The 8 Gen 3 also still uses the previous tile-based A7x GPU architecture, while newer chips use the "A8x family of GPUs based on the new Slice architecture".
While the methods are similar in that they both ray-march through the scene to compute per-pixel fluence, the algorithm presented in the blog post scales linearly with the number of light sources, whereas Radiance Cascades can handle an arbitrary distribution of light sources with constant time by benefiting from geometric properties of lighting. Radiance Cascades are also not limited to SDFs for smooth shadows.
Yeah, and I believe Radiance Cascades accurately calculate the size of the penumbra from the size and distance of the area light, which also means that point light sources, as in reality, always produce hard shadows.
The technique here seems to rely more on eyeballing a plausible penumbra without explicitly considering a size of the light source, though I don't quite understand the core intuition.
Payment services like credit cards demand a significant fee for a (nowadays) technically trivial service: instant cash-free payments. These could be replaced with modern instant bank transfer standards, like FedNow in the US:
These don't require external middle men (like credit card companies) and are therefore almost free. Unfortunately the US is late to the party (in India and some other countries these are already widely used for years), so many banks don't support FedNow yet.
Hot take: text-to-image models should be biased toward photorealism. This is because if I type in "a cat playing piano", I want to see something that looks like a 100% real cat playing a 100% real piano. Because, unless specified otherwise, a "cat" is trivially something that looks like an actual cat. And a real cat looks photorealistic. Not like a painting, or cartoon, or 3D render, or some fake almost-realistic-but-cleary-wrong "AI style".
FYI: photorealism is art that imitates photos, and I see the term misused a lot both in comments and prompts (where you'll actually get subideal results if you say "photorealism" instead of describing the camera that "shot" it!)
I've heard chairs of animation departments say they feel like this puts film departments under them as a subset rather than the other way around. It's a funny twist of fate, given that the tables turned on them ages ago.
Photorealistic models are just learning the rules of camera optics and physics. In other "styles", the models learn how to draw Pixar shaded volumes, thick lines, or whatever rules and patterns and aesthetics you teach.
Different styles can reinforce one another across stylistic boundaries and mixed data sets can make the generalization better (at the cost of excelling in one domain).
"Real life", it seems, might just be a filter amongst many equally valid interpretations.
If Midjourney is a niche, then what is the broader market for AI image generation?
Porn, obviously, though if you look at what's popular on civitai.com, a lot of it isn't photo-realistic. That might change as photo-realistic models are fully out of the uncanny valley.
Presumably personalized advertising, but this isn't something we've seen much of yet. Maybe this is about to explode into the mainstream.
Perhaps stock-photo type images for generic non-personalized advertising? This seems like a market with a lot of reach, but not much depth.
There might be demand for photos of family vacations that didn't actually happen, or removing erstwhile in-laws from family photos after a divorce. That all seems a bit creepy.
I could see some useful applications in education, like "Draw a picture to help me understand the role of RNA." But those don't need to be photo-realistic.
I'm sure people will come up with more and better uses for AI-generated images, but it's not obvious to me there will be more demand for images that are photo-realistic, rather than images that look like illustrations.
> If Midjourney is a niche, then what is the broader market for AI image generation?
Midjourney is one aesthetically pleasing data point in a wide spectrum of possibilities and market solutions.
Creator economy is huge and is outgrowing Hollywood and the Music Industry combined.
There's all sorts of use cases in marketing, corporate, internal comms.
There are weird new markets. A lot of people simply subscribe to Midjourney for "art therapy" (a legit term) and use it as a social media replacement.
The giants are testing whether an infinite scroll of 100% AI content can beat human social media. Jury's out, but it might start to chip away at Instagram and TikTok.
Corporate wants certain things. Disney wants to fine tune. They're hiring companies like MoonValley to deliver tailored solutions.
Adobe is building tools for agencies and designers. They are only starting to deliver competent models (see their conference videos), and they're going about this a very different way.
