Maybe they render their models to a low-res texture, then render that texture as a 2D quad. They could get the single pixel black outline by rendering the quad black with an offset or scale, then rendering the colour version on top.

Maybe they render their models to a low-res texture, then render that texture as a 2D quad. They could get the single pixel black outline by rendering the quad black with an offset or scale, then rendering the colour version on top.

If you want to look this up, it's often called the "impostor" technique.

It's often used so that you can draw a huge crowd of (billboard) characters while only actually rendering one 3D model, or so that you can draw many (billboard) characters, but not have to actually render all of their 3D models every frame (caching the impostor textures).

The effect shown in those screenshots is nothing more than 2D sprites in a 3D scene. 2 triangles for each character and an animated 2D image.

They are just sprites, drawn with an overhead look, and placed over the road with a 3D transform instead of an orthogonal 2D transform.

Update: After watching a YouTube video of the iOS version, I can confirm that that is all they are doing.

L. Spiro

The effect shown in those screenshots is nothing more than 2D sprites in a 3D scene. 2 triangles for each character and an animated 2D image.

They are just sprites, drawn with an overhead look, and placed over the road with a 3D transform instead of an orthogonal 2D transform.

Update: After watching a YouTube video of the iOS version, I can confirm that that is all they are doing.

L. Spiro

yeah, by the time i got down to your post i'd pretty much come to that same conclusion.

so, why do you think they did it that way? for the look (retro gameboyish) ? not enough power to render a model ? they must have done it for the look. if they did sprites due to lack of processing power, surely they could have made better looking sprites. the "look" definitely says "GTA" to me.

that's what it looked like to me too,

looks to me like they render normal polygon characters and as a post process, create black borders around those character areas. if you look at e.g. http://www.el33tonline.com/images/cache/8240.jpg

you see the characters intersect properly with the waterplane (check the char on left most), yet there is a black outline, in motion it also looks fluid like MD2 animated meshes, very smooth, viewed from several angels without popping like prerendering would do, without false intersections like sprites would do. I'm quite convinced it's polys+edge detection..

looks to me like they render normal polygon characters and as a post process, create black borders around those character areas. if you look at e.g. http://www.el33tonline.com/images/cache/8240.jpg
you see the characters intersect properly with the waterplane (check the char on left most), yet there is a black outline, in motion it also looks fluid like MD2 animated meshes, very smooth, viewed from several angels without popping like prerendering would do, without false intersections like sprites would do. I'm quite convinced it's polys+edge detection..

The character on the far-left is probably just drawn to be sinking into the water. Another animation.
The way they walk is not as smooth as you think. You can clearly see the frames of 2D animations on clear videos, and they never change perspective whether they are high on the screen or low. You never see any other angle than that which has been drawn onto the 2D sprite. All there is is warping due to the different perspectives, but it still shows the same character from the same angle.

After that, the outline is blocky and oriented with the sprite.
If it was blocky but oriented with the screen, it would be a low-resolution post-processing effect. Since it is oriented with the sprite (see images in first post), it is drawn into the sprite, and the sprite itself is low-resolution. You can even see this in your screenshot on the middle red guy drowning. If you scale the image up for a closer look, his outline is blurry because it has been rotated with the sprite. The others’ are very sharp because the sprites are aligned mostly with the screen.

The original data was likely 3D models with effects to draw outlines around them and an either artistic touch to make them look flat or they just used flat shaders.
Those were then pre-rendered onto sprites for the final game. This is why the objects look 3D and why they move smoothly. Since the sprites were generated from 3D objects, it is hard to tell the difference, and the animations are smooth, but they are still jerky if you look closely enough.

It is basically Donkey Kong Country but with an outline for style and from overhead. That’s all.

yeah, by the time i got down to your post i'd pretty much come to that same conclusion.

so, why do you think they did it that way? for the look (retro gameboyish) ? not enough power to render a model ? they must have done it for the look.

Same reason we did something similar for Valkyrie Profile: Covenant of the Plume.
The first goal was the style, which requires an outline.
Rendering that many 3D models and then adding the outline (or rendering 3D models with the outline already added by doubling the vertices and reversing the winding and extruding them) is too much work for Nintendo 3DS and especially low-end iOS devices. If they could do it on every device the game supports, they would have.
Due to the style they want to achieve, it became too complex, so they pre-rendered it.


L. Spiro

Jeanne D'Arc for the PSP uses fully rendered models with outlines. I would see up to about a dozen characters with idle animations drawn in this manner. Drawing the outlines they probably used one render to a sprite, tinted it black and then drawing that sprite offset several times to get the outlined look.

Perhaps some more optimization may have been done in this game with caching the idle animations as sprites, because unlike GTA, in gameplay, typically only one character moved and attacked at a time.

Chinatown Wars was originally Nintendo DS game. The DS has a set limit of 2048 triangles per frame.

So if you want to make an open world GTA game on DS hardware, it means you place the camera above to control exactly how much of the world you can see, use most of your polygon budget on the cars, and then use billboard characters with 2 triangles each.

The DS also has fully 3D RPGs, but every character has 3 sided limbs in most cases.

Nintendo DS models are like a crash course in polygon reduction and silhouette preservation!