I think I have noticed something that is partially responsible for the uncanny valley feeling in the faces. The lack of forehead movement. Because of the rig you guys wear to capture the facial movements, that strap and headgear basically cut off your forehead movement. And humans have a lot of movement in their forehead with the skin and the muscles that move and wrinkle along with the rest of your face. Those little nuances of forehead wrinkles or furrowing of the brow are missing because its just a flat forehead due to the rig. That could be a huge improvement on the overall feeling if you could find a way to get the forehead movement back.
I noticed that too. The face camera angle is facing up, capturing a lot of mouth movement but not as much of the eyes and forehead which is noticeable on the final product. The render is almost a little over-the-top on the mouth expression and underwhelming on the eyes and forehead. I would like to see them experiment more on the face cam angle.
Also, a tiny bit of « noise » in the tracking data might help. Tracked movements always tend to be smoothed a bit (and often need to be a minimum to clean them up), just adding or subtracting afterwards a very small random number (let’s say between -1 and +1 pixel for example) can help make it look more natural.
You also talk with most of your head if you use facial expressions. The ears go to move, your scalp needs to move and the forhead. It maybe subtle, but it's enough.
Now that I watched the short, I can pinpoint that what bothers me the most in both characters is the nose. They are very stiff, nostrils and the tip of the bridge between the eyes definitely move a lot more when you talk. Also, I find that the mouths are a bit too wide… Overall, the majority of the expressiveness is solely located under the nose, and it’s really pretty good, but the top of the face oddly stays quite still in comparison of all those big mouth movements.
Makes me appreciate just how good the characters look in Cyberpunk 2077. They are stylized just enough where they don't fall into uncanny valley while remaining really expressive.
This is promising! However, on top of muting performances, the facemorph also has a bit of an "airbrushed" quality to it in my opinion, which is a common factor in the uncanny valley. If you can find a way to make the facemorph match performance intensity and have more blemishes/imperfections, I think you'll have it. Also, watching back, the eyes are a big issue. 11:26 is a clear example, there's just a deadness behind the eyes of the facemorph. This can be chalked up to the muting of the performance, but I think it may prove to be a seperate hurdle you'll have to jump.
It's a weird thing I've noticed in AI art like what Facemorph does. You can create perfectly photo real people, but if you tell it to make the character's eyes look up or down, it just refuses. I struggled with Stable diffusion for 15-20 minutes once, trying to get eyes to simply "look downward," before I decided it simply couldn't be done without physically editing it my self.
Exactly, eyes never move smooth its always skipping from one point to another unless you looking on moving object, this lack that stabilised connection of eyes to one point even while moveing head, and also blinking, there are those slow half blinks that are completely unnatural, then widening or narrowing of eye pupil, all of this is very important
i second the airbrushed thing. it just has that typical AI feel where it went through the thing frame by frame, replacing each image, instead of feeling like a 3d object with volume that’s moving around, kinda like their rock paper scissors anime, but on a much more refined level. still, you can tell, and it’s rather off-putting within the context of the scene imo. i also liked that jordan’s character not only retained more of that expressiveness, it also had more blemishes and wrinkles on the skin which matched the rest of the visuals perfectly. the other character looked straight up photoshopped lol
These guys are true professionals. They always test the waters on new tech but don't just run and jump in the boat. Their full exploration of the tech and pushing it to the limits to see if it's useable, then deciding against it based on how it would genuinely feel, that takes guts. I also watched a podcast of BRCC and they complimented Corridor. They don't just randomly compliment channels. So I believe this group of guys are a premium, professional group of guys.
They are quite literally not professionals yet continue to speak on the state of the industry as if they've ever stepped foot onto a professional union job. That's their biggest problem. This comment is insane.
@@hugboat808 This is an interesting comment. I never got the impression anyone in the crew was passing off their commentary as people who work in Hollywood, just VFX artists who are familiar with the tools and work that is often used in movies. What did they say that gave you the impression they are claiming to be union professionals?
@@ScyrousFX Dude, you can't take click bait titles seriously. This is UA-cam. I see you're a content creator, surely you know how titles can be used to try and drum up traffic.
especially at 14:30 you see the skin move in a way, that a texture on a 3D surface does. The facemorph sort of has no real understanding of true depth (and also as its originated as a tool for stills has not much information over time across frames). So the facemorph has too much of a "smoothing" of the skin and therefore makes it harder to understand stretching and general movement of the skin and its surface details or imperfections. (wrinkles and shadows also)
Yeah, I saw a post on Reddit criticizing the last video and I feel the face smoothing is too much and makes it look like plastic rather than what the original looked like. I feel like the pores on the characters faces are covered with the xenomorph applied.
Speaking of cloth sims, do they take into account the atmospheric resistance of the air as the cloth moves through it? It always feels like the cloth is swinging around in far too "lively" a fashion to be believeable.
