New chicken portrait

[Foaly]

Hello everybody!

I have been following the 0 AD development for a while and I find this project just absolutly amazing!

The other day while playing I saw something, that I thought I could improve. So I gave it a shot. The current portrait for the chicken, is just a render of the low poly chicken model. The model works really well in the game, but the image looks a bit too low poly for a close up image. This is what the current portrait looks like:

I wanted to replace the image with a nicer looking one. Since I am not a very skilled painter, I simply took an available chicken model in a higher resolution under CC0 license (http://opengameart.org/content/chicken-animated), rendered it and edited it with GIMP. The result looks like this:

 

Since this is my first contribution I don't really know the whole process of adding things to the repository (I hope this is even the right thread...). But before I get into all that, I wanted to get some feedback on whether it looks good or not. And maybe some advice on how to continue to get this submitted.

 

Thanks in advance,

Foaly

[niektb]

Hi Foaly!  Thanks for your interest in contributing to this project!

Hehe, certainly the old one can be improved  This thread works perfectly fine. When your contribution is in the final version (and good enough ofc) one of the team's artist could commit it to the repository.

I see two possible points of improvement:

1. The glow is not so much like the in the other portraits in the game (when I come home I'll post a few FX presets, I use Gimp too)
2. The proportions of the chicken, it would be nicer if it had some more width (with respect to the height).

[niektb]

Alright. Normally portraits consist of 5 different layers. I assume that you've installed the LayerFX plugin from here: http://registry.gimp.org/node/186 (make sure you install the  layerfx.2.8.py  and not one of the others). I apply the effects on 256x256 and then size the image down to 128x128 afterwards.

The bottom layer is just straight black. Nothing special here.

Now starts the fun. The second bottom layer is an Noisy Outer Glow effect. The color used is 7e8862. The exact glow size can be varied depending on the shown object but here are the parameters that work in almost all occassions:

Above it is a small Smooth Outer Glow layer. 

Last effect is a Drop Shadow. This is to create a sense of 3D in the image:

Then on top comes the portrait itself. Always make sure that the FX are applied to this layer. Also when resizing the layers, make sure that do a 'resize layer to image' action on the FX layers.

Here are a few examples on what the final portrait could look like (the bottom two aren't sized down, and these also used exact the same parameters as listed above. The others have slightly different sizes / colors / noise amounts):

 

I hope this helps!  

 

[niektb]

Oh, I was asked to link to this topic too  

 

[Foaly]

Hi!

Thank you very much for the nice welcome and the feedback! Also thank you for the detailed description on how to do the glow around the object. I gave it a try and it worked perfectly. Maybe the explanation should be put on the wiki, because its very good and I couldn't find it before (maybe I didn't look properly though...)

So here is my new chicken model. I tryed to implement all your suggestions. Tell me what you think:

I am a developer, so I do know how to commit things to repositories, its just that I usually work with git not svn. But mostly I am not familiar with your process (open a trak ticket first?, send a pull request?...). I am sure you guys can tell me the way to go, once the image is finalized.

But first people need to tell me what the think

Also I signed the legal waiver you linked.

[GunChleoc]

It is nice and crisp now - I think it looks too dark though. Probably best grab a few icons and position them around the icon you're working on to make the brigtness/contrast/saturation click?

[niektb]

No, it can't be found on the wiki (yet) I've put that post together just for this topic.  

Glow looks much more decent now! Chicken Pose is also a good improvement!  But agreed with GunChleoc that the brightness and contrast (of the chicken layer) should be tweaked a bit. (and maybe I would make the bottom noisy glow layer slightly bigger, but that's not a big issue)

You could open a Trac ticket, and post the image there. You could post it here. (as long as I can find it  ). Pull requests aren't possible, only patches. But these struggle with binary files (such as images)... (hence these normally get committed directly by a team member.)

