Re: [Gimp-developer] Blend Tool UI

[Simon Budig <simon budig de>]

peter sikking (peter mmiworks net) wrote:
> > Hmm, you might have misunderstood what I meant by shapebursts. The
> > shapebursts are special gradients that mimic the shape of your
> > selection (currently labeled "Shaped (angular)", "Shaped (spherical)",
> > and "Shaped (dimpled)"). Anyway, at this point I'm really conflicted
> > as to what should be done with them. I'm not sure whether they belong
> > in the blend tool or not right now.
>
> OK, I should have consulted the manual and now I have done it.
>
> I am now completely convinced that the shape burst belongs
> in the gradient tool. there is nothing contradictory about
> that. the gradient tool applies a gradient fill (as everything:
> modified by the selection) and Shape is a fill ‘mode’.
>
> and when talking interactive: next move would be to control
> these funky fill shapes on the canvas with a handle or two.

A little background on shapeburst modes to get an idea about what might
be adjustable by the user.

When filling with a shapeburst mode, a distance map is computed, where
for each somewhat-selected pixel a distance to the nearest
unselected pixel is computed. The "distance" is computed using a
manhattan metrics (dx+dy), with some funky shit for partially selected
pixels. For the latter part I am currently not convinced that it works
correctly and consistent.

Computing the distance map is an expensive operation (and unfortunately
the manhattan-metrics approach introduces some quite visible weird
unexpected artefacts). But this actually needs to be done only when the
selection changes, I think right now it is computed every time.

A huge improvement would be to change the distance metrics to something
euclidean. That'd give way more predictable and useful results. However,
this is not easy at all apparently...

The currently selected gradient is then mapped to the distance (think
gradient-map). The "angular", "spherical" and "dimpled" variants are
basically specifying what function is used to map the distance to a
position in the gradient.

This *could* be something more funky, similiar to how I used the curves
tool to determine the shape of the golden text in my ancient
golden-text-tutorial: http://www.home.unix-ag.org/simon/gimp/golden2.html

"angular" would be a linear mapping and I believe that "spherical" and
"dimpled" basically would be quarter-circle mappings (in different
directions).

Changing this to a generic (editable?) curve should be fairly
straightforward.

Bye,
        Simon
-- 
              simon budig de              http://simon.budig.de/