so “no one will have to pay human artists for that work ever again.” , but nowadays nobody complains about that… why ?
I bet AI generators will find their place too, but that doesn’t mean it will work for everything.
Paid assets are great and it saves a lot of money on a project, again money that could have gone to local artist on that project …
Free resource are great too, especially if you are a student or hobbyist , generally what you don’t pay you compensate by time and energy, that why commercial projects will probably always rely on high quality payed material.
If AI can generate some stuff, there will always be talented artist to take these to the next level and sell them, isn’t it ?
All that said, external assets got their limitation too, on many projects I worked on there was no point in buying assets because the style of the show was so particular that everything needed to be modeled / textured manually …
1 Like