ZylonBane

Godzilla, step carefully away from the technical terminology, and leave it to the people who actually know what it means.

Godzilla-- the 2600 has had full-screen bitmaps since Combat, an animated sprite does not constitute "FMV", and what is this "hi-res mode"? Thomas-- Interesting... I've gotten better results when I increase the contrast. Otherwise images tend to just run together into a blob of color. What I do in PSP is go to Levels and set the Input levels to around 30, 1.00, 225. Gets rid of distracting dither in the solid dark and bright areas of the image. This is fun!

No, only ONE more question-- Why don't you just download it and answer all these questions yourself?

Uhh... do "SuperCharger" or "Cuttle Cart" ring a bell?

Ah, that's your problem. You're using such a desaturated palette that the primaries are bleeding into each other at split time. You should go with a pure saturated palette. Consider that all the color values are really doing is to flag which pixels go in which color separations. By using a desaturated palette and cleaning up the gray pixels in the split images afterward, you're achieving exactly the same effect as using a saturated palette in the first place. doron... you're an odd one for sure.

Dragon's Lair is a great cartoon, but as a game it sucks ass. Pure trial-and-error and rote memorization. But I admit it was a zilliion times better than BrainDead 13.

Let's see if I can explain this... I look at picture of strange girl, and don't care (except maybe to wonder what's in the part that got cropped off). I look at picture of strange girl, reduced to 8 flickery colors in horribly low resolution, and care even less. This technique is so all-around low-rez that it relies heavily on the viewer to mentally "fill in the blanks" in what they're seeing. And this is most effective if the image is one they immediately recognize. That's (mostly) what this is all about-- the thrill of recognition. Any image munged into this format is NOT going to be appreciated for its subject matter or composition, but for the fact that it's been rendered on the quaint little Atari 2600. It's like they say about Russian dancing bears... it's not the dancing that's remarkable, it's that they dance at all! Say, that reminds me... Who can be the first to identify all the pictures in the montage I posted?

Be careful what you wish for.

I've tried ordered dither a few times. The results are uniform all right... uniformly horrible. :wink: I've updated the image above to include an ordered-dither example. Thomas, that's weird about you getting gray pixels in your split images. Are you sure your custom palette only has 8 values in it? Are they pure (255,0,0) values?

Which paint program are you using for this? I've been using Paint Shop Pro, using the Stucki error diffusion filter. Floyd-Steinberg seems to create too many on pixels. And do you split first, then dither down, or dither down, then split? Seems to be a toss-up which way is better. In the sample below, I think the clown came out a little clearer using split-first, but all the "montage" pics earlier in this thread used the palette-loading method.

Ah, I see where you're coming from. What you say is true, but doesn't matter because this system has such incredibly pitiful color resolution that subtlety just doesn't enter into it. The test images I did above, I got the best results by cranking the brightness and saturation up to cartoony levels before converting. The only reason I desaturated my palette was to get a general preview what the end result would look like. I could have used pure RGB combos and still gotten the same color splitting... when there's only one red to choose from, the paint program doesn't care how saturated it is, just that it's the closest match. I suppose if you had some pics that were all dim or all bright and desaturated then it would be worthwhile to tweak the RGB that way, but those would be special cases. And as for this-- "And if the maximum amount of e.g. Red in a picture is quite low, wouldn't it make sense to choose different values for Red then. Right?" If I'm understanding what you're saying here, then NO. If a picture has no red in it, but it does have white, then you still need the red to create the white. The same goes for all the other color combinations.

It's weird but, for the purposes of this display technique, the 2600 palette is completely irrelevant except for Red, Green, and Blue. This is because, just like the monitor phosphors, all other colors emerge from combining these three. So, you can see white in an image even though the 2600 is never displaying white in any given frame. The custom palette on the PC side represents nothing more than all the possible combinations of RGB, and is only a convenience to ensure that the dither pattern is in sync for all three layers. It lets you dither, then split to RGB, as opposed to splitting and then dithering. You could create images that only interlaced two colors, but they would be very limited, eg-- Red-Green would only give you red, green, yellow, and black. It took me a while to get a firm mental handle on all this back when ColorView first came out for the 8-bits. Just forget about the palette and think of it as managing your own jumbo RGB phosphor triads.

Umm, don't you think it would make more sense to include pictures that more than one person would recognize? You know, stuff like this--

Nukey, where did you get the idea that anyone was talking about using playfield pixels? This has been based off the 48-pixel "6-digit-score" routine the whole time.

Paul-- Yeah, a 48x80 image done this way consumes ~1.5K. But for your RPG you could go for something a lot simpler, like a 16x32 image using only 2 layers. That would only take 128 bytes per image, plus you'd be able to set real colors for each layer. Eh, probably be simpler to just go for regular color bands. Another possible form of the CV technique is to go with 3-scanline-tall pixels. That way you could display all the RGB components at the same time, and just use the rolling interlace for color blending. EDIT: D'oh, Thomas responded while I was composing-- Here's the clown image: http://rsb.info.nih.gov/ij/images/clown.gif This pic has a bit of history. It's one of the images that was used a lot in print ads back when 256-color graphics cards were new. That's why it's only 320x200... MCGA resolution! Fortunately this is already the correct aspect ratio for the 2600. Baboon.jpg (on the same site) was also used for these ads. As far as individually tweaking the RGB colors... this is exactly the same as going into a paint program and adjusting the RGB levels. You end up affecting the tint of the entire image. It would only be practical on 3-color images that didn't use any blending (or very carefully controlled blending). Another thing, it turns out Robert M was initially correct about being able to load a custom palette. I was so caught up in the mechanics of channel splitting that I didn't see it at first. So the way I'm creating test images now is: 1. Crop/downsample image to 48 x whatever 2. Load a custom palette I've created that only contains RGBCMYKW (slightly desaturated), using error diffusion dithering. 3. Split to RGB. 4. Well, here it gets more involved. Progressive scanline masking and recombining and such. Before (as with the Clown demo), I was splitting to RGB first and then downsampling, which caused the diffusion dither to be out of synch in the three images. Whoops!

Wow, rapid turnaround there Thomas! Looks nicer than I thought it would, what with the low resolution dithering. Might be fun to convert a few more images and put out a little slideshow demo. Any suggestions?

I didn't say you were. And it's not my method. A nice bloke by the name of Jeff Potter came up with it years ago. I also didn't say your method doesn't provide any color blending, just that it doesn't offer the type of blending that ColorView does. The Atari computer demo scene has been over this territory a LOT, and they almost always go for interlacing for the pseudo-color screens. I have to think there's a good reason for that.

Interesting! Unfortunately, since you're swapping between solid R/G/B frames it does flicker severely and you don't get the color-blending benefit of the ColorView approach. Attached is a ZIP of three pre-interlaced clown images, and a GIF preview of what it might look like on a 2600 (highly doubtful any browser will animate it at the specified speed). 2600clownrgbs.zip

Yeah, it's almost as if Xevious is some kind of unreleased prototype or something!

Since this is a one-color-per-line thing, you can't just reduce the whole thing to a small palette. You have to split each image up into 3 grayscale R/G/B images, reduce them to your target color depth (for a 2600, that would be 2-color), then interleave them back together. You're essentially creating meta-triads... RGB triads at the pixel, rather than phosphor level. And this is how the above image would look on a 2600. Scary, isn't it? This is why you need active interlacing to recover the resolution you lose.

Rob-- Exactly! Although it's 8 colors (2*2*2), not 6. I'm attaching a proof-of-concept image. Take a look at it, and try to guess how many colors it's using. Then load it up in your paint program of choice and be amazed.

Ok, I'm not really going to have to explain the concept of additive color blending, am I? Haven't you ever looked at a color TV really close? "The TV can display a seperate color on each phosphor already...why would you need to incorporate 3 phosphors into one? Just to increase the color pallette? You would still only have one color per phosphor, right?"

Where did you get "without interlacing" from? This method requires interlacing and I've mentioned it several times.

The pixels aren't colored independently... the scanlines are. And it's more of a shimmer than a flicker. At any given jiffy you've got a complete interlaced RGB image onscreen. The interlacing is just done to increase the vertical resolution and blend the colors together.