prophet

I'm always a fan of using an image as it's own Mask if possible.
Here's my rough take…
• Dupe the top layer and Rasterize
• Apply various filters to get a high contrast Black and White proto-mask (used Exposure and Black & White and merged as I went)
• Clean up
• Invert
• Clean up again (you could spend a lot of time cleaning and retouching depending on how precise you need the Mask to be)
• Rasterize to Mask and nest into top layer
 
Screen Recording 2023-08-18 at 1.47.17 PM.mov

Just roughly masked with Flood Select on the dark areas + Brush.
For the water reflection in blend mode Lighten: a sky copy (flipped, moved, stretched) with a separate mask.

I think "minimize" is the wrong term here. I can see that the OP uses the scroll bars to shift the image while selecting. In one instance, the selection process remains active, in the other it doesn't.

I'm not suggesting any functionality of the Gradient tool be moved or changed in any way. Merely, that rather than having to make the extra click to pop open the fill (or stroke) gradient context toolbar editor, that interface could live in it's own panel.
It would make my workflow a bit quicker as it would align those options with existing panels in the Right Studio rather than appearing on the opposite side of the workspace. But to each their own.
Valid, but a panel could give more options in that regard without taking up much more space. Or remove the color popup and use the existing Color Panel in the same way as one does when just making adjustments with the tool itself.

Yikes! That's a pretty big UI blunder.

I also think a Gradient Panel would be nice. Redundant? Maybe a little, but being able to quickly see all properties of a gradient fill of a selected object in a panel saves the need for an extra click on the Context Toolbar.

A fair point. I don't suggest doing away with the context toolbar stuff. As @thomaso says, there are multiple places where the same thing happens. Why not add Gradient to that methodology as well?
 

Then perhaps "fix" is the wrong word. I would like to see…
 

Brilliant! The perfectionist in me will want to go in a tweak all the little crevices, but this technique is a great starting point.
I think it also proves the capability for auto-trace already exists in Affinity, just needs some polish to bring it up to the "full tool/feature" level.

To smooth the jagged edges, go to Designer persona and apply multiple steps of the following until you're satisfied with the result:
select all nodes > apply Corner Tool > Rounded > ~20 px (or about 2 mm @ 300 dpi) > Bake Appearance select all nodes > click the Smooth Curve button a few times select all nodes > click Convert: Sharp button repeat steps 1 > 2 > 3 > etc. > 1 > 2 optionally apply Contour Tool or similar to your liking Works surprisingly well within a minute or so.
Black is the initial jagged "million-nodes" vectorization of the hidden pixel layer, red overlay is a smoothed copy using several passes of the above workflow:

Not all that shabby, huh?  

alternate step 5 in @Aammppaa's method
5. Select all triangles, make sure "Transform Objects Separately" is on, scale one triangle as desired

Well spotted. I can get even spacing between 2 triangles in a "row", but always a bigger gap between 2 "rows".
Contour is definitely the way to go.

I keep viewing this thread to get it out of my "Unread Content" feed.

Well spotted. I can get even spacing between 2 triangles in a "row", but always a bigger gap between 2 "rows".
Contour is definitely the way to go.

I keep viewing this thread to get it out of my "Unread Content" feed.

This was my initial thought, but it is difficult to get this to work, as the Transform center would have to be at the center of the triangle, which is not the center of the triangle's bounding box. As a result you end up with an uneven gutter. 

alternate step 5 in @Aammppaa's method
5. Select all triangles, make sure "Transform Objects Separately" is on, scale one triangle as desired

And for a spot color background, no "empty" Levels layer is needed, but one must still make the underlying image into a group of 1 and apply multiply to the group.

A small "improvement" on this issue with 2.1.
It would appear one no longer needs to add an "empty" levels layer, make the image into a group of 1, and set that group to multiply.
Now you can merely duplicate the image, add the shadow to the image at the bottom, and set that image layer to multiply.
A bit simpler, but you know what would be even more simple?…I bet you do.

I don't do this kind of work much, but I find working with a curve directly in Photo at the desired size lets me see the actual pixels/hinting in real time. I can see where trouble spots are and move/nudge nodes to minimize problems.

I like to use the "Transform Object Separately" in my workflow.
- Grab everything
- Flip
- Click "Transform Object Separately"
- Flip again
Just have to remember to untoggle "Transform Object Separately" before proceeding with other transforming work

+1 for this.

How would you know how much "thought" was put into a piece of art?

I can duplicate this on occasion. I think it has to do with the zoom level of what your viewing and what the dropper is actually capturing. I can even get the dropper to sample the anti-aliased color when it doesn't actually show in the magnifier.

Lovely, but the orbital mechanics and gravity fluctuations of that planetary system must be a nightmare.