Difference between revisions of "Talk:Curve Gradient Layer"

Latest revision as of 09:22, 29 January 2008

IRC log

pxegeek: Hey dooglus - does the broken heart fix also fix creases in perpendicular curve gradients?

dooglus: pxegeek: I've avoided looking at perpendicular curve gradients. the fix may fix them, but I really don't think it will - the changes were inside an"if (!perpendicular) { }"block

dooglus: pxegeek: but a similar fix for perpendicular gradients will probably be quite easy. could you report it as a bug, and include a URL for an example creased .sif file?

pxegeek: You can just check perpenidicular on Rore's heart, and see the crease at the top of the screen

dooglus: ok, but rore's heart *is* creased, so you would expect the perpendicular gradient to reflect that...

dooglus: or am I missing something?

pxegeek: So if you draw any concave curve, there's a crease, but it's not all the way to the line - it starts a little way out

pxegeek: Shouldn't there be a way to control how smoothly the gradient transitions?

pxegeek: Or should I just stick a blur layer over the top?

dooglus: the gradient transitions along its whole range, from the start of the bline to the end

dooglus: there's not a lot of scope for specifying anything there

pxegeek: I'm sure what it's doing is technically accurate, it just doesn't look elegant.

dooglus: heh - I'm not sure of that at all :)

pxegeek: Like changing gear without a clutch

dooglus: sure

pxegeek: Stick a blur layer 16ptx16pt on top of rore's example with perpendicular checked

pxegeek: That's the smoothness I'd hope to see instead of a jagged line.

pxegeek: Want me to post a'before and after'?

dooglus: I'll make an example picture that we can talk about

dooglus:

dooglus: the green'ray'and the red'ray'are both perpendicular to the bline

dooglus: since the bline is convex, they point towards each other.

dooglus: the place where they meet will be make a'crease'

pxegeek: Right, but within the radius of curvature, the gradient is smooth

pxegeek: I guess what I'm thinking of would be a more advances type of gradient that fades off the further you get from the line

dooglus: is this what you're looking for?

dooglus: I put a super sample layer over it

dooglus: what you're asking for is a perp.curve gradient and a regular curve gradient combined somehow - one of the compose modes would do that for you wouldn't it?

pxegeek: Maybe

dooglus: do you think it's about the radius of curvature? I'm not sure

pxegeek: /me still futzing with blur layets

pxegeek: the radius of curvature just means that the perpendiculars to the line don't meet within that zone...

pxegeek: obviously

pxegeek: Ah synfig's finished building!

dooglus: oh, I see what you mean

pxegeek: And much smaller size without the debug options set

dooglus: the gradients meet at the centre of the circle

pxegeek: yup

dooglus: I think this is exactly what supersample is for

dooglus: you'll get a smoother crease...

pxegeek: Yes - that looks smoother

dooglus: it's red meeting green - that's never going to be'smooth'

dooglus: I used clashing colors to illustrate the point - typically it'll be grey meeting a slightly different grey - but still a meeting of different parts of the gradient

pxegeek: But you can still do a smooth fade between red&green....

pxegeek: :)

dooglus: with antialiasing, yes

pxegeek: It's just a question of how much area you want spread it over

dooglus: the curvelayer does this: for each point, which bit of the curve is closest to it. draw a line through the curve to the closest point.

pxegeek: I'll explore the compositing linear and perpendicular curves idea

dooglus: I've found a bug in the perendicular c.g. though

dooglus: pxegeek: shows how the perpendicular gradient works (and how it breaks - what's the big white patch??)

dooglus: pxegeek: if you can suggest how to'blend'the colors in that example, I'll have a go at implementing it... you've got a whole bunch of different colors shooting out at different angles - how do you make that look smooth?

dooglus: I'll check the logs tomorrow - I'm off to bed now. maybe the wiki's curvegradient talk page would be a good place for this, so we can find it easily later.

dooglus: night all :)

pxegeek: 'night dooglus

pxegeek: Yes, that example may be tricky to smoothly transition ;)

Example Buggy SIF File

The black and white lines indicate the extreme left and right edges of the gradient. The large red area between them shouldn't be there:

File:Perp-curve-gradient-2.sif

Fixed the bug. It wasn't using the last bline in a blinelist when the blinelis was looped: