Non-photorealistic rendering projects

Layzej

Layzej wrote:
Here's a non-photorealistic rendering of a 3d model of the NPR guy:

Forgot to link to the project: 3D NPRGuy Scribble Render

gtoal

I'm so glad I left the NPR gallery open to anyone to post because there's been a lot of impressive work posted while I've been gone!

I dropped in to ask your help; I'm trying to remember a web site from some years ago that I used to use a lot but now either can't find or it is gone…

it was a huge long page of examples of graphics filters - maybe 4 to 6 per line. The sort of filters that produce effects like some of the things in the NPR gallery.

It was for a package like ImageMagick or GraphicsMagick or G'Mic, or Anti-Grain Geometry or maybe some Gimp or script-fu plugins.

I've found these and none of them are it:

http://www.imagemagick.org/script/examples.php
https://gmic.eu/gallery/
https://gmicol.greyc.fr/menu.php?folderId=0
https://web.archive.org/web/20090308091125/http://antigrain.com/demo/index.html
http://www.fmwconcepts.com/imagemagick/index.php

although they're similar in style. If anyone remembers or has a bookmark for a web page that fits this description could you post it please?

thanks

Graham

Retr0id

Sorry to necropost, but does anyone have the “full-res” (480x360) original version of the NPR guy as an image file?

Last edited by Retr0id (April 30, 2022 13:41:49)

awesome-llama

Retr0id wrote:
Sorry to necropost, but does anyone have the “full-res” (480x360) original version of the NPR guy as an image file?

http://www.gtoal.com/scratch/nprguy.jpg

--Eggs--

where did the image come from

badatprogrammingibe

Anyone else here try out Stable Diffusion? You can do a lot of cool NPR using the img2img feature, and the img2text is really great as well.

Last edited by badatprogrammingibe (Sept. 24, 2022 04:58:23)

gtoal

badatprogrammingibe wrote:
Anyone else here try out Stable Diffusion? You can do a lot of cool NPR using the img2img feature, and the img2text is really great as well.

Yes! Some truly exciting things going on in this area. I've been using dreamlike.art while its still free… unfortunately I don't have a computer at home good enough to run Stable Diffusion myself, though I suspect some of you kids with high powered graphics games machines might do!

In fact you've given me an idea - I've just now created a studio where scratchers can upload images to (wrapped in a project) to show off what you've created with these tools.
(It's NPR but I want to keep that studio for software that implements NPR, not use it for posting images created elsewhere)

The new studio is at:

https://scratch.mit.edu/studios/32241985/

What would be especially neat is if you can use the images you create in your projects, eg characters, backdrops, etc.

If any of the long time scratchers who I recognise from when I was a regular here would like to help moderate this new studio (i.e. remove inappropriate images should any turn up, so that the MIT folks don't close the studio down), post in the comments section to let me know: https://scratch.mit.edu/studios/32241985/comments - sorry newer users that I haven't met - since moderators can take down any image that they think is inappropriate I don't feel comfortable giving that privilege to anyone I don't know.

Graham

Last edited by gtoal (Oct. 15, 2022 22:45:29)

badatprogrammingibe

gtoal wrote:
badatprogrammingibe wrote:
Anyone else here try out Stable Diffusion? You can do a lot of cool NPR using the img2img feature, and the img2text is really great as well.
Yes! Some truly exciting things going on in this area. I've been using dreamlike.art while its still free… unfortunately I don't have a computer at home good enough to run it myself, though I suspect some of you kids with high powered graphics games machines might do!

Surprisingly, there's a way to set it up for CPU, and that's what I've been doing as well. It's orders of magnitudes slower than using a GPU: it takes me about 10 minutes or so to render a single image on my laptop with an ULV processor, as opposed to around 10 seconds on a GPU, but it's usable enough. I haven't tried upscaling or higher res though. It's probably just better/simpler to be using dreamlike.art or one of the other many options, though, at least while it's still free.
I can't link the CPU guides here as the one I used uses language that's not so appropriate for this website, but they shouldn't be too hard to find.

gtoal

badatprogrammingibe wrote:
Surprisingly, there's a way to set it up for CPU, and that's what I've been doing as well. It's orders of magnitudes slower than using a GPU: it takes me about 10 minutes or so to render a single image on my laptop with an ULV processor, as opposed to around 10 seconds on a GPU, but it's usable enough.

I did try that in mine but failed due to insufficient RAM :-/ I have a very underpowered computer by modern standards. It's been good enough for everything I wanted to do … until now!

G

_nix

gtoal wrote:
badatprogrammingibe wrote:
Surprisingly, there's a way to set it up for CPU, and that's what I've been doing as well. It's orders of magnitudes slower than using a GPU: it takes me about 10 minutes or so to render a single image on my laptop with an ULV processor, as opposed to around 10 seconds on a GPU, but it's usable enough.

I did try that in mine but failed due to insufficient RAM :-/ I have a very underpowered computer by modern standards. It's been good enough for everything I wanted to do … until now!

G

Maybe there's some nastily tricky way to convince your computer to let hard-disk space stand in for additional RAM

I haven't messed with much NPR recently, but I love to hear folk are still interested and exploring it more now and then!

Scratch-Minion

badatprogrammingibe wrote:
Anyone else here try out Stable Diffusion? You can do a lot of cool NPR using the img2img feature, and the img2text is really great as well.

I tried out a Stable Diffusion Demo version.
All I had to do was type in a line of text and it would produce 4 images in 15 seconds on my PC.
It didn't have any other features available.

I was most impressed by the Stable Diffusion Demo program.

Here are 3 of my projects with images generated by Stable Diffusion:

Crispydogs101

How do people code it in a way that it makes a specific image?

PullJosh

Crispydogs101 wrote:
How do people code it in a way that it makes a specific image?

If you look inside any of the NPR projects, you will see that they have a list inside that is massive and full of numbers. Each list item corresponds to one of the pixels on the screen, and the item's value stores the color of that pixel. So with some reasonably simple code, it's possible to draw the original image just by reading the pixel colors from that list.

From there, NPR projects usually do some additional math of some kind to manipulate the image before drawing it. It's really fun to play around with different modifications and see how they look!

Crispydogs101

PullJosh wrote:
Crispydogs101 wrote:
How do people code it in a way that it makes a specific image?
If you look inside any of the NPR projects, you will see that they have a list inside that is massive and full of numbers. Each list item corresponds to one of the pixels on the screen, and the item's value stores the color of that pixel. So with some reasonably simple code, it's possible to draw the original image just by reading the pixel colors from that list.

From there, NPR projects usually do some additional math of some kind to manipulate the image before drawing it. It's really fun to play around with different modifications and see how they look!

Kay. Thanks for the help.

just_notfan373

Layzej wrote:
Here's one that's a little more “non-photorealistic” than I'd like: https://scratch.mit.edu/projects/70172336/ The idea was to generate a picture from 20 triangles using a genetic algorithm. It didn't really work. It got to 40% and then stopped climbing. It had the broad outline of the reference picture, but looked more like a mess of triangles if you'd never seen the reference.

I'm running it again with only 10 triangles.. Maybe I'll add more triangles each time it stops climbing. Gtoal had an interesting post about using a mesh of triangles. Maybe that's the right approach?

W H O O P S O U R S E R V E R I S S C R A T G H I N G I T S H E A D

Layzej

Fixed

Games-n-Goodness

cool

Discuss Scratch