@@sandwiched depends on the settings you can use in UE5, there is a cloth simulation that simulates cloth for you, you can even turn it into animation so it keeps the same cloth simulation
@@DeletedDevilDeletedAngel Okay...? My point is all cloth sims (that I've seen... and that I've noticed that they're CG... 🤔) feel overly wiggly, as if the cloth isn't being slowed down by atmospheric friction.
Anyone in or around a design industry has probably heard the phrase "kill your babies/darlings" and I think this is an excellent case study for that. A great idea, with near-perfect execution, but not suited for the project. Letting it go was ultimately the right decision, the ability to receive feedback of that weight is what makes projects excel
10:50 the Unreal render looks miles better than the Facemorph one. Facemorph just looks way too smooth and ends up looking more digital than Unreal with the face tracking.
Definitely, it looks like the filters you use on your phone to smooth out your face and get rid of pimples and wrinkles. It doesn't add to the realism, and really just makes them look like they are made of plastic.
I agree with your final decision of not including the face-morf. I think a part of the reason why it doesn't work is that the face also doesn't match it's environment, especially for the character with the scarf which were always right next to his face, reminding us that the face doesn't belong.
One of these times, we're going to watch an entire Crew episode like this and the reveal at the end is that the entire video and all the shots and interviews were CGI and that the video itself was the creation and it was completely past the valley.
The performance of the expressions on unreal were so good and on point, that they make the video great. The microexpressions and ticks is what sells it.
The facemorph moves me back into uncanny valley. The lipsync and the eyemotions are better, but it looks like someone animated an instagram filter (or like god was a Square Enix fan).
I run virtual reality training for police training. The uncanny valley is the prime complaint from my guys. This video gives me hope for a brighter future!
There are two key things that are missing - 1st - The larger characters body is built sturdy and robust, but his expressions are that of someone who is lighter and thinner. Think of Shaq, he's a big lovable goofball, but when he's making a joke or acting silly he's not airy or floaty in his movements and expressions, even when he's being "lighthearted" there is still his weight and mass that factor into how expressive he is and how quickly he can move about the space he occupies. At times the accelleration from one pose to the next was too "quick" or "light" for a character with a larger frame. 2nd - Obviously ya'll know body language is a huge part of communication, but one of the things i often notice is the movements in the face, head, and neck, do not align with the movements of the shoulders and the chest, specifically how a person raises/lowers their shoulders and how someone will "close" or "open" their shoulders/chest with certain movements etc
I need a step-by-step breakdown on how you guys did this. Is that something I can find on your site?. Also you guys need to do a UFO breakdown on skinny Bob. Thanks for all the great content!
Everybody on the crew rocks, goes without saying obviously, however it was really nostalgic and nice in a way to see sam and niko involved together so much in this one. Just like the good old days.
Every episode I watch makes me want more and more to meet these guys, I mean imagine working at a place that inspire so much creativity. I personaly chose to major in CGI (In Brazil) because of you guys, thank you so much for making this job so much more interesting and fun!
For me, the focus on putting together a realistic face morph seems a bit like running before you walk. Watching the short, I got immediate uncanny valley vibes before I even looked at any faces, because the body movements just aren't there. You can get the most photorealistic face with great expressions, but when their movements are super unnatural none of that matters. Even more so when the head feels weirdly floaty on top of the body, the guy's chin constantly clips into his scarf, everyone's default expression is this really disconcerting show-all-the-teeth smile, and a lot of the expressions themselves feel unnaturally over exaggerated or out of place with the dialogue. I know the main point of this project was to test the face morph stuff, but you've gotta climb out of the janky mocap valley before even starting to worry about the uncanny valley.
This is a tricky thing to pull off with mocap. If the CG character has a different body type and costume (such as the bulky armor here) then the performance is going to be very different. When Toho did mocap for Shin Godzilla, they had the actor drag a weight behind him to simulate the massive tail, otherwise he'd look too lightweight.
@@AdrianParkinsonFilms Oh for sure. Mocap definitely isn't as easy as just putting on a suit and voila. It's a lot of effort on both the actors and the people doing the motion cleanup. But at least for me, when its bad it drags things way further down the "can I empathize with this character?" chart than Bethesda face. And the more realistic you make the model the bigger that effect, because not only is their face acting weird, but their entire body is acting weird. I'm not even talking actor nuance. Basic stuff like fully extended limbs snapping into place, elbows squeezed awkwardly in towards the waist, the body doing a big gesture while the face is angled straight at the camera the entire time. Things that would take a good amount of work to manually clean up, but would remove a lot of that "guy in a mocap suit" feeling.