[Foaly]

Yeah I totally agree. It looks fine on its own, but next to the other portraits it looks kinda faint and dim. I went ahead an redid the lighting in the original blender model and also adjusted the brightness and contrast in the final rendered image. I also increased the glow size a bit. Here is the result:

If the normal process is that a team member commits the image, then I guess there is not much else to do for me Unless somebody suggests some more improvments.

If my name is included in the credits (Foaly), I think it would be fair when the creator of the original chicken model (CDmir - Čestmír Dammer) would be included as well. Even though it is under CC0, it feels kinda wrong to take all the credit if somebody else has done 80% of the work. Or do you have a different policy for derived work?

Edited June 19, 2016 by Foaly

[sanderd17]

I'm wondering if we shouldn't bring all new portraits to 256x256 instead of 128x128. If we plan to display them in a sort of encyclopedia, it could probably take up a bit more screen space than currently.

[wowgetoffyourcellphone]

10 minutes ago, sanderd17 said:

I'm wondering if we shouldn't bring all new portraits to 256x256 instead of 128x128. If we plan to display them in a sort of encyclopedia, it could probably take up a bit more screen space than currently.

Hmm, maybe. Biut will resizing (scale down) for match GUI cause artifact or other undesired thing (like look smudgy)?

[sanderd17]

There are already 256x256 portraits, and IMO, they look fine: http://trac.wildfiregames.com/browser/ps/trunk/binaries/data/mods/public/art/textures/ui/session/portraits/units/256

The advantage of halving the width and height in the GUI is that it exactly merges 4 pixels to 1.

[wowgetoffyourcellphone]

2 hours ago, sanderd17 said:

There are already 256x256 portraits, and IMO, they look fine: http://trac.wildfiregames.com/browser/ps/trunk/binaries/data/mods/public/art/textures/ui/session/portraits/units/256

The advantage of halving the width and height in the GUI is that it exactly merges 4 pixels to 1.

So, unit portrait in match GUI is presented in 128x128?

[sanderd17]

Hmm, apparently not, the big icon currently is 94x94, and the small icons are 30x30. But as it doesn't give issues, it certainly shouldn't give issues with a 256x256 image.

[Foaly]

I found another model form the same artist, also under CC0 (http://opengameart.org/content/old-deer-male).

I thought it would be a nice replacement for the deer portrait. This is what the old one looks like:

This is the one I did:

Tell me what you think!

 

edit: this morning I tryed another version with different lighting and contrast. I like it better. What do you guys think?

Edited June 13, 2016 by Foaly

[niektb]

@Foaly: Would you be interested to participate in a small test? (Yesterday we had some debate in the team about the desired portrait image size and decided to do a little test.)

In case you do, would you render the chicken in 2 more sizes (256x256 and 512) and post them here? (I can put the glows around them).

We would like to investigate if there's visible difference in-game between a native-sized render (128x128) and a portrait where the render engine scales the image down (from 256 or 512) 

 

Nice deer btw, definitely the second version! (though I'd prefer a little more glow, are you sure that you applied the suggested parameters at 256x256 and not a higher resolution?))

[Foaly]

First of all thank you for the feedback! I have to say I am quiet impressed by your eyesight I indeed applied the parameters at the wrong size... (512 instead of 256)

Of course I would like to help out with your test. I did some new renders of both the chicken and the deer at different sizes. For the deer I also fixed the glow. I found for 512x512 that size parameters between 60 and 50 worked well for the bigger glow (the one with the noise) and size values between 20 and 18 worked well for the screen glow. Also I generally made the size of the drop shadow about 3/4 smaller than what you suggested. Otherwise it seem to dominant to me and didn't bring out the 3D'ish look as well.

So here are the new images:

If your post gets transformed into a wiki page I would like to suggest one improvement. After you created the three glow layers (larger noisy glow, smaller glow, drop shadow), it is important to right click on each of the three layer in the Layers Tab and click "Scale layer to image size" before scaling the layers down. Since those three layers are bigger than your object layer, you get weird offset errors by 1 or 2 pixels otherwise (probably due to rounding). This is espacially noticable with the drop shadow layer. Shrinking the layers to image size fixes this.

