BCON2026: AI-Enabled Tools & Workflows SIG

General Forums Blender and CG Discussions
1

Preface/disclosure: Hi everyone! I’m Oscar and I’ve got almost 2 decades doing all the Blender things. Wrote the book on Blender, brought Blender workflows to Pokemon, and I made the Seattle Blender User Group. For the last 2 years I’ve been working on a mesh reconstruction suite with www.unrealizex.com where we make 3D assets in as little as 1 credit, 1 image, 1 minute. Our process doesn’t use generative AI, but this work means learning new AI workflows and engaging with the nuances and concerns of AI. I show some of this work, but it’s not the focus of the talk.

Here’s my talk from the 2026 Blender Conference, the Blender + AI-Enabled Workflows SIG. The SIG covers some broader AI literacy, Blender’s history with AI, the current guidelines and projects, and also reframing this discussion through the lens of “The Blender ethic.” Namely: Software freedom, project sovereignty, open culture, and our history as technological tinkerers.

More personally, to this specific forum: The Blender community doesn’t agree on everything, but I do think we gravitate together historically because of a shared ethos around software freedom, user and foundation sovereignty, hackers/tinkerers, and creative freedom. AI doesn’t dictate that, contradict that, or supercede that.

I hope this has some useful information for you, and true to the SIG format, I’d love to hear your answers to the questions!

2

What does this mean?

3 Likes
3

It completely contradicts that, and Blender’s relationship with Anthropic was the final nail which closed the door on my donations.

You talk about freedom and sovereignty, yet ignoring the countless artist and musicians and authors who have had their work stolen by the AI development community.

3 Likes
4
  1. Per the talk, Blender’s had computer vision in it since 2011.
  2. Depending on who’s drawing the semantic line in the sand, anything related to anything photoscanning is, technically, AI. Which highlights how some of our societal difficulties with AI isn’t related to the technology, but the more recent society-disrupting skyrocket in technology.
  3. If you are working in the tech sector, the technology has become ubiquitous. So while there are alarms about AI displacing jobs, there are also economic realities of many jobs now requiring AI usage in production. So, yes, been using Claude at work, and learning AI workflows discussed in the vid. AI’s a big subject that, like I said, can hog all the oxygen in the room, but as a bare minimum of honesty. One takeaway from the conference for me was AI use in professional Blender workflows are much more frequent these days, but due to the societal debate around it, we struggle to embrace a Blender-style open and honest dialog where we share workflows. No 2 people using Blender+AI are approaching it in the same way, but there’s a stigma that prevents us from sharing and learning about it with each other, when normally Blender is all about sharing and learning with each other.
  4. I still gotta show why we’re NOT generative AI! Will do so on the other thread too. But in the short term, I actually think our errors are the best visual evidence for our lack of generative AI. If you use Clone and upload pics from multiple angles, you get a more accurate triangulation. But what about when you use just 1, where does the info on the back come from? Hence: Florg the All-Spine. This was made using experiments investigating our errors when I accidentally included a pic of a bird in a cat dataset and got a weird mutant. So Florg was made to push that: I uploaded a horse, a dog, a cat, a spider, a baby, and a human, and it made this result. Take a look for yourself, but the answer’s there: we are reconstructing from information in the inputs, even when that information is incorrect or compromised. https://unrealizex.com/s/2i4btcobxm/
5

That’s a lot of words, and I appreciate the depth of your response, although I think breadth would be a better term because your response doesn’t actually say anything. As I mentioned elsewhere, in order to create a 3D object from an image, something has to generate data that is not included in the image. You don’t explain what that something is, you just say it’s not generative AI but it’s also AI related. Honestly, it’s hard to take that at face value, especially because the website is entirely AI generated.

You’re welcome to keep your secrets, but if you want to have the open honest conversation you mention in your reply, you do have to do some of that yourself if you want other people to do the same

1 Like
6

Your website says, and I quote directly:

What you can do
Six tools, built around one job: making 3D models. You don’t call them — you just ask.
From a prompt
Describe an object and get back a textured 3D model — no reference image needed.
Connect the UnrealizeX MCP server to Claude, ChatGPT, or your terminal. Describe an object, sketch it, or hand over a photo — get back a textured 3D model without leaving the conversation.

This is what would broadly be understood as “text to model”, where a piece of text is converted by an algorithm into a 3D model. In order for this conversion to take place, the intent of the text must be parsed and converted to vertex data and colored pixels as texture on that vertex data. There is no algorithmic connection between the word “cube” and the following .OBJ data:

# Blender 4.4.3 Goo Engine
# www.blender.org
o Cube
v -1.000000 -1.000000 1.000000
v -1.000000 1.000000 1.000000
v -1.000000 -1.000000 -1.000000
v -1.000000 1.000000 -1.000000
v 1.000000 -1.000000 1.000000
v 1.000000 1.000000 1.000000
v 1.000000 -1.000000 -1.000000
v 1.000000 1.000000 -1.000000
vn -1.0000 -0.0000 -0.0000
vn -0.0000 -0.0000 -1.0000
vn 1.0000 -0.0000 -0.0000
vn -0.0000 -0.0000 1.0000
vn -0.0000 -1.0000 -0.0000
vn -0.0000 1.0000 -0.0000
vt 0.375000 0.000000
vt 0.625000 0.000000
vt 0.625000 0.250000
vt 0.375000 0.250000
vt 0.625000 0.500000
vt 0.375000 0.500000
vt 0.625000 0.750000
vt 0.375000 0.750000
vt 0.625000 1.000000
vt 0.375000 1.000000
vt 0.125000 0.500000
vt 0.125000 0.750000
vt 0.875000 0.500000
vt 0.875000 0.750000
s 0
f 1/1/1 2/2/1 4/3/1 3/4/1
f 3/4/2 4/3/2 8/5/2 7/6/2
f 7/6/3 8/5/3 6/7/3 5/8/3
f 5/8/4 6/7/4 2/9/4 1/10/4
f 3/11/5 7/6/5 5/8/5 1/12/5
f 8/5/6 4/13/6 2/14/6 6/7/6

The translation between human language and numerical matrices is done by neural networks. In common parlance, neural networks with a human language interface are referred to as LLMs, or more generally, as generative AI.

There are two possibilities here.

One- you’ve discovered an algorithm that can interface human language with spatial data matrices without using LLMs, in which case, you’ve just solved the entire economic problem facing the AI industry- a multi-trillion dollar problem, at the very least.

If that’s the case, why would you waste this Earth-shattering discovery on a paltry subscription service when you would be the richest man in the world overnight if you sold it to any of the large tech companies? And how have you made such a discovery when teams of tens of thousands of researchers have failed, after burning billions of dollars, to do so?

Two- you’re using LLMs and you’re attempting to obscure that fact. However, you’ve clearly stated that, and I quote: “I still gotta show why we’re NOT generative AI!” followed immediately by “… our lack of generative AI”, and, most clearly- “Our process doesn’t use generative AI”. So, how do you answer the questions posed above since this option is not possible per your statements?

To restate those questions in one clear query:

  • How have you solved the human-language to spatial data matrix challenge without using LLMs when teams of tens of thousands of the technology field’s greatest researchers have failed, after burning billions of dollars, and working on the problem for decades, to do so?
7

As i said in the other topic.. there is a very blurry “border” between “simple” neural networks to get some better noise reduction filter (which are not that simple) and the “haluzination” of areas without actual data in the original image (“learned” from data noody paid for).

So you might get some “customers” but also it’s a hard way to get some trust from artists. One question might be: Are uploaded images used for anything else ? And why should one trust you when you say: no.

Because this “unrealizer” is an online service and not some downloadable app (without conneting to the internet) :person_shrugging:

It’s some very hot topic one could burn their fingers one :sweat_smile: .

( Also your “errors” might be “proof” that you didn’t “collected” enough data yet.. you know ? That trust thingy :wink:. Sadly some very very big companies did ruin this badly :grimacing: .)

8

Annnnd that’s where I tune out. Like, that is a form of generative-AI. There’s a difference between a smart denoising algorithm and an LLM that churns out plagiarized text, or a diffusion model that subsists off of the stolen work of countless artists, or agentic AI that regurgitates lifted code and burns our planet alive. Programming, whether people acknowledge it or not, is an art form, and agentic AI like Claude is also causing harm in that sector.

Your third point literally contradicts this.

2 Likes
9

Cloning. Clone stamp. Think along those lines. Content-aware fill. Not AI, just sampling from the inputs.

Ever used content-aware fill in PS?

Do that, but 3d instead of 2d on a splat.

But IDK, startups gonna startup and we’re probably gonna close up shop soon :man_shrugging: If you’re playing with the free version enjoy it while it lasts. GPUs getting turned down soon, then off.

10

But clone stamping won’t give you the back of someone’s head from a face-forward portrait.

4 Likes
11

If you’re worried about losing clients and struggling to make your business work, why not try being honest and answering questions, instead of selectively answering only a few that you can spin how you want and ignoring the rest? I don’t care what you’re selling, the way you are responding to posts here would not make me want to buy anything from you

3 Likes
12

I am honestly not sure where to start, or rather, finish…? Classical heavy PG compute + gaussian splat scientific developments + contentaware fill but do it in 3D instead of 2D + look at the 3D model results (good AND bad), pretty darn similar to 2D content aware fills (good AND bad) = some people I talk shop about this to look at it and go “yup, as someone familiar with all those components, now I get it.” I am getting the sense here that this is more of an issue where I’d need to onboard you more about those individual ingredients, rather than the sum total of the ingredients.

“Where did it get the back of the head?” Why does content-aware fill put new grass in the grass spots and new sky in the sky spots and new trees in the tree spots? Except when it messes up and puts the dog in the bench spot?

Willing to explain it more if it’s still not making sense.

13

You keep throwing around highly technical terms like you’re trying to come across as smarter than us. You’re not. We know what these things mean and how these systems work.