100% agree. Just the delay between sound and animations immediately ruins the animations and Jordan's overacting makes it far worse. Actually I think the biggest problem here is that they try to emote too much instead of just being normal humans. For example, when I'm talking with someone else, I rarely see their teeth if at all. And for some reason Sam's character has a mouth fit for a frog with how wide it is.
Simple tricks from photoshop I've found that works better than it should for video: Layer up facemorph video on top of the render and any/all: -Opacity between 30% -70% -Adjust basic levels on lower opacity for depth -Blend Mode - Multiply on new layer between 10-20%
Man vid’s like this just make me love Corridor more and more! The knowledge and process you guys share with the community and those outside of the community is seriously amazing! Also I love the whole crew but it was very sweet to just see Sam and Niko at the end discussing the process and giving their final thoughts! It’s really cool to see how far their partnership has come and how they have developed as creatives!
I agree 99%. The main thing to consider is the "all or nothing" argument. Given the range, the sweet spot is going to shift. Someone needs to go in, not unlike an animator, and adjust the blend from expression to expression.
I watched the corridor animation video first and all i could think about was how good the facial animations look and I come to this video and see that was whole goal of the video. Just want to say, mission accomplished & great work! Also a great story too. You guys are killing it.
Don’t know if you’ve tried it, but if the problem with Face morph is that it « smoothes » expressions, is it possible to exaggerate the face movements a bit (let’s say by 10%) before putting it through Face morph ? I think it’s possible to process the data from the face capture to augment the amplitude of points displacement. Also, I notice that your photoreal AI renders are (like I experienced it myself) always a tiny bit off-place from the original image. I personally always take the time to retouch the AI generated image in Photoshop using the liquify tool in order to align extremely precisely the facial features to the original frame. In my own experience, it really does a noticeable difference with those kind of image processing. Because if, for example, the corner of the mouth of the character is off by a few pixels from your reference frame, not only will it be off in all processed frames but the calculated displacement will be attenuated in proportion to how much it’s off.
You gained so much respect for seeing the flaw in this facemorph thing. It felt really awkward, almost like everyone pretended to like it, just because Niko said so. Glad to see your hunger for innovation hasn't ruined your taste or worse, your humbleness.
I think very soon if not already, they'll have the facemorph there to get the nuance, but I think they need to full commit to making a single scene absolutely photo real. Like thinking tie fighters in the atmosphere levels of nuance; haze, subsurface scattering, cloth sims with no clipping, photo real textures, foley effects, everything.
I was impressed all things considered. I saw all of Sam's facial nuances come through. Other than being a little too smooth comparatively, I'd say facemorph shows a lot of promise.
I love the way you guys keep pushing the bounds, especially as we see soo much fearmongering about the new tech, and while it is legitimately scary how advanced the tech is getting (I always think of how technically, now is the "worst its going to be", its only going to get more and more impressive), you guys finding ways to make interesting and enjoyable content shows the bright side of things instead
Success or failure. You guys are pushing the movie and game industries forwards. I wish they would give you guys more props for your innovations and fusion of tech and art. Everything you guys have been doing, has been helping the industries (art, film and games) to a whole new level. Just keep at 'er.
I honestly feel like you made the right decision. While it's great in specific circumstances, overall it's taking away more then it's giving, which is no surprise since it wasn't designed for this task. But it's great that you test everything and see what you can do with it, that's how true innovation is done. I'm always glued to the screen whenever you guys experiment with something new, it's always so exciting.
A failed experiment is still valuable because it provides you with more knowledge than you had before, even if it's just of what doesn't work yet, and what the limitations are
I feel like you guys definitely made the right choice. It looks way better with the nuanced facial expressions coming through. I would love to see more videos like this. Keep it crunchy!
The thing I always find that holds back on realism is the lack of imperfections. Everyone's face has different blemishes, small indents etc. If you have a face that is smooth or has a forced scar it just doesn't look real
I watched the video before watching this BTS and I want to drop my opinion here as a viewer and not a professional. I really like that you used both in the same video. Giving a different art style to each character actually gave them more character. Sam's character being so clean and "perfect" really gave that vision of a video game hero while the unreal engine being used on Jordans character made it feel like a more gritty NPC. I think you guys should experiment more and think less about using one or the other and instead, think about using both dependent on the characters. Using both gives the video an art style that is not seen elsewhere and really adds more to the characters and how you want the viewer to see them. Bravo guys, this was fantastic work and I hope you read this.
