วีดีโอ

Could this be used for graphs representing models other than space graphics? Let's say we can represent a musical scale with an input graph. And from it we create primitives -> tree -> grammar and form new "words" which could be interesting new scales? Of course as long as the input thing can be represented with a graph btw great stuff. I'm an undergrad student that just got into stuff like this and am following your model synthesis paper to create a comprehensive opensource extension for godot engine so just saw this new thing from you and honestly cannot follow it that well yet, so sorry if the question seems stupid....

This is so funny. Truly a hidden gem!

amazing research. adding physical constraints per different materials will be more useful

Thanks for this video I just started writing a report 3 days ago for the final submission of a year 3 subject of my undergraduate degree based on your Example-Based Procedural Modeling Using Graph Grammars paper and the overall algorithm. Its really interesting stuff this video did a good job explaining some of the concepts that have been taking me a while. I have been trying to code my own version of some of these steps as part of my learning process I have been able to create primitives from shapes as that's relatively straight forward but I feel like I am missing some step when it comes to how you label faces and therefore get the left and right values for your edge labels. One point of confusion for me is figure 3 a where you show labeling of faces but I am confused as to why the faces that a bar intercept are both labeled/coloured as 0/white and are not their own colour. This might be a a common method that I am just not familiar with but I would apreciate any pointers if you have any. I have also subscribed to your source code emailing list and greatly look forward to you releasing something to play with when you do.

this is awesome.

amazing

Very cool

Fantastic work

in the 3d model case, how do you decide where to cut the model to generate the primitives? both in terms of position, orientation and scale. In the example at 15:38 there seems the choice seem to follow a grid

WAO

Have you thought of turning this into a product like City Engine by Pascal Mueller?

I came across this orbifold notation before. Thank you for illuminating them.

Nice and succinct.

I love that you're using differential geometry to determine which primitives are valid candidates for each other. Takes me back to the turning a sphere inside out days!

Should put the enterprise into this to generate new starships!

Corrections: (1) A basketball actually has a 2*2 symmetry. (2) The defining characteristic of the "X" is really the orbifold has a mobius band / cross cap, not the number of glide reflections.

Really appreciate the effort that went into this!

9:30 two times two too many two's

This video rules. It’s funny how the math of symmetry is so interesting, yet symmetry itself looks boring in reality. Well, symmetry is still important, but if we show perfect symmetry, f.ex the frontal view of the human face, it looks boring.

Bro I've been waiting for months for another Paul Merrell video

Another Paull Merrell video 🎉

Insta-sub, thanks for your great work!

how did you manage to do so much visuals? It’s incredible.

I would say this is one of the best game trailer I've seen.

One even could say, that C. Alexander pioneered the concept of "generative grammar" in relation to architectural/geographic auto-modeling, which the methods you describe elaborate further...

It makes sense to call it "grammar". It is common to speak of design "language", or per Christoph Alexander "Pattern Language" related to Architecture. Makes perfect sense to me to invoke "grammar" to refer to the underlying combinatorial logic.

Hello Paull Merrell, I really like your algorithm and its research paper. Do you think it would be possible to use a neural network to carry out the task of constructing the output model, an AI capable of determining the adjacency constraints and randomly drawing one of the possible labels. And also could this AI model be more efficient in terms of execution time and memory costs?

Beautiful : }

Am I correct in my intuition that using parallel lines from a non-periodic tiling would be completely impossible? By its very nature I think you would run into the same precision problems you mentioned in the arbitrary points version but to be honest I'm not sure. I guess the reason I was thinking about that is because if you limit the angles of the arbitrary shapes, in any configuration they would exist in exactly that configuration at some point in the non-periodic tiling. Well any specific set of vertexes would at least. But as it's an infinite grid you couldn't find that place without rounding something. So taking from that thought, if you had a tiling that repeats every 16 parallel lines in all directions then you could "round" the arbitrary shapes onto the psudo periodic grid and get exact values of that shape now that is is locked onto the grid. That's probably not the best explanation and I don't know what I talking about but it's very interesting. At any rate thank you for that well made video and the work you have done. Every time I see wave function collapse in the wild it makes me happy. Good luck.