How about less grandstanding and more honesty, and you explain in 100 words or less how this isn’t generative AI when I ask your site for a “regal cat with snake body and lizard feet sitting in a throne made of broccoli with a scorpion crown” creates the following concept image and then makes a 3D model from it?

concept_image
concept_image1024×1024 980 KB

This is directly from your website as a result of a text-to-model prompt. And here’s the model:
Screenshot_20261015-191501
Screenshot_20261015-1915011080×2400 1.09 MB

Are you saying publicly and unequivocally that this image was not created with generative AI? That this image, which your website created, was not generated by a neural network or LLM? Is this your official, on the record, public statement?

The AI policy of this site has absolutely zero tolerance for lying about AI use. You need to explain in clear simple language how this text-to-model process is not generative AI

8 Likes
14

Ok, so you keep mentioning Content-Aware Fill, as a comparison point of why your tech isn’t AI.

How does ADOBE describe Content-Aware Fill?

In Adobe Photoshop, Content-Aware Fill removes unwanted objects by analyzing nearby pixels and filling the selected area to match the surrounding texture and color.

It duplicates texture and color, based on pixel proximity to the selection. You aren’t duplicating vertex borders with a prompt; you’re creating raw output from text description. Which is what most every other AI generative software does; sometimes with an input reference image, sometimes not.

Now, let’s see Adobe’s own words for Generative Fill.

Generative Fill in the Photoshop app now has the improved Adobe Firefly Image model. Generative Fill is a generative AI tool powered by Firefly that enables you to add and remove content from images nondestructively using simple text prompts to achieve realistic results.

AI, which sounds a lot like what I’m seeing here.

"Our process doesn’t use generative AI, but this work means learning new AI workflows "

So it doesn’t use AI, but to use it we need to learn AI workflow. Oooook.

2 Likes
15

@ogbog this is not a request. You are required to answer if you want to keep using this site. You have three choices:

  • Refuse to answer- you’ll be banned 48 hours from now
  • Answer explaining clearly how this image created by your site from a text prompt is not using generative AI- the staff will consider your arguments and discuss next steps
  • Admit that you’re lying and apologize for it- this post will be closed and unlisted, but your account will not be subject to further restrictions at this time
3 Likes
16

I’m gonna go ahead and give myself permission to comment on this.

Just because you don’t call something “generative AI” doesn’t mean it’s not generative AI. It’s AI. It generates. That’s what I call generative AI. So I would say that this is all generative AI. Personally, I would even go further and call it slop generator. I see this talk as a poor attempt to deceive.

AI technology has its uses and its issues, it’s here though, most of us use it in one way or another, but stuff like this is annoying. It’s just another attempt to make quick easy money. It’s not too difficult to take some open source model like Trellis and host it with an AI vibe-coded website. This is what this seems to be. And there are plenty of those already. Personally I would like not to see stuff like this on Blenderartists.org. It’s just a personal wish that I would like to express. There are so many of them that if this stuff is allowed here, it will soon be only this.

I don’t think that if you are an artist, AI should be forbidden for you. No, of course we can all use whatever we want, it’s just that in my opinion this has nothing to do with one actually being an artist. It’s a separate subject. One can be an artist and enjoy cooking for example. But just because of that I don’t think platforms like this should be filled with frying pan commercials.

I think Blender is a tool made by people for people to create. The idea of people creating is nice. All this generative AI stuff… OK, let it be. But it’s a separate thing. I think it’s off topic in the context of Blender. That’s just my thoughts.

9 Likes
17

Oh yeah, that’s definitely generative AI. That’s you, the user, using generative AI. This process is a printer, and your input is the only thing we derive 3D from; your input in, your input out. If you feed it a copyrighted picture of Spider-man, we spit out a copyrighted 3D model of Spider-man; if you tried to sell that piece of paper, or that 3D model, it’d be your copyright violation, even if we’re the printer in the middle. If you used generative AI somewhere else, and uploaded it as an image, the 3D inherits that image’s legal status. Feed it a CC BY-NC-SA image, we spit out a CC BY-NC-SA 3D model.

I ain’t going to say it’s easy to navigate. Steam sells games without AI and games with AI. It’s totally fine to refuse to buy the games with AI. It’s also totally fine to boycott Steam entirely due to their allowance for games with AI. I ain’t gonna be mad at you; news of the morning is Linux users switching to FreeBSD cuz Linus is on his AI BS again. Software freedom is a way of life, I ain’t forcing you to use this. The whole of the law shall be do what thou wilt.

The plan in the next phase was, in fact, to segment the two generative AI tools out (One-shot text and One-shot sketch), which was my own early recommendation when working on this project. Pretty much all the issues you guys are raising came out of my own mouth. A process that doesn’t use generative AI is less appealing when there’s a button that says “but if you DON’T care about generative AI, here you go.”

You’re not wrong, and I actually agree with you about segmenting or removing the generative AI modes. Rather, we’re just out of runway as a business to make those changes.

18

Yeah, a giant wall of text claiming that I the user am the one using generative AI when your site created this content wasn’t one of your three options. I gave you a chance to be honest and you chose to lie again. Locking this thread now. The moderators will weigh your case and come to a consensus

3 Likes
21

So your site/product doesn’t do Gen AI…except when it does. But that’s the user’s fault.

Or something.

The whole of the law shall be do what thou wilt.

Quoting Crowley is so edgy…

2 Likes