Max Liani

@Jiayin_Cao It’s the right technique. You let the door open to use SIMD to vectorize shader execution of sorted batches and get real performance improvements. I did some Siggraph presentations with Alex Welsh a while ago about this (when I was still leading the RenderMan XPU effort)

@maxliani Exactly what I did. I took some detour by making an attempt to SIMD the whole shading language first, only to realize that I had to introduce too many new concept to deal with divergency. The AD solution is quite neat. Though it has its problem, but it should be good for my use

After years of thinking about it, I finally got a reasonable approximation for mipmap selection working in my renderer. The biggest challenge, by far, was supporting procedural UV generation in my shading language. I believe I have enough content to put down a blog post next.

@kenpex @Jiayin_Cao One can still propagate usable derivatives using ray footprints.

@Jiayin_Cao @maxliani Is this for secondary rays? Wouldn't these need path differentials anyways?

@yiningkarlli So, more similar origin than I thought; both accelerated through by the crisis. The difference perhaps is that I went “rogue” and started the project on my own time, because Animal Logic wouldn’t have given me time off production to work on it.

@maxliani The original plan was like, maybe write a cool paper, and then see from there. Nobody expected Big Hero 6 (fall 2014) to made using Hyperion; the plan originally was optimistically either Zootopia (spring 2016) or Moana (fall 2016) might be the first to use Hyperion.

Maybe I see this in rosy retrospection, but I feel that the work I did many years ago to create the Glimpse production renderer still feels to me the most insane project I ever started.

@yiningkarlli Glimpse origin was more 2nd job startup like: I started writing the engine as hobby project, at night and weekends. Then a production crisis formed and I was asked to help solve it. I pitched we could use my renderer as solution.

@yiningkarlli Is it true that Hyperion started from a pitch by Brent Burley to Ed Catmull about his idea of a new rendering architecture? The project got green-lit and the team began working on it, so a proper studio initiative.

@vandenbog_art It’s a personal view with personal driving factors. I am not attached to code, but it’s the knowledge and sense of satisfaction for designing and solving the problem. And that I find is missing.

@maxliani If caring about the hobby means caring about the code, rather than its effects, then that makes sense to me. Suppose language models could actually write beautiful and efficient code... I didn't write it so what do I care.

New blog post. Just thoughts this time. maxliani.wordpress.com/2026/05/11/pas…

@VrKomarov Thank you. Yes, those conditions are very rare, and sometimes studio collapse because of missteps, hence the understandable risk aversion.

@maxliani As I understand it, many stars had to align in the sky to achieve this result

@christofsalis When AI codes it up, nothing cements in my head. Next iteration I wont know how to create a better design. Thanks by the way.

@christofsalis That’s not impostor syndrome, that is engineering. You do, you try what works, you measure, analyze, ponder, iterate; throw away several prototypes, until the problem is crystal clear and the design falls out from it. That’s how any valuable product is made.

@vandenbog_art My hobby project is gargantuan, for a hobby project. If I’d pour ambition in there I’d have a competitor for Maya :D The thing is that I struggle to care for what AI produces, and very few parts of a project of mine can fit in the “I don’t care” category.

@maxliani Perhaps larger, more ambitious, hobby projects would put AI back into a smaller role. If an AI can one shot it, it’s probably not very interesting on its own. But as a building block to something bigger? Maybe.

@Kiiiri7 Scope of the project, limited resources, limited time, level of sophistication and the outcome. I mean, if you have seen the Lego movies, you see what I am talking about.

@maxliani From which aspect insane? Difficulty or visual fidelity?

@ana_dodik @OdedStein @vincesitzmann @JustinMSolomon Currently I cache triangulation and associate triangle id and its barycentrics to the vertex, but from your paper it looks like there are much better methods to do that.

@ana_dodik @OdedStein @vincesitzmann @JustinMSolomon That’s very interesting. In the simplest of the applications, I had a problem implementing the face split tool in Workbench: how to express a split vertex in the middle of a polygon if the polygon is allowed to deform (or even animate) while I draw the split.

Can neural fields help unlock new understandings of an old geometry problem? Excited to announce our latest SIGGRAPH Asia work: Variational Barycentric Coordinates! 🧵 - with @OdedStein @vincesitzmann @JustinMSolomon

@ana_dodik What??

👀

@human_klo And dancers

@maxliani We are the music makers :)

@yiningkarlli Thank you. I have more questions than I have convictions here. Trying to keep an open mind and understand how I feel about it.

@maxliani This is a really thoughtful, nuanced take, which means that unfortunately you should probably prepare for both the rabidly absolute-anti-AI people and the ultra-pro-AI bros to attack you for it. Well okay mostly the anti-AI people; most of the pro-AI bros don’t read anymore.

@ssh4net It is relevant. The transition from geometry to microgeometry is still very much unsolved problem.

Looks like that also relevant. But I thought that mainly from fresnel equitation that is a self-contained, and do not take a normals into account. Or just can’t due to averaging, that flatten all micro-details into a single vector. (Valve had workaround that estimate roughness on the fly from normals. And Apple use samilar method to estimate roughness maps on their 3D scanning API).

About 3.5 years ago I did some experiment to improve the behavior of microfacet normals at glancing angles. Many surfaces become highly reflective even though having high roughness. Check the discussion in the thread. Here is my implementation reusing equations you already have: