Square Enix Artist Discusses Rebirth's Modernization of Final Fantasy VII 3D Assets

[T0@st]

It'd be fair to say Final Fantasy VII Rebirth's next-gen makeover of characters, monsters, and more from the 1997 original has been a spectacular glow-up. The modern console era has returned an iconic cast and world to us with a level of realism in gameplay that even pre-rendered cutscenes over 25 years ago couldn't match. We asked Square Enix if they could crunch some numbers and share some insight into the changes nearly three decades of technological advancement have wrought. Here, main character modeler and lead character artist Dai Suzuki walks us through a selection of characters, creatures, weapons, and more.

Dai Suzuki: When people think of Cloud, most think of his gigantic sword and his unique hairstyle. Because it is so iconic, we needed to put special effort into creating Cloud's hair for Final Fantasy VII Remake, to properly express his personality. The hair was an extremely high-priority element and in fact accounted for half of the total polygon count for the whole model. In Final Fantasy VII Rebirth, the hardware has been changed to PS5, allowing for a higher polygon count to be used than in Final Fantasy VII Remake.




Creating fur - Red XIII and Cait Sith
Dai Suzuki: Red XIII and Cait Sith's entire body is also rendered by the placement of many hair cards. Since their fur is one of their unique characteristics, it was necessary to not only just let their hair cards grow, but also to pay attention to the flow of the fur.



Their blocky appearance is adorable in the original too, but don't you just want to pet the new Cait Sith?

Character facial animations
Dai Suzuki: Polygons are essential not only in depicting a character's physical appearance, but also in expressing their inner personality. When expressing emotions through facial animation, facial distortion is greatly influenced by the total number of polygons.



The characters in the original version had relatively expressionless faces, but in Final Fantasy VII Remake and Final Fantasy VII Rebirth, their facial expressions allow us to convey the emotions of the characters much more powerfully.



Weapons - Buster Sword
Dai Suzuki: Weapons also have more polygons due to the increased detailing. The Buster Sword has a simple design, but most of the polygons were used around the materia slots and the leather wrap around the handle.



The materia that attaches to the weapon also has mesh on the weapon side, and out of the 8,000 polygons, approximately 1,000 polygons are used for the two materia slots.

Monsters - Midgardsormr
Dai Suzuki: Next, let's take a look at the modeling of the monsters. Midgardsormr's numerous spikes are modeled all the way down to the tip of its tail. These spikes, which were absent in the original version, make its silhouette even more menacing.



These details play an important role in conveying to the player the feeling of incurring lots of damage from the spikes when the creature lashes its large tail.

A matter of scale - from card keys to battleships - Largest and Smallest Polygon Counts
Dai Suzuki: Now, which of Final Fantasy VII Rebirth's objects uses the least number of polygons, and which uses the most? I don't have all the data myself, but the numbers here should not be too far off.

• Smallest count: 12 polygons
• Largest count: Approx. 2,300,000 polygons

The smallest was a box used as a dummy model. This is data that isn't visible during the gameplay. Excluding this, I think the smallest would be a tiny note or a card key.

The largest is the battleship Relnikha, which Scarlett rides. This is used in the cutscenes, but it has such a large polygon count since we had to treat the interior not as a background but as an all-inclusive character.

In this article, we focused on polygon count differences between the original version and Rebirth. With the polygon counts, you can compare them through the numbers, so it's relatively straightforward. In game development, we often create a high-poly model with more than 10 million polygons for baking a Normal Map, but I wonder if we'll eventually have hardware that can effortlessly run such models in a game? While I'm not sure whether that would be a good idea, I do have high hopes for the future evolution of hardware.

• Dai Suzuki, Lead Character Artist, Square Enix

Note: "Card" refers to polygons with hair-like design.

View at TechPowerUp Main Site | Source

 

[Shihab]

Would be an interesting topic, if we had gone from PSX FFVII to Remake/Rebirth. But Square Enix milked this IP till it drew blood in the last two decades. Cloud's hairstyle has been standardized this way since at least the 2005 movie, and in the entire design has seen (minor) iterations in games including the Dissidia spinoff.

 

[dragontamer5788]

Polygon counts undercount the extra work that goes into modern games.

Extra-high polygons are made for the designing process with modern "computer sculpting" applications (ex: ZBrush). I don't know what tools Square Enix uses, but they almost certainly had a very high (~1 million) polygon of these characters at the sculpting phase. Then, they optimize and decimate the "virtual sculpture" down. Normal-maps, bump-maps, displacement-maps. They add layers on top of layers of 2D data (cheaper to calculate in a video game) to emulate effects (shadows can be emulated with 2d normal maps for example).

In the end, the 8000-polygon modern buster sword is "built" from the 1-million (or more) model. The final "lower poly" 8000 sword is an optimization, a way to compress all of this effort into a format that a modern GPU can model and actually show you at 60fps.