ChatGPT gets the social trend. Ghibli. Sora memes.
> Porn, obviously, though if you look at what's popular on civitai.com, a lot of it isn't photo-realistic.
Civitai is circling the drain. Even before the unethical and religious Visa blacklisting, the company was unable to steer itself to a Series A. Stable Diffusion and local models are still way too hard for 99.99% of people and will never see the same growth as a Midjourney or OpenAI that have zero sharp edges and that anyone in the world can use. I'm fairly certain an "OnlyFans but AI" will arise and make billions of dollars. But it has to be so easy a tucker who doesn't learn to code can use it from their 11 year old Toshiba.
> Presumably personalized advertising, but this isn't something we've seen much of yet.
Carvana pioneered this almost five years ago. I'll try to find the link. This isn't going to really take off though. It's creepy and people hate ads. Carvana's use case was clever and endearing though.
> Porn, obviously, though if you look at what's popular on civitai.com, a lot of it isn't photo-realistic.
I don't have an argument to make on the main point, but Civitai has a whole lot of structural biases built into it (both intentionally and as side effects of policies that probably aren't intended to influence popularity in the way they do) that I would hesitate to use "what is popular on Civitai" as a guide to "what is attractive to (or commercially viable in) the market", either for AI imagery in general or for AI imagery in the NSFW domain specifically.
Well, as I said, if I type "cat", the most reasonable interpretation of that text string is a perfectly realistic cat.
If I want an "illustration" I can type in "illustration of a cat". Though of course that's still quite unspecific. There are countless possible unrealistic styles for pictures (e.g. line art, manga, oil painting, vector art etc), and the reasonable thing is that the users should specify which of these countless unrealistic styles they want, if they want one. If I just type in "cat" and the model gives me, say, a water color picture of a cat, it is highly improbable that this style happens to be actually what I wanted.
If I want a badly drawn, salad fingers inspired scrawl of a mangy cat, it should be possible. If I want a crisp, xkcd depiction of a cat, it should capture the vibe, which might be different from a stick fighters depiction of a cat, or "what would it look like if George Washington, using microsoft paint for the first time, right after stepping out of the time machine, tried to draw a cat"
I think we'll probably need a few more hardware generations before it becomes feasible to use chatgpt 5 level models with integrated image generation. The underlying language model and its capabilities, the RL regime, and compute haven't caught up to the chat models yet, although nano-banana is certainly doing something right.
> what is the broader market for AI image generation?
Replace commercial stock imagery. My local Home Depot has a banner by one of the cash registers with an AI house replete with mismatched trim and weird structural design but it's passable at a glance.
I really hated the depressing grey GUI of Windows 95/98/NT/2000/Me. It looks like working in a dull grey concrete office with a grey PC and grey monitor while wearing a grey tie. I get that Windows XP and 8 look too colorful for many people (I like them), but Vista & Co, with their glass design, managed to avoid colors while still not making everything a drab grey.
All those versions had customizable color themes, which windows has gradually weakened to almost nothing. There was nothing stopping you from rocking hotdog stand back then except your own good taste and the bleeding from your eyes.
I liked it although that was maybe because Windows 2000 was a great OS. Having seen the awful mess the Unix vendors made of CDE/Motif, 2000 felt more professional.
This begs the general question: why were PCs (and monitors, keyboards, mice etc) always sold in ugly grey, for decades? Before they finally relented and switched to black?
I would do unsavory things to get a matching keyboard, mouse, and monitor in translucent plastic akin to an Atomic Purple Gameboy Color. I've tried retrofitting an iMac G3 multiple times, but they're so old now that taking them apart literally breaks them because the plastic's too brittle.
Apparently (from a layman's perspective) the difference between conventional RGB ray tracing and spectral ray tracing is this:
RGB assumes all light sources consist of three RGB lights, where the brightness of red, green, and blue varies. E.g. a yellow light would always be a red and a green light.