You guys should once a year do a video that's basically one of these behind the scenes videos but where you mocap you all into it, using the most up to date tech to create your digi-double, and do them yearly until the technology advances to the point where a VFX artist wouldn't even be able to tell that it's digi doubles in the video
The last debate really depends on the angle you're going for - in drama, you want nuance and detail to suck you in and engross you; in comedy, realism is secondary to expression - over-emphasis works, you don't need to go all out on details to immerse the viewer
The facemorphs being on just the face is creating a contrast with the rest of the hair/model, which brings its own uncanniness. Additionally faces look too smooth, nevermind the muted expressivenes which you already highlighted. A tech to watch perhaps, but nowhere close to ready for use. The performances with the unreal renders are mindblowing tbh
I know this wasn't the focus of this experiment in particular, but something that broke my immersion was the audio. Without it, the shots were very impressive. Perhaps in a future video, you could look into perfecting this aspect? I'd love to see how these groundbreaking experiments could be supported by solid audio engineering :)
It's impressive how nicely Facemorph refines the image. And if you guys didn't had such a good and nuanced facial mo-cap, I would have preferred the one with Facemorph. But so... This one goes to the raw one.
its always really cool to see the kind of stuff that you guys can do! one thing i think has stood out to me from the beginning of your virtual production stuff, and tbh its not just you bc this is stuff ive noticed on actual hollywood cgi, is the facial animation. i know youre using facial capture here, but with the lower fidelity characters, and these which are probably some of your best ive seen yet, i think the performance doesnt quite carry through to the render.
The perfect example of the Uncanny Valley question is the animated Starship Troopers TV show Roughnecks. That show ran for a few seasons and the first season the animation is as basic as it comes and it's super relatable then in the later seasons you really run into The Uncanny Valley. Then the CGI movie Starship Troopers invasion came out decades later and finishes off the series takes the performances back up the other side of the valley. It's a great case study for the evolution of CGI and would make for a great animators react. Also just who doesn't want more starship troopers XD
I think I have noticed something that is partially responsible for the uncanny valley feeling in the faces. The lack of forehead movement. Because of the rig you guys wear to capture the facial movements, that strap and headgear basically cut off your forehead movement. And humans have a lot of movement in their forehead with the skin and the muscles that move and wrinkle along with the rest of your face. Those little nuances of forehead wrinkles or furrowing of the brow are missing because its just a flat forehead due to the rig. That could be a huge improvement on the overall feeling if you could find a way to get the forehead movement back.
I noticed that too. The face camera angle is facing up, capturing a lot of mouth movement but not as much of the eyes and forehead which is noticeable on the final product. The render is almost a little over-the-top on the mouth expression and underwhelming on the eyes and forehead. I would like to see them experiment more on the face cam angle.
Also, a tiny bit of « noise » in the tracking data might help. Tracked movements always tend to be smoothed a bit (and often need to be a minimum to clean them up), just adding or subtracting afterwards a very small random number (let’s say between -1 and +1 pixel for example) can help make it look more natural.
You also talk with most of your head if you use facial expressions. The ears go to move, your scalp needs to move and the forhead. It maybe subtle, but it's enough.
I noticed the same thing. I think Corridor is making leaps, but the animation is still growing. You might even say it's coming to life!
Now that I watched the short, I can pinpoint that what bothers me the most in both characters is the nose. They are very stiff, nostrils and the tip of the bridge between the eyes definitely move a lot more when you talk. Also, I find that the mouths are a bit too wide… Overall, the majority of the expressiveness is solely located under the nose, and it’s really pretty good, but the top of the face oddly stays quite still in comparison of all those big mouth movements.
Makes me appreciate just how good the characters look in Cyberpunk 2077. They are stylized just enough where they don't fall into uncanny valley while remaining really expressive.
I was thinking the same thing
really, you feel good enough to enjoy the story, instead of trying to make it "super real" but end up being over-engineered.
Herd mentality. Go to the next mainstream thing, even if it was shat on so badly when it was the mainstream to do so.
@@kenz2756this comment is herd mentality 😂 no one was shitting on Cyberpunk for the graphics
@@zintack5998 they were on the older gen consoles
This is promising! However, on top of muting performances, the facemorph also has a bit of an "airbrushed" quality to it in my opinion, which is a common factor in the uncanny valley. If you can find a way to make the facemorph match performance intensity and have more blemishes/imperfections, I think you'll have it.
Also, watching back, the eyes are a big issue. 11:26 is a clear example, there's just a deadness behind the eyes of the facemorph. This can be chalked up to the muting of the performance, but I think it may prove to be a seperate hurdle you'll have to jump.
It's a weird thing I've noticed in AI art like what Facemorph does. You can create perfectly photo real people, but if you tell it to make the character's eyes look up or down, it just refuses. I struggled with Stable diffusion for 15-20 minutes once, trying to get eyes to simply "look downward," before I decided it simply couldn't be done without physically editing it my self.
Exactly, eyes never move smooth its always skipping from one point to another unless you looking on moving object, this lack that stabilised connection of eyes to one point even while moveing head, and also blinking, there are those slow half blinks that are completely unnatural, then widening or narrowing of eye pupil, all of this is very important
i second the airbrushed thing. it just has that typical AI feel where it went through the thing frame by frame, replacing each image, instead of feeling like a 3d object with volume that’s moving around, kinda like their rock paper scissors anime, but on a much more refined level. still, you can tell, and it’s rather off-putting within the context of the scene imo. i also liked that jordan’s character not only retained more of that expressiveness, it also had more blemishes and wrinkles on the skin which matched the rest of the visuals perfectly. the other character looked straight up photoshopped lol
These guys are true professionals. They always test the waters on new tech but don't just run and jump in the boat. Their full exploration of the tech and pushing it to the limits to see if it's useable, then deciding against it based on how it would genuinely feel, that takes guts. I also watched a podcast of BRCC and they complimented Corridor. They don't just randomly compliment channels. So I believe this group of guys are a premium, professional group of guys.
They are quite literally not professionals yet continue to speak on the state of the industry as if they've ever stepped foot onto a professional union job. That's their biggest problem. This comment is insane.
We are their Guinea pigs. They know if stuff they do doesn’t work, we’ll tell them.
@@hugboat808 This is an interesting comment. I never got the impression anyone in the crew was passing off their commentary as people who work in Hollywood, just VFX artists who are familiar with the tools and work that is often used in movies. What did they say that gave you the impression they are claiming to be union professionals?
@@ScyrousFX Dude, you can't take click bait titles seriously. This is UA-cam. I see you're a content creator, surely you know how titles can be used to try and drum up traffic.
@@hugboat808 Being a professional means being in an union? Most VFX artist jobs aren't even unionized.
especially at 14:30 you see the skin move in a way, that a texture on a 3D surface does. The facemorph sort of has no real understanding of true depth (and also as its originated as a tool for stills has not much information over time across frames). So the facemorph has too much of a "smoothing" of the skin and therefore makes it harder to understand stretching and general movement of the skin and its surface details or imperfections. (wrinkles and shadows also)
Yeah, I saw a post on Reddit criticizing the last video and I feel the face smoothing is too much and makes it look like plastic rather than what the original looked like.
I feel like the pores on the characters faces are covered with the xenomorph applied.
Exactly where my uncanny activates, the "smoothing" is just wierd
PLEASE PLEASE add a simple cloth sim to Jordan's characters bandana, it would sell the scene more than some additional face filter ever could.
Speaking of cloth sims, do they take into account the atmospheric resistance of the air as the cloth moves through it? It always feels like the cloth is swinging around in far too "lively" a fashion to be believeable.
@@sandwiched depends on the settings you can use in UE5, there is a cloth simulation that simulates cloth for you, you can even turn it into animation so it keeps the same cloth simulation
@@DeletedDevilDeletedAngel Okay...? My point is all cloth sims (that I've seen... and that I've noticed that they're CG... 🤔) feel overly wiggly, as if the cloth isn't being slowed down by atmospheric friction.
@@sandwiched have you seen the cloth simulation preview for unreal engine 5.3-4? Just wondering in case we're thinking about the same cloth thingy
@@DeletedDevilDeletedAngel No, I'm not in the CG field; just a nerd and casual viewer of CD. :)
Anyone in or around a design industry has probably heard the phrase "kill your babies/darlings" and I think this is an excellent case study for that. A great idea, with near-perfect execution, but not suited for the project. Letting it go was ultimately the right decision, the ability to receive feedback of that weight is what makes projects excel
10:50 the Unreal render looks miles better than the Facemorph one. Facemorph just looks way too smooth and ends up looking more digital than Unreal with the face tracking.
I have felt that with this full Facemorph series. I have never liked their facemorph's more than the footage beneath it.
Yeah I agree on the smoothness. It washes away the details that convey emotion.
@@JDP5127 it's looked a tiny bit better on one or two imo, but yeah in general I really don't get why they're trying to hype up facemorph so much
i agree, it looks like a photoshop faceswap
Definitely, it looks like the filters you use on your phone to smooth out your face and get rid of pimples and wrinkles. It doesn't add to the realism, and really just makes them look like they are made of plastic.
The rock having a picture with the rock having a face like the rock is my favourite thing ever.
The “uncanny” to “canny” to “can ye” transition was truly astounding.
Oh and the whole “pushing the boundaries of SFX thing” was cool too.
USE IT FOR CUTSCENES. That’d be a fun gag to make a super realistic cut scene, then cut to a gameplay “scene”.
I agree with your final decision of not including the face-morf. I think a part of the reason why it doesn't work is that the face also doesn't match it's environment, especially for the character with the scarf which were always right next to his face, reminding us that the face doesn't belong.
One of these times, we're going to watch an entire Crew episode like this and the reveal at the end is that the entire video and all the shots and interviews were CGI and that the video itself was the creation and it was completely past the valley.