Do you have a discord channel?

Why bro sound like 3blue1brown lmao 😮

hum... i have to code a bit and test, but i think this has some application to my problem...

Question, i see alot of this being applied to map creation but could you apply it to things like enemy creation, character creation, etc How granular of an application space does it have? How much would it have to be changed to work in applications outside of map/level creation?

Nice to finally meet you! Great research, thank you.

You work is very inspiring! One additional think you could work out in the future is space proportions. Either inside or outside of your generated objects become too thin. For example too narrow spaces between skyscrapers, but there could be example almost in every object you presented. If you could obtain the constrains of minimal space and then generate with them, final results would much more believable. Thank you for your work and this presentation, I hope to use this knowledge in the near future.

your work is amazing , your results are impressive

your work is amazing

Your method is very ingenious and it definitely seems like it worked out most of the positive/generative part. All that really remains is figuring out how to model the constraints/rules that steer/converge the generative part into something desirable. Keep up the good work! You're definitely on the right track to something incredibly useful! In Control Theory there are the dual concepts of Positive and Negative Feedback, all you need is the negative/regulating part.

very cool!

This is awesome! I am curious how this compares to the 3d performance of the model synthesis algorithm (I think this was NP hard and required back tracking), and if you could compare the two algorithms and the models they can create. A lot of this still goes over my head - but I still find this fascinating for game dev. Thanks for the content!

Impressive work!

is there any example of the style template out there?

fastantic, similar to matthias mueller's videos on XPBD, "mysteriously underviewed". championed proceduralism in music dsp for a wile. first ? lyrical music generation 1994 (a day per piece to assemble to tape). i love how WFC can be communicated as "assign every possibility to every instance and reduce until soluted" takes 40 minute videos to "divulge" along with a bunch of crypto hogwash. it's why no one in dsp likes me, i'd basically punch everybody. the whole lodge. generation, make it up.

i am gonna try implementing it in bifrost.

seems like a very interesting concept, unfortunately its not supported on latest android devices.

This is great stuff. It reminds me loosely of a class I took at UC Berkeley's School of Environmental Design back in the late 1980's taught by Horst Rittel. It was more of a seminar on Design Theory, actually, but not what I was expecting when I signed up. I expected the material to cover famous Architects and their philosophies/theories of designing buildings and so forth. After all, you generally look to the range of historical precedents for inspiration. But, Rittel had a completely unconventional approach (at the time). He described a method for solving design problems through using a "morphological box" that would essentially list all of the various possible design elements in an array of rows stacked vertically. So, one row would list possible floor plans, another would list window styles, another would list roof profiles, etc. Then, the designer would essentially run through and evaluate all possible combinations, discarding the ones that were non-sensical, and grading the ones that could work. Presumably, since all possibilities would be considered, the best options would percolate to the top. At the time, as a student of design, I had a real problem with this approach, not only because it seemed incredibly tedious for all but the simplest of design problems, but also because it ignored artistic inspiration and individual creativity. But, in retrospect, I think Rittel was simply decades ahead of his time. With the power of modern computers, and clever algorithms such as yours, I suspect approaching design in such a way is more practical than ever, even for complicated scenarios. I think the type of automation you are investigating is going to be a game changer in a variety of design fields very soon. Providing mechanisms for a designer to influence the automation, directing it in some manner, seems to be a key element that could make this kind of thing very useful. I'll be honest, I thought Rittel was kind of out of his mind. His approach would be so burdensome to implement on paper for a real-world situation, that I felt a little hostile toward it; I mean, it was almost absurd. Nobody in Architecture would actually work that way. But, now, with the computer at hand to do the heavy lifting sorting through the variations, it makes a lot more sense.

really inspring stuff!

Just remember to use your powers for good instead of evil.

很棒的算法!

here: future matrix architect