Great, entertaining image. Bags of character in the face, powerful looking buttocks .

Big things; the pose/anatomy strikes me as a bit stiff/awkward, everything looks fairly human apart from the legs and head. I'd get a nice silhouette going from a thumbnail before rendering and also endeavour to incorporate a more bestial aspect; both in the anatomy and pose. I gave it a quick go moving the tail somewhere more interesting/distinct, giving him a crouch and lending more weight to the face which I absolutely love.

Little things; the definition/shine on the muscles kind of make it look like he's wearing a grey wet-suit; not very furry, but I guess you could find a nice toony compromise. There's a discussion of this issue and example solutions here.

Yeah, as nice as some of the temple stuff looks, perhaps distilling it down to a few show-pieces and going flatter elsewhere might be better overall as well as being less labour-intensive . My first reaction and I think that of other guys earlier was that select simplification could read nicer, especially in the area above the eye-socket.    

There doesn't seem to be much in the background that really takes advantage of the isometric angles as per the foreground. It might help both with simplification and also consistency if some of the other temples and rock formations were seen to exaggerate this alignment/aspect.

I didn't mean sow doubt as to perspective; you're pretty invested now and it does look good already.

Improvements a gogo them some sexy temples, much better rocks and the reinstatement of the palette brings the whole thing alive. Minor niggle; I preferred the mirror-like still water.

To clarify what I meant about the perspective; I'm advocating having the isometric projection of the island resemble the skull and not so much the island itself. I think this would be more effective in terms of the duality of the image and also the symbolism; it's not a skull-shaped island but the viewer is interpreting it as such.

I am suggesting a (small) loss of readability with your current approach. You mentioned earlier you didn't want it to be 'obvious iso', I guess a lot of people will just see a distorted skull. Removing detail from the equation by squinting, or looking at it from a distance as you might well do on a shirt, is it as skull-like as it could be?

Edit- Actually that sounds a bit strong. I'm just trying to explain what I meant and offer an alternative vision, not dictate that one way is objectively 'better'    

I see what you mean now about the map vs projection but I'd reckon the idea will be lost on the great majority of viewers and I think the image of the skull only becoming apparent from the one viewpoint, Holbein style, might be more effective.

Regarding trees etc. sorry, I must have skimmed a bit; the first image is well catered for in this regard . Other stuff; to my mind the choice of more rounded rock formations is puzzling considering the pronounced planar qualities of the source image.

You seem already to have incorporated more of that rock-cut temple idea, sounds great to me

Yeah, I love the highly detailed stuff and I meant it might be effective to script more of the glitchy effects on top of sound pixel technique

My examples are all by hand, so to speak after having a look at some actual glitches. It's pretty simple stuff that you could easily do manually if you had to (basically just dicking about with the rectangular selection tool) but scripting them should be relatively simple to do and have the advantage of that de facto randomisation and malleability as I was talking about (maybe a real programmer could chime in here ).

Edit - Ah, that was a bit blasé sorry, blame the hour I get what you mean now with the potential of the glitch to blur the lines from collision masks; that would definitely take some tinkering to get right. Glad my example was helpful, here's what I did -

• Gave bestial heads full compliment of range, shortened ram's face & positioned snake's head at more of an angle, could have done this with the former also
• Layered offset is just shifting sections an arbitrary amount in either direction; for tile scrambling I put up a 8x8 grid and Control-C/V'd that sucker. For the palette swap I copied the whole image, stuck it onto a separate layer underneath and selectively deleted the original.