In contrast, spectral rendering allows light sources with arbitrary spectra. A pure yellow light (~580 nm) is different from a red+green light.
The physical difference is this: If you shine, for example, a pure yellow light on a scene, everything looks yellow, just more or less dark. But if you shine a red+green (impure yellow) light on a scene, green objects will be green and red objects will be red. Not everything will appear as a shade of yellow. Conventional RGB rendering can only model the latter case.
This means some light sources, like high-pressure sodium lamps, cannot be accurately rendered with RGB rendering: red and green surfaces would look too bright.
(Also note that the linked post has also a part 1 and 3, accessible via "next/previous post" at the bottom.)
> RGB assumes all light sources consist of three RGB lights
Another way to say this is that conventional 3 channel renderers pre-integrate the spectrum of lights down to 3-channel colors. They don’t necessarily assume three lights, but it’s accurate to say that’s the net effect.
It’s mostly just about when you integrate, and what you have to do when you delay the integration. It’s kind of a subtle distinction, really, but rendering with spectral light and material data and integrating down to RGB at the end more closely mimics reality; the cones in our eyes are the integrators, and before that everything is spectral. Or more accurately, individual photons have wavelength. A spectrum is inherently a statistical summary of the behavior of many photons.
I guess it mainly makes a difference for light sources that are very yellow but not very red and green (sodium lights) or very cyan but not very green and blue (no realistic example here). Considering that actual sodium lights are being largely replaced by white LEDs, which can be modeled quite well with RGB ray tracing, spectral rendering might not offer a significant advantage for most applications.
Yeah exactly. Spectral isn’t often giving you a very different result from RGB. It rarely matters for entertainment rendering like films & games, but it’s useful for scientific predictive rendering.
Sodium lights are a problem not because they’re yellow, but because they have very spiky spectra. Smooth spectra, whether it’s lights or materials, will tend to work fine in RGB regardless of the color.
> Sodium lights are a problem not because they’re yellow, but because they have very spiky spectra.
A side effect of this (and other low-CRI lights) is that it's hard to take pictures in them, because if you take a picture of a person you want their skin tone to look just right or else they look weird and sickly and unattractive.
Regular white balance algorithms are not quite able to handle this. So you might imagine why phone cameras are motivated to do AI processing or people processing or other things that make the picture look overprocessed. Because the people are temporarily literally the wrong color in that lighting, and an AI model may be capable of knowing what color they "actually" are.
(That said, the main reason photos look overprocessed is that for some reason nobody on Earth ever implements sensible sharpening algorithms. They always use frequency-based ones that cause obvious white halos. Learn about warp-sharpen and median filters, people.)
It also becomes important for rendering glass and other highly refractive substances. Some conventional RGB rendering engines can mimic dispersion, but with spectral rendering you get it "for free."
One issue with Hero Wavelength sampling (mentioned in article) is that because IOR is wavelength-dependent, after a refraction event, you basically lose the non-hero wavelengths, so you get the colour noise again through refraction.
Sure, but most of that is from industrial production, and really should be debited on importing nations’ CO2 accounts. Whereas in the US transportation, heating and construction are the main consumers.
Importing nations are already paying for the imports themselves, from which China profits. It seems reasonable that this leaves the responsibility for the energy used on China's side.
> Now, the research team is conducting systematic studies on the key scientific issues related to adding thorium, and aims to completethe construction of a 100-megawatt TMSR demonstration project, and begin operation by 2035.
For comparison: A commercial nuclear power plant is 1 gigawatt, a 10x difference. I assume this would be the next step.
The typical 1 gigawatt rating for a nuclear power reactor is measuring electrical output. Given the various inefficiencies, the actual reactor output (as heat) is something like 3x that amount. Whereas a research reactor will be quoted as thermal output.
That to say, a typical commercial reactor might be 30x the power of a 100 MW research device.
reply