---------

There was a great .gif talking about this effect with Super Mario Odyssey.

 

[Waweq]

Too bad it flopped in Japan and US, PS5 is a weak selling point for exclusives, look at Ronin

 

[InVasMani]

Polygon counts undercount the extra work that goes into modern games.

Extra-high polygons are made for the designing process with modern "computer sculpting" applications (ex: ZBrush). I don't know what tools Square Enix uses, but they almost certainly had a very high (~1 million) polygon of these characters at the sculpting phase. Then, they optimize and decimate the "virtual sculpture" down. Normal-maps, bump-maps, displacement-maps. They add layers on top of layers of 2D data (cheaper to calculate in a video game) to emulate effects (shadows can be emulated with 2d normal maps for example).

In the end, the 8000-polygon modern buster sword is "built" from the 1-million (or more) model. The final "lower poly" 8000 sword is an optimization, a way to compress all of this effort into a format that a modern GPU can model and actually show you at 60fps.

---------

There was a great .gif talking about this effect with Super Mario Odyssey.

View attachment 340934

Disturbing contrast between polygon usage between Nintendo and Square Enix. I'm far more impressed with the per polygon level of scene optimization Nintendo achieved in the end. Actually it even begs the question of how would the original FF7 models look I if they had just been redone like Super Mario Odyssey approached it.

 

[konga]

Disturbing contrast between polygon usage between Nintendo and Square Enix. I'm far more impressed with the per polygon level of scene optimization Nintendo achieved in the end. Actually it even begs the question of how would the original FF7 models look I if they had just been redone like Super Mario Odyssey approached it.

Normal maps and PBR only go so far. The PS1 FF7 models were insanely low-poly and are unsalvageable.

I don't think FF7 Rebirth's polygon count is too high or anything at all, by the way. Modern consoles are more than capable of processing that many polygons, and FF7 Rebirth in particular have some of the best-rendered characters in any game out there, so they're doing something right. (I refrained from saying "best-looking" since looks are subjective, but it's hard to deny how impressive S-E's character rendering is)

 

[dragontamer5788]

Disturbing contrast between polygon usage between Nintendo and Square Enix. I'm far more impressed with the per polygon level of scene optimization Nintendo achieved in the end. Actually it even begs the question of how would the original FF7 models look I if they had just been redone like Super Mario Odyssey approached it.

The important thing here is that the Super Mario Odyssey coin has far more "effort" put into it, despite technically being at a lower polygon count. There's also a bit of the Cel-Shading style vs the slightly more realistic style in Super Mario Odyssey at play here.

There's also the "Uber Shader", or PBR shader (as @konga point outs). When you make a modern model, you need to build it with full understanding of PBR, which may have existed as a theoretical PH.D level paper back in 1996 (or whenever PS1 / FF7 came out first), but today PBR became absurdly more popular to learn / teach at schools after Disney's Wreck it Ralph. (A single "Uber-shader" in the PBR style made all the different characters in that movie, and Disney published many papers on how PBR / Uber-shader worked).

Everything post 2012 or so (aka: after Wreck-it Ralph) is going to be PBR. A bunch of things will be PBR before that (there's plenty of games in the 00s that had physically based rendering), but the "Wreck it Ralph" moment really brough that technique into the mainstream. (Physically Based Shading At Disney (disneyanimation.com))

 

[Gmr_Chick]

Polygon counts undercount the extra work that goes into modern games.

Extra-high polygons are made for the designing process with modern "computer sculpting" applications (ex: ZBrush). I don't know what tools Square Enix uses, but they almost certainly had a very high (~1 million) polygon of these characters at the sculpting phase. Then, they optimize and decimate the "virtual sculpture" down. Normal-maps, bump-maps, displacement-maps. They add layers on top of layers of 2D data (cheaper to calculate in a video game) to emulate effects (shadows can be emulated with 2d normal maps for example).

In the end, the 8000-polygon modern buster sword is "built" from the 1-million (or more) model. The final "lower poly" 8000 sword is an optimization, a way to compress all of this effort into a format that a modern GPU can model and actually show you at 60fps.

---------

There was a great .gif talking about this effect with Super Mario Odyssey.

View attachment 340934

This sounds a lot like how players of The Sims 2-4 (myself included) make their custom content - mainly clothes and hair. They might start with a high poly reference - let's use Cloud's hair as an example here -- in Blender, but it would inevitably have to be "decimated" (the mesh would have to have its poly count and thus detail, lowered) to a reasonable number for LOD0...and then reduced further for LOD1...rinse and repeat for LOD 2 thru 3, taking care not to decimate it too much.

Also, bump/normal maps are amazing!

 

[KLMR]

Next month we will have a piece where LOD is discussed as a new technology to handle high poly assets in games... FFS.