Corridor Crew is slowly becoming a sci-fi channel
And I’m so here for it
Just without the fi 😂
Scifi channel when it was good
Peak early 2000's entertainment!
Bro they were sci-fi from the very beginning
Sam's Unreal engine virtual production videos are AMAZING
The performance of the expressions on unreal were so good and on point, that they make the video great. The microexpressions and ticks is what sells it.
Still waiting for ‘Adipurush’ movie VFX reaction… 😂
The facemorph moves me back into uncanny valley. The lipsync and the eyemotions are better, but it looks like someone animated an instagram filter (or like god was a Square Enix fan).
I love that you all are always pushing things and thinking outside the box for making progress.
I run virtual reality training for police training. The uncanny valley is the prime complaint from my guys. This video gives me hope for a brighter future!
There are two key things that are missing -
1st - The larger characters body is built sturdy and robust, but his expressions are that of someone who is lighter and thinner. Think of Shaq, he's a big lovable goofball, but when he's making a joke or acting silly he's not airy or floaty in his movements and expressions, even when he's being "lighthearted" there is still his weight and mass that factor into how expressive he is and how quickly he can move about the space he occupies. At times the accelleration from one pose to the next was too "quick" or "light" for a character with a larger frame.
2nd - Obviously ya'll know body language is a huge part of communication, but one of the things i often notice is the movements in the face, head, and neck, do not align with the movements of the shoulders and the chest, specifically how a person raises/lowers their shoulders and how someone will "close" or "open" their shoulders/chest with certain movements etc
I need a step-by-step breakdown on how you guys did this. Is that something I can find on your site?. Also you guys need to do a UFO breakdown on skinny Bob. Thanks for all the great content!
U guys are amazing, u inspire me ❤
The editors again, did great on this video to help the story and portray it. Such a good video and story of a new creative process.
Everybody on the crew rocks, goes without saying obviously, however it was really nostalgic and nice in a way to see sam and niko involved together so much in this one. Just like the good old days.
Every episode I watch makes me want more and more to meet these guys, I mean imagine working at a place that inspire so much creativity. I personaly chose to major in CGI (In Brazil) because of you guys, thank you so much for making this job so much more interesting and fun!
For me, the focus on putting together a realistic face morph seems a bit like running before you walk. Watching the short, I got immediate uncanny valley vibes before I even looked at any faces, because the body movements just aren't there. You can get the most photorealistic face with great expressions, but when their movements are super unnatural none of that matters. Even more so when the head feels weirdly floaty on top of the body, the guy's chin constantly clips into his scarf, everyone's default expression is this really disconcerting show-all-the-teeth smile, and a lot of the expressions themselves feel unnaturally over exaggerated or out of place with the dialogue. I know the main point of this project was to test the face morph stuff, but you've gotta climb out of the janky mocap valley before even starting to worry about the uncanny valley.
This is a tricky thing to pull off with mocap. If the CG character has a different body type and costume (such as the bulky armor here) then the performance is going to be very different. When Toho did mocap for Shin Godzilla, they had the actor drag a weight behind him to simulate the massive tail, otherwise he'd look too lightweight.
@@AdrianParkinsonFilms Oh for sure. Mocap definitely isn't as easy as just putting on a suit and voila. It's a lot of effort on both the actors and the people doing the motion cleanup. But at least for me, when its bad it drags things way further down the "can I empathize with this character?" chart than Bethesda face. And the more realistic you make the model the bigger that effect, because not only is their face acting weird, but their entire body is acting weird.
I'm not even talking actor nuance. Basic stuff like fully extended limbs snapping into place, elbows squeezed awkwardly in towards the waist, the body doing a big gesture while the face is angled straight at the camera the entire time. Things that would take a good amount of work to manually clean up, but would remove a lot of that "guy in a mocap suit" feeling.
100% agree. Just the delay between sound and animations immediately ruins the animations and Jordan's overacting makes it far worse. Actually I think the biggest problem here is that they try to emote too much instead of just being normal humans. For example, when I'm talking with someone else, I rarely see their teeth if at all. And for some reason Sam's character has a mouth fit for a frog with how wide it is.
I also feel like the fact that it is a video game having the “Main Character” being higher fidelity could actually work!!!!!!!
Dude because of anime rock paper scissors I have gotten into blender and I’ve been using cascadeur! I feel like I’m here with you guys lol.
Somehow in my brain when I couldn’t get the ai to be stable enough it went to 3d models lol
Me myself I prefer the straight out of unreal with more emotion
Simple tricks from photoshop I've found that works better than it should for video:
Layer up facemorph video on top of the render and any/all:
-Opacity between 30% -70%
-Adjust basic levels on lower opacity for depth
-Blend Mode - Multiply on new layer between 10-20%
Man vid’s like this just make me love Corridor more and more! The knowledge and process you guys share with the community and those outside of the community is seriously amazing!
