Carrara Non Photo Realistic Works

Dartanbeck

Hmmm... do you (HW) know if Topaz Simplify will work on sequences (animations)?

I agree about the whole tonal range thing. Have you seen DanRitchie's video on painting trees? He says the same things - artists generally buy 'these' colors, and that's it. So the pallet is automatically much more limited. It's also just cool to watch him paint ;)

I saw you give an example of ToonPro. I love the overdraw feature! 

Always looking for a new way to try and get animated renders to become more of a painting. Some photo-to-painterly effects are random, hence we can run the same image through many times, each time will come out different. I think that's somewhat why my curly (shader) dynamic hair is so glitchy - the shader-created curls are applied randomly to each frame, so as the frames pass, there's not continuity, and it turns out looking like crap.

 

Diomede

Thought this was an interesting video on the science of non-photo rendering. 

Headwax

Hey that's interesting with the tone mapping. 

For the Oloneo program I am just using it at the moment for fine adjustment at the end. Colour saturation and detail 'pop'. 

That said, I'm am rendering with lots of ambient etc, so I think doing a relight and using the HDR options would be going against what I am after ???

Headwax

Dartanbeck said:

Hmmm... do you (HW) know if Topaz Simplify will work on sequences (animations)?

I agree about the whole tonal range thing. Have you seen DanRitchie's video on painting trees? He says the same things - artists generally buy 'these' colors, and that's it. So the pallet is automatically much more limited. It's also just cool to watch him paint ;)

I saw you give an example of ToonPro. I love the overdraw feature! 

Always looking for a new way to try and get animated renders to become more of a painting. Some photo-to-painterly effects are random, hence we can run the same image through many times, each time will come out different. I think that's somewhat why my curly (shader) dynamic hair is so glitchy - the shader-created curls are applied randomly to each frame, so as the frames pass, there's not continuity, and it turns out looking like crap.

hi, well thanks for that vid - he's certainly a good painter !

for the topaz simplify you could do a batch process on your sequenced images, 

via 

https://support.topazlabs.com/hc/en-us/articles/200097758-How-to-create-an-action-in-Photoshop-using-Topaz-Labs-

The best results seem to be with 'simplify'

drop down the opacity of the layer till happy (you only what a bit of simplify)

then run it through a Topaz Impression filter

then adjust the opacity - or layer parameter 

- in a nut shell it's best to use both programs together - they have a demo I think

 

 

Dartanbeck said:

 

 

Headwax

Diomede said:

Thought this was an interesting video on the science of non-photo rendering. 

wow that's a lot of meat for thought, thanks for that link :)

 

Headwax

Out of interest here's examples of an orginal render and the result after a few million years of post work, using Topaz Simplify and Impression and Fotosketcher plus lots of different render passes. And Oloneo for the final tuning .

Best to use Simplify before Impression

The end result just depends how sketchy you want to go. I painted over the clothes to give a bit of variation to the drabness.

It's a quick process getting the render out of Carrara because the lighting isn't complicated etc - plus you dont worry about artifacts like bright pink handles...

 

 

 

 

Stezza

how cool is that ... you must share your paint over technique.. awesome 

Diomede

I know that the science video is too long for most people to watch (more than an hour).  But there are a few places that I think people might like because they discuss common features of styles and strokes.  I was really interested in the algorithm dicussion starting at the 24 minute mark and then for the next 10 minutes or so.  I now want to go back and take another look at Carrara's NPR tools and parameters.  For example, can the orientation choices of horizontal, vertical, U, V, Normal, Triangle and Hue allow for some control over cross hatching?  Maybe not.  But it sparked some ideas.  For example, could I render 2 NPR versions, one wth thick lines and one with thin - and then use a depth pass to have the final image use the thinner lines for features that are relatively far away?

 

EDIT - the thickness discussion doesn't start until a little later.

Stezza

Diomede said:

  For example, could I render 2 NPR versions, one