[niektb]

Ah, you're great!  

Actually, I export the image as a png and size that one down separately (so I always have a higher res source)

[Foaly]

Thanks! Let me know how the test goes.

I saved the gimp .xcf file at 512x512 and scalled down before exporting to .png, so I guess it comes out to the same.

[niektb]

Alright, here are the results. Make sure to view the attached screenshots at native resolution so you can see the results properly :

I've used the 128 icon for the house, the 256 icon for the storehouse and the 512 icon for the farmstead so I could compare them side by side.

When viewing the buildings in the build in the production queue, the 128 chicken looks best, the others show aliasing (512 has the most aliasing)

However, when looking at the portrait in the information panel (as shown below) it looks like the 128 is slightly blurred (when compared side-by-side, on its own it looks pretty decent). The 512 on the other hand still looks a bit aliased. 256 gives the best result.

So clearly, higher resolution does not give necessarily better results. Instead, one should search for the size where the least resizing is needed to get the best looks. 

So, what resolution is better? 128x128 or 256x256? Actually, I think that 128 looks slightly better. I do however think that 256 is more future-proof (higher resolution screens get more and more common so I think I wouldn't be surprised if the GUI gets scaled up (to prevent from getting too small) in the nearby future for HDPI screens.

What resolution looks best according to you, guys?

[Palaxin]

The engine doesn't support anti-aliasing? Can we change that?

I think 128x128 and 256x256 are pretty close, so I would also favor 256. At least 4K displays will need this resolution. According to this site 4K displays already reach 23% market share in the TV segment. I think in a few years this will be standard for both TV and PC displays...

[Foaly]

I actually have to say that for the icons I can't see a difference at all.

For the portrait (in the info panel) I think the 256x256 portrait actually looks the best. It looks sharper and more crisp than the 128 one. The argument that it is more future proof if the GUI gets scalled up later on also speaks for 256.

Have you played with OpenGL texture minifying and magnification functions? There is nearest neighbour (default, GL_NEAREST) and bi-linear interpolation (GL_LINEAR) for both minifying and magnification. For minification there is also mipmapping with different combinations of interpolation methods. You can read more about it here under GL_TEXTURE_MIN_FILTER. Maybe mipmapping is a little overkill for this use-case though, as it can get quiet memory intersive and there is a bunch of portraits, for which you usually only see a portion. But the linear interpolation might be something worth looking into. I've had good results using it in the past. Then again maybe it is already done. I don't know the engine you use at all Just throwing some ideas out there!

Edited June 14, 2016 by Foaly

[Foaly]

Hi I went over some more of the portraits and the one for the horse and one for the tiger really stood out to me. They seem to be created from the games 3D models. I really like them. They look great even up close, but I think they would look even better with some different posing, lighting and the glow added. Here is what they currently look like:

Now my question is: Where do I find the models for that? I looked around the art repository for the tiger model. The only thing I could find was this blend file, but it is an untextured version and it also looks a little different, so I suspect it might be an old model. I also tryed to open the tiger.dae from the game repository, but blender wouldn't open that file. Similar for the horse. I found a bunch of horse files in the .max format in the art repo, but I can't open them in Blender. So I don't know where I should get the models from. Could somebody help me out?

[niektb]

My Blender is able to import the .dae so could you describe what's going wrong?

 

[Foaly]

I downloaded the tiger file from here. Then I open blender, hit import and select the file. Blender then tells me that there was an error and I should go check the console. the console prints:

Schema validation error: Critical error: Additional: Entity 'larr' not defined

Schema validation error: Critical error: Additional: Entity 'rarr' not defined

One thing that come to mind is that maybe I shouldn't download just the file but clone the whole svn repository. Are there maybe some dependecies?

Edited June 17, 2016 by Foaly

[niektb]

That's strange... I've imported it for you. Can you open it now?

tiger.zip