Also I love the whole crew but it was very sweet to just see Sam and Niko at the end discussing the process and giving their final thoughts! It’s really cool to see how far their partnership has come and how they have developed as creatives!
they really nailed the faces for this one
Honestly, I feel like rendering both styles and layering them with a blend mode would be the sweet spot.
I agree 99%.
The main thing to consider is the "all or nothing" argument. Given the range, the sweet spot is going to shift.
Someone needs to go in, not unlike an animator, and adjust the blend from expression to expression.
Both vids were awesome! Sick work Corridor🔥🔥
That video ended with the same concrete resolution as Burn After Reading. ... Loved it!
The crazy thing is, in a year this will be surpassed 10x over with the rate at which this stuff is evolving.
In a few years, this might be the foundation for epic vr games
It's like funny cartoons. The jokes land despite the smiley face level detail to the expressions.
I like the process you ended up with. It removed the part of the jank that was bothering me, and left the goofiness. Very nice.
I watched the corridor animation video first and all i could think about was how good the facial animations look and I come to this video and see that was whole goal of the video. Just want to say, mission accomplished & great work! Also a great story too. You guys are killing it.
Don’t know if you’ve tried it, but if the problem with Face morph is that it « smoothes » expressions, is it possible to exaggerate the face movements a bit (let’s say by 10%) before putting it through Face morph ? I think it’s possible to process the data from the face capture to augment the amplitude of points displacement.
Also, I notice that your photoreal AI renders are (like I experienced it myself) always a tiny bit off-place from the original image. I personally always take the time to retouch the AI generated image in Photoshop using the liquify tool in order to align extremely precisely the facial features to the original frame. In my own experience, it really does a noticeable difference with those kind of image processing. Because if, for example, the corner of the mouth of the character is off by a few pixels from your reference frame, not only will it be off in all processed frames but the calculated displacement will be attenuated in proportion to how much it’s off.
You gained so much respect for seeing the flaw in this facemorph thing. It felt really awkward, almost like everyone pretended to like it, just because Niko said so. Glad to see your hunger for innovation hasn't ruined your taste or worse, your humbleness.
I think very soon if not already, they'll have the facemorph there to get the nuance, but I think they need to full commit to making a single scene absolutely photo real. Like thinking tie fighters in the atmosphere levels of nuance; haze, subsurface scattering, cloth sims with no clipping, photo real textures, foley effects, everything.
Amazing video! Final vid looked incredible Sam! Loved Jordan's ad too!
What about having the faces over expressive to compensate for the mutedness?
I was impressed all things considered. I saw all of Sam's facial nuances come through. Other than being a little too smooth comparatively, I'd say facemorph shows a lot of promise.
I love the way you guys keep pushing the bounds, especially as we see soo much fearmongering about the new tech, and while it is legitimately scary how advanced the tech is getting (I always think of how technically, now is the "worst its going to be", its only going to get more and more impressive), you guys finding ways to make interesting and enjoyable content shows the bright side of things instead
“In a world where you can do anything, nothing means anything.”
Artistic honesty is a beautiful thing itself
Corridor Crew is the only thing that recaptures that feeling I had watching Saturday Morning Cartoons as a kid.
Success or failure. You guys are pushing the movie and game industries forwards. I wish they would give you guys more props for your innovations and fusion of tech and art.
Everything you guys have been doing, has been helping the industries (art, film and games) to a whole new level. Just keep at 'er.
Absolutely mind blowing. Really glad you guys didn’t stick with FaceMorph in the end
So good. I love seeing the process and challenges you all overcome!
I honestly feel like you made the right decision. While it's great in specific circumstances, overall it's taking away more then it's giving, which is no surprise since it wasn't designed for this task. But it's great that you test everything and see what you can do with it, that's how true innovation is done. I'm always glued to the screen whenever you guys experiment with something new, it's always so exciting.
honestly, both looked great, the only thing that took me out was the scarf clipping the face
A failed experiment is still valuable because it provides you with more knowledge than you had before, even if it's just of what doesn't work yet, and what the limitations are
You guys absolutely nailed the mocap acting! I can't tell you how much I enjoyed it. And Sam, you made the right decision on the render. 👏👏👏
So relevant too see if things don't work out. It makes us human. Keep up the good work!
Jordan is so good on camera i actually watched the entire sponsored message.
“Let’s talk about what’s canny. Can ye make your own website?” Is the best lead in I’ve ever heard!
I feel like you guys definitely made the right choice. It looks way better with the nuanced facial expressions coming through. I would love to see more videos like this. Keep it crunchy!
Hoping we get some ghost/debunk videos for Halloween. (I mean, I want a whole series of that, but I know you guys have other priorities)
Love the boundary pushing videoes.
