Why is it so hard to make digital fire look good?
vulture.com
vulture.com
I can sympathize with the writer: they're often paid by the word, and finding/producing good illustrative images and securing the rights to them can easily double or triple their work, with no additional pay.
On the other hand, I have zero sympathy for the editor: it's their job to realize the article needs pictures, and pay the writer extra, or pay someone else, to source them!
The publishing industry is just in a very bad place right now, mainly because it's so unprofitable for all but the biggest few players and the spigot of VC money has been turned off.
I'm not sure, but maybe the reason is related to the source of this article (and again, a missing license agreement)?
(Repost of <https://news.ycombinator.com/item?id=30349455>)
A related phenomenon is the same kind of article, but full of irrelevant stock images. (<https://news.ycombinator.com/item?id=37167492>)
https://www.latimes.com/business/hiltzik/la-fi-hiltzik-getty...
That's a particular extreme example, but Getty are notorious for being sue-happy.
>Friedlander notes that the fire scene in No Hard Feelings was part of reshoots that took place just a month and change before the film hit theaters — giving the effects team only a couple of weeks to work.
Literally the only reason. I'll bet you anything that team didn't have a few weeks. More like hours. A set has to be built, cameras transported, crew flown in. That overhead is accounted for. CG is considered a bunch 1s and 0s we can conjure from thin air with a few clicks. Bad directors expect miracles, don't get them, then scream "CG BAD!".
As someone who worked as a camera person I want to emphasize another aspect: Shooting the real thing (fire) is itself not without its challenges. You really need to match light levels well or else you won't even notice fire in the picture. Fire doesn't always burn in visually pleasing ways, and especially outside wind can make or break how it will look.
(skeptical)
Fire is a fluid simulation and this has been in consumer tools like houdini for a long time now. Even in avatar from 14 years ago there were plenty of CG explosions that people didn't think about.
The moment you comp it in to a live plate our uncanny valley spidey senses kick in
I haven't seen this be the case if the scale is right and the resolution is good. After that it is a matter of compositing. There is actually a decent amount of cg fire out there now. I found this after 10 seconds of searching on youtube:
https://www.youtube.com/watch?v=1Ec3hrkwh8Q
Doing live action integration is tough in general, but fire is additive and that helps quite a bit. LEDs that can be add lighting to the photography helps too.
And, I realise. I first learned Houdini c. 2004 so I’m aware of the technology.
Your YouTube link is a perfect example. It looks good but certainly doesn’t pass the bar for mistake-it-for-the-real-thing realism.
Also houdini in 2004 didn't have anything like the fluid simulations it has now.
Can you really? On the clothing of a human body that's in motion in a strong wind? That accurately illuminates that human and the environment around them? And the fire and illuminated objects are all properly reflected on adjacent shiny surfaces like car window and car hood?
Because that's the scenario presented at the very top of the article. And it does indeed sound dastardly difficult. Or while it's maybe possible with enough painstaking work, it's difficult enough that no movie would ever have a large enough budget for it. (Although I really do wonder if we have accurate enough physics models of the igniting gases to truly realistically show the way licks of flames ignite in a strong wind -- as opposed to a more bog-standard campfire type.)
I wasn't saying all they needed to do to get perfectly realistic fire in all scenarios is hire one guy with a 1950x. Our pipeline still ends on a supercomputer. I'm saying if you see a garbage fire in a film you don't need to write an article about "why fire hard". The answer is "they were unwilling to pay for a realistic fire". We're not talking billions of dollars here. Not everything is Avatar. But like I said, studios often expect miracles for little to no money. Fire isn't like skin/faces which is actually genuinely ridiculously hard to pull off.
>Although I really do wonder if we have accurate enough physics models of the igniting gases to truly realistically show the way licks of flames ignite in a strong wind -- as opposed to a more bog-standard campfire type
That would make an excellent article because you don't actually want a "realistic" fire. Don't forget we're doing entertainment, not science. Turbulence in media is just a noise function. I want a pretty vector field visual to indicate where and how my sim will move. Pyrotechnics is the same. We want control. Realism is undesirable. It's like asking for realistic silencers. John Wick is amazing because the silencers aren't realistic. The explosions and fires you see in film/tv aren't functionally realistic (though manipulated to appear so) and that's a good thing.
I think OP is saying he has that freedom and time to do exactly that on his desktop.
Just because we can easily make realistic-enough fire in some scenarios doesn't mean we can in all. And clothing on fire on a human body in motion in the wind sounds like one of the hardest scenarios of all.
TEST AH,65h ; Wanna know why 60h? Me too.
JNZ @@nx ; This is pure experience.
; ok but it's better with 65
[1] https://github.com/2point21/lba1-classic/blob/main/SOURCES/F...It's hard because of the randomness associated with flame physics. It's near impossible to look good because of the lack of interaction of the set and performers.
Where VFX fire shines is in layering additional fire elements into a shot with with real fire. It makes a bigger scene without adding safety concerns to the crew.
I'll list some examples below.
Look at the different movements and interactions I have with the fire and the fire has with itself as I move and spin. Not impossible to replicate digitally, but would require so much painstakingly detailed work that the cost of it outweighs the cost of doing it practical. Budget wise, you have 2 options. Pay for a vfx version that is good but not perfect, or pay for a practical fire burn.
Practical burns come with an added benefit of getting a better performance from your actors - https://www.instagram.com/p/Cy31ycnv300/?img_index=2
The second slide of this series is a very small burn. It would have been easy to do with a light effect and vfx. What you would likely lose is the performer reacting in an authentic way. The head tilt from feeling the warmth. The delicateness of removing the jacket swiftly without touching the fire. All these things get lost when there is not a real fire element to influence it.
I'd figure it wouldn't be too hard to record couple of hundred of hours of different kinds of fires burning, maybe "re-encode" / transform all the frames from pixel data to a more convenient representation like motion vectors or something and turn that into some sort of model.
Maybe a ControlNet- or motion model to use with Stable Diffusion? Or some kind of proprietary output model which could be used with video editing software / motion capture?
Usually the hardest part with AI seems to be coming up with the data to train with. In this case, that should be the least of one's problems.
You could even record fires in front of a green screen and, hell, setup it like so that you can control variables like wind direction with fan(s), intensity of the fire, temperature of the fire,...
I might even be possible to train the model "in reverse", in such a way the the goal for the AI is to come up with correct set of motion vectors, when given wind direction and speed, intensity and temperature as inputs. Since you can control these variables and record the real result, you have the ground truth for something like reinforcement learning.
So the question is; am I just overly optimistic and dumb, or isn't this a relatively easy thing to do?
There are videos showing the results with/without the AI detail pass.
I highly recommend people watch the film and then read about it's production after, as I did.
It also serves to prove the thesis of TFA, since they certainly don't look photorealistic. Despite that, I do think they look gorgeous.
How does a low texture res mushroom cloud with a few dozen polygons from a 26 year old video game "prove the thesis since it doesn't look photorealistic"?
This has no connection at all to how modern fire effects are done. That's like seeing a cave drawing and claiming no painting can ever look realistic.
Obviously there’s this which pops in every thread on the matter. Rightfully so. https://madebyevan.com/webgl-water/