Hi,
There is nothing unordinary in such setup. You can link any color information to the GLOW output.
If you provide some example here, I'm sure forum artists might help with the node setup.

Добрый день, и добро пожаловать на форум!

Судя по всему, вы используете старую версию движка. Поддержка компрессии gzip - новая фича в Blend4Web 17.04.

Hi!

Actually, this type of limitations is a quite tricky thing, as Target Camera presumes the freedom of motion as it is an overview camera.
I've played with such setup for a little while, and I must admit that it is almost impossible to make a stable physical block for the camera. You never know how quickly the user will move the camera.

Attaching the example. This is the best I was able to achieve. No floor and ceiling - they are even less stable. Also, I recommend turning off worker in the app.init method with physics_use_workers: false. This gives slightly better results.

So anyway, this approach is not production ready and I would recommend using camera limits if possible.

I've checked the blend-file you provided.
Images have the greatest impact on the file size. For instance, the nappa_normal image has a size of 12.3 MB. I believe it is not required in your case as it can be replaced with a very small tiled image.

Furthermore, there are several images packed into blend-file with similar file paths. For example, cubemap_rought and cubemap_rought.004. Currently, Blend4Web unpacks them as separate files, which leads to exported file size growth. We'll fix this in the nearest release.

As for now, you can just remove duplicated images and use the original images instead.

[EDIT]
An even easier solution is to use unpacked images. You can click File-External Data and switch off the Automatically pack flag. Then just press Unpack All into Files. I'm not sure if this is fine for your development pipeline, but it will reduce the export size in a similar manner.

Yes, sure.
Just send your file to the support email you got when you acquired the license.

The problem with GL_LINES and GL_POINTS is that they are deprecated in modern OpenGL and their possibilities were truncated since OpenGL 3.0 and hence in WebGL 2.0. You can't set their width to a value greater than 1.0.

This is why I think the efforts to implement them in Blend4Web do not worth it.

Instead, it is better to further develop the existing draw_line API. You can always make a pull request to our GitHub repository.

This might have happened because meshes have UVs. Maybe, even several.
When there is no material applied to the meshes, UVs are being omitted. After attaching the materials UV-data goes into the final json/html.
This is just an example, but anyway, Blend4Web tries to export as least data as possible. Some situations like this are also possible.

Это можно сделать скриптингом.
Совсем без программирования тут не обойтись. Вот здесь показано, как менять курсор средствами JavaScript.

Если скомбинировать это с этим вызовом при движении мыши, то можно добиться того, о чём вы говорите.

В свою очередь добавлю, что действительно настройка ноды Parallax сейчас далека от интуитивной. Это обусловлено невозможностью выбирать текстуру в нодовых группах. Всё изменится в Blender 2.8, куда войдет наш патч добавляющий поддержку кастомных нод, написанных на Python.

В связи с этим, пока приходится применять такой хак с подключением текстуры и предполагать, что высота находится в альфа-канале.

Since "pick_center" reacts to SELECTABLE objects, I can achieve this by setting ALL objects, including walls etc., to "selectable". But is this the right way to do it? A large scene could have a massive number of objects… Or is there a way to test specifically if the object returned by "pick_center" is obscured?

Actually, this is a correct approach. Basically, testing object's visibility means rendering other objects which might block this object. And this is why you need to make walls selectable as well. If the walls are static and have the same material, this won't seriously harm the performance. Hint: to set some property for selected object you can ALT-click it.

What I would recommend anyway, is to try to minimize the amount of pick_object()/pick_center() calls. If you can do this only on click, this will be a good optimization.

And if you need the best performance possible and at the same time you want the approach to be as generic as possible, then physics is your choice. This requires more efforts though. You can check how this is done in Petigor's Tale (sources available in the free SDK under apps_dev/petigors_tale). Find the start_menu.js module and check the intro_load_cb method.