That og rock with a written smiley face is my favorite member of the crew now
It's insane what you produce now. You are rally getting close, I look forward to seeing your next step!
There's never a problem you can't throw money and time to solve. Well done.
The thing I always find that holds back on realism is the lack of imperfections. Everyone's face has different blemishes, small indents etc. If you have a face that is smooth or has a forced scar it just doesn't look real
This reminds me when you’re trying to draw a picture, and you can’t tell when to stop, then you go too far and ruin the picture.
I watched the video before watching this BTS and I want to drop my opinion here as a viewer and not a professional.
I really like that you used both in the same video. Giving a different art style to each character actually gave them more character. Sam's character being so clean and "perfect" really gave that vision of a video game hero while the unreal engine being used on Jordans character made it feel like a more gritty NPC.
I think you guys should experiment more and think less about using one or the other and instead, think about using both dependent on the characters. Using both gives the video an art style that is not seen elsewhere and really adds more to the characters and how you want the viewer to see them.
Bravo guys, this was fantastic work and I hope you read this.
You guys should once a year do a video that's basically one of these behind the scenes videos but where you mocap you all into it, using the most up to date tech to create your digi-double, and do them yearly until the technology advances to the point where a VFX artist wouldn't even be able to tell that it's digi doubles in the video
I think face morph was the best every time. It looks so good
The last debate really depends on the angle you're going for - in drama, you want nuance and detail to suck you in and engross you; in comedy, realism is secondary to expression - over-emphasis works, you don't need to go all out on details to immerse the viewer
The facemorphs being on just the face is creating a contrast with the rest of the hair/model, which brings its own uncanniness. Additionally faces look too smooth, nevermind the muted expressivenes which you already highlighted. A tech to watch perhaps, but nowhere close to ready for use. The performances with the unreal renders are mindblowing tbh
The face just looks too smooth! With some post this could be better than Disney.
At 12:30, when it's an obvious choice, but you've sat on the project for so long and can't decide which is better.
I love using Kitbash and Cargo. Cargo made scene design so much easier.
This is the first time i have understood why it's called "uncanny VALLEY." Thanks once again, Crew!
I just realized that Sam and I started learning Unreal Engine around the same year that's pretty awesome.
Damn, that's really good!
1:11 I can't tell you how The Rock is feeling, but I can smell what The Rock is cooking!
I really liked seeing the process of determining which approach was right for the project. I think you guys picked the right one.
I know this wasn't the focus of this experiment in particular, but something that broke my immersion was the audio. Without it, the shots were very impressive. Perhaps in a future video, you could look into perfecting this aspect? I'd love to see how these groundbreaking experiments could be supported by solid audio engineering :)
Sam's comment gave me an idea. VFX artists make Roller Coasters R-rated
I'm very glad you didn't go with the face morph, it just makes them look buttery smooth and inconsistent compared to the rest of the frame
Using facemorph for one character but not the other was the right call. It's not a catch-all solution, but when it works it works.
It's impressive how nicely Facemorph refines the image. And if you guys didn't had such a good and nuanced facial mo-cap, I would have preferred the one with Facemorph. But so... This one goes to the raw one.
For a software designed specifically for still images, it does surprisingly well with video.
This explains why all the characters in Ashoka were so emotion-less and stoic, they were done in Anamorph
Is going in and hand animating the facial animations to be even more exaggerated before applying the morph a viable option?
The self-destructive perfectionism that synced up the video with the audio at 14:30 is relatable :|
Entire gaming industry is learning new things from this video
One of your best vids yet wow
Honestly the biggest distraction to me was the guys chin clipping through his bandana
Am I the only one that thinks the face-morth isn't really that much better? Even on the last video.
That facial performance just straight off the camera is really impressive! I really liked this video!
1:50 Rocco has evolved but he is still just A ROCK.
its always really cool to see the kind of stuff that you guys can do! one thing i think has stood out to me from the beginning of your virtual production stuff, and tbh its not just you bc this is stuff ive noticed on actual hollywood cgi, is the facial animation. i know youre using facial capture here, but with the lower fidelity characters, and these which are probably some of your best ive seen yet, i think the performance doesnt quite carry through to the render.
impressive. Simply impressive. This is the new gold-standard
The perfect example of the Uncanny Valley question is the animated Starship Troopers TV show Roughnecks. That show ran for a few seasons and the first season the animation is as basic as it comes and it's super relatable then in the later seasons you really run into The Uncanny Valley. Then the CGI movie Starship Troopers invasion came out decades later and finishes off the series takes the performances back up the other side of the valley. It's a great case study for the evolution of CGI and would make for a great animators react. Also just who doesn't want more starship troopers XD
Great watching implementation of new stuff in your lab. Greetings from Norway.
Great video like always keep up the good work. 👍