Best Of
Re: Sketchbook: Celosia
I had a sudden thought this weekend and solved three unrelated workflow annoyances. 😄
First one: Sometimes I sink edge loops of clothing to create the appearance of invisible seams.
My usual approach involved manually prepping the mesh in Marvelous Designer with extra edge loops, exporting an unwelded mesh and manually grouping edge loops in Blender to offset them along normals. Do you need to alter some part of the clothing? Shit luck, enjoy redoing all of that.
I got Blender to do all of it by dropping a single geo node on the clothing now; no prior prep in MD or Blender required. It's configurable to accept manual loop additions and removals and works by using boundaries from uv islands, which due to using actual sewing patterns to model clothes tends to make UVs coincide with most invisible seams.
(these clothes are direct exports of early drafts in MD hence the weird topology)
It was a pain to get the loop cut and slide to work correctly and without destroying UVs, geonodes still don't have anything for either op.
The same boundary edge loops selection logic led to the solution of another issue: Sometimes I want to just smooth surfaces without subdividing them; or subsurf them without changing the shape of their boundary loops at all.
Now I can.
Custom smoothing node with pinned boundaries ◼ x standard subdivision surface ◼
No edges grouping or manual creasing required, although you can manually include boundaries by marking the area if you wish. This node can also restore some of the volume lost to the smoothing process.
When I went down this rabbit hole it wasn't with the intention of solving my crumpled hair cards issue, but the solution is somewhere in there. The custom surface smoothing works wonders for small deformations of the cards:
And I got something more or less working for the edges. I'm not happy with it because they lose too much volume (the trick I'm using for surfaces doesn't work in this situation), and I'll probably have to pull data from each boundary loop individually because being lazy and using values from the control curve to restore volume didn't fly. It's either too little or too much:
First one: Sometimes I sink edge loops of clothing to create the appearance of invisible seams.
My usual approach involved manually prepping the mesh in Marvelous Designer with extra edge loops, exporting an unwelded mesh and manually grouping edge loops in Blender to offset them along normals. Do you need to alter some part of the clothing? Shit luck, enjoy redoing all of that.
I got Blender to do all of it by dropping a single geo node on the clothing now; no prior prep in MD or Blender required. It's configurable to accept manual loop additions and removals and works by using boundaries from uv islands, which due to using actual sewing patterns to model clothes tends to make UVs coincide with most invisible seams.
(these clothes are direct exports of early drafts in MD hence the weird topology)
It was a pain to get the loop cut and slide to work correctly and without destroying UVs, geonodes still don't have anything for either op.
The same boundary edge loops selection logic led to the solution of another issue: Sometimes I want to just smooth surfaces without subdividing them; or subsurf them without changing the shape of their boundary loops at all.
Now I can.
Custom smoothing node with pinned boundaries ◼ x standard subdivision surface ◼
No edges grouping or manual creasing required, although you can manually include boundaries by marking the area if you wish. This node can also restore some of the volume lost to the smoothing process.
When I went down this rabbit hole it wasn't with the intention of solving my crumpled hair cards issue, but the solution is somewhere in there. The custom surface smoothing works wonders for small deformations of the cards:
And I got something more or less working for the edges. I'm not happy with it because they lose too much volume (the trick I'm using for surfaces doesn't work in this situation), and I'll probably have to pull data from each boundary loop individually because being lazy and using values from the control curve to restore volume didn't fly. It's either too little or too much:
Celosia
2 ·
Re: Sketchbook: Celosia
Thank you, good catch! I used to be more attentive when reviewing messages. 😆iam717 said:congrats on getting this far.Nothing really to say but i had to figure out the 1st post above at this part:" but in these awful this seems a slightly more secure" awful - times, was missing.Nothing major just stood out to me, neat thread, i am decidedly on an off with blender so one day this might come in handy.
------
Tiny update just to say there's one!
I scratched a hair alpha directly in Blender with a mouse and no stabilizer to better understand the type, scale and amount of variants I need for each type of hair layer. Looks better than I expected, hahah, I'm shocked the baby hair isn't a full disaster, and it's not as see through as I expected given the low amount of base layers.
For context, these are going to be my first proper realtime hairs. I've done one or two solid mesh bun styles for game jams before, and I'm used to grooms or hair for sculptures like in that first bust, but realtime? That's new.
Speaking of sculptures, I'm doing my best to not get sidetracked testing the system to create hair for them. 😂
This one was my first Blender bust and you can see that with the min twist annoyances and everything at the time I just gave up on the back. I feel using these nodes would go so smoother, I'd have room to experiment with placement and giving the hair a nice flow instead of spending 90% of the time pushing control points around. Refrain from doing thin loose strands, throw a remesher over it to join them into a mass and I might even have a way to do some very fine-looking hair for printing.
Edited to add today's QOL node: Hair clamping for baking.
Long overdue since I baked those hair locks weeks ago and got a bit of clipping at edges. Runaway strands are clamped and flagged so I get to decide how smooth them. Now I can pack as many card variants as possible without anything getting visibly clipped or accidentally invading their neighbours' real state.
Celosia
1 ·
Re: Zero views on my ArtStation profile despite having a Pro subscription—any advice?
It seems that somewhere in the last 4 years, the way in which Artstation presented new art changed significantly. I'm not sure if it was their allowance of AI art or the stronger promotion of AAA studio art(or both), but I have noticed a substantial shift. My uploads from 6-7 years ago received far more views and likes(and subsequent followers) than my much more recent work. This is despite my older work being vastly inferior in terms of modeling, baking, texturing and rendering.
I think this has caused alot of smaller artists which don't work at an EA, Rockstar, Naughty Dog, 343i etc to just drop off. Leaving mostly AAA artists whose work is insanely good and who hardly view other peoples work. Just my opinion.
I think this has caused alot of smaller artists which don't work at an EA, Rockstar, Naughty Dog, 343i etc to just drop off. Leaving mostly AAA artists whose work is insanely good and who hardly view other peoples work. Just my opinion.
guitarguy00
1 ·
Re: How do you make the 3D pipeline less painful?
Now they are sending in AI so lifelike they can pass as humans.
Lamont
1 ·
Re: Error analysis Blender has stopped working / Analysis of similar errors / How to protect your files
"Solving the problem without pumping."
This doesn't make sense in English. I think you mean instead:
"Solving the problem without dumping data."
This doesn't make sense in English. I think you mean instead:
"Solving the problem without dumping data."
Eric Chadwick
2 ·
Re: AI Art, Good or Bad? A (hopefully) nuanced take on the subject.
zetheros said:Programming is more about problem solving than actually writing the code. AI could be a tool in your kit to help with finding solutions, but no way should it actually be doing the code writing
Actually, finding solutions is exactly not the use case for LLMs. Good coding is more that writing lines that perform a task successfully, it's about designing solutions. That requires thinking. Thinking is something LLMs aren't able to do, and will never be able to do. They're a probabilistic algorithm autocompleting your inputs by drawing from a large but damn well compressed pool of data. That's why they're not creative and are unable to output they don't "know" something unless it's hard-coded into them or that happens to be what's in the data, eg the gross majority of answers in the training set were "I don't know", so that's the pattern adopted for this association. They don't have a concept of anything, and no matter how much data or computing power is thrown at them it won't change, they'll just become a better put-together illusion of thought.
So you might be able to use LLMs to help create quick and shitty proof of concepts, and even do some rubberducking, but not to think up solutions for you.
Celosia
2 ·