Hacker News Reader

Comment by wlesieutre | Hacker News Reader

Not as subtle, but Star Citizen's NDA test servers have hidden watermarks to track where screenshots come from. Someone pulled up the edges to show one here: https://i.imgur.com/FXzwlYd.png

You still get leaks because people take screenshots for their bug reports, upload them somewhere (even with an unlisted URL) for discussion on the bug reports forum, and then other users under the same NDA can download your screenshots and share them (rather than getting busted for leaking their own). So having a screenshot from a bug report leak isn't necessarily actionable for banning people.

But it's still useful to prevent, say, users under NDA streaming the super buggy builds on Twitch. Once they're relatively stable, CIG drops the NDA requirement, opens testing to a wider audience, then eventually push out to the main servers.

Doing this with a disguised UI component is a pretty cute implementation.

View on HN

As an early backer of Star Citizen, I wish they focus more on making the game instead of chasing down leakers.

Because SC is not actually a game. It has evolved into a dream business. They are selling dreams. So secrecy is most important, else the glamour collapse. They lost me when they hired Gillian Anderson, but DRM ridiculousness is on the same level. Such things should not be the focus of a studio working to provide an actual game.

I have thought about this and realized I don't have the knowledge of how Gillian Anderson and the other big name actors have been compensated for their involvement to be critical. Maybe the actors receive the rights to their hirez digital scans, for example. Maybe the actors are getting a cut of the eventual 'box office take.'

The digital scans on their own probably aren't worth much, doing anything with them is a lot of modeling and shader work tied to the game engine. My bet's on good old fashioned money, possibly a cut of Squadron 42 sales, but probably not a cut of the giant pile of crowdfunding.

Mark Hamil and John Rhys-Davies might have been an easier sell, having been in Wing Commander. For the rest, maybe they were excited to work on something that hasn't been done before. The quality of the characters shown in the trailer is crazy.

Such actors do not work for cuts of crowd-funded games. They may get residuals or a share of sales, but they were paid cash for their time. You can hire C and D-list with promises of future pay, but A/B demand money.

I was thinking more “both” than “only potential future money if we ever finish it”

A cut of future money means no money. Anyone who is even a bit savvy and carries the kind of clout that comes with a big bank account doesn't need to work for empty promises. If they were promised both, it would be really surprising to me if the immediate money wasn't already enough to pay them for their time. Given that that is the case, "both" suddenly doesn't make sense anymore.

Is this that No Man's Sky everyone keeps talking about?

I fire it up for a couple hours every patch, but you're right it's not much of a game yet. There's a certain sense of awe from the scale that other space games don't give me though, so I'm still excited about what it might become. And from a tech perspective, watching all their engine work has been really interesting to me.

Have you checked it out since they added Hurston?

What does that screenshot look like without edge detection?

Tracked down the original source on that image, here are both versions:

Unedited https://media.discordapp.net/attachments/271503305283338250/...

Watermark edges https://media.discordapp.net/attachments/271503305283338250/...

The grid of noise in the highlighted edges is from the jpeg compression on the screenshot, if we had the full quality original to work from it'd look cleaner.

Subtly visible in fotoforensics:

http://fotoforensics.com/analysis.php?id=9c41c242d475089ea8c...

How did they create the watermarked image without the watermark being visible to users? What kind of a filter or blending mode are they applying so that it can only be detected with edge detection? Anyone know?

See this post https://news.ycombinator.com/item?id=18649805

It's a low-alpha checkerboard, and it's actually quite visible.

What do you mean by pulled up the edges?

photoshop -> filters -> find edges -> some tweaking of the tone curve

I was being deliberately vague because I don't know exactly how the person who made the example processed it, but it's probably this. Find Edges and then maybe something else to clean it up or make the text stand out.

Someone else linked the image and its original file name was "gimp-2.png" so it wasn't Photoshop :P

Doing it in Gimp is easy. Load the image; Filters->Edge Detect->Sobel

I was curious and tried to reproduce it from other screenshots :) https://i.imgur.com/FBfwoLR.png

I guess https://en.wikipedia.org/wiki/Edge_detection ?

Damn... burning rendered tracking ID into the background. That's pretty crude.

They must've been pressed on time.

They already have a text rendering system for the UI, and it only needs to render to a texture once, then just keep it around and overlay it on every frame. Stuffed into a square format because that's how GPUs like textures.

At that point, the GPU doesn't care if it's overlaying text or something like a barcode. The main priorities would be resilience to image/video compression, and being large enough that you can't just crop it out of screenshots, while not impeding gameplay. Since it's only for the unstable testing builds, why spend more time than necessary implementing it?

Is there something I'm missing about what's wrong with this approach?

> something I'm missing

Instead of hiding an image of a text in a layer, they could've encoded the text itself (i.e. the ascii values) into the texture. That's trivial to do and doesn't produce visually-discernible artifacts as per above.

I agree - Steganography applied to the textures. I haven't seen steganography used that way before but if its done right you don't see it at all.

That wouldn't be as robust against compression artifacts and rescaling.

It probably could be if you represented the data with fairly large blocks, but I don't see the point either way. Sure they could have more directly encoded ASCII into texture data, but why bother when they already have a system in place for rendering ASCII to a texture as text?

It'd be overkill if they were building this feature from the ground up starting with nothing but a string, but they weren't.

Why wouldn't it?

They render ID into rasterized text (blocks of pixels) and blend in these blocks. But they could've just as easily rendered raw 0s and 1s of the ID into blocks the exact same way. It just would've not looked like text, but as a random pattern.

Maybe I'm not getting what you are saying.

But a rasterized text has a lot more information about each character than the raw ASCII values. If you halve the resolution or add compression artifacts an OCR will still pick it up.

If they rendered raw binary data they would have to add lots of redundant information as well as spreading it out over a large enough area. Maybe that's what you meant but didn't fit the 'trivially to do' part for me. Seems a lot easier to just render the text.

Reply on news.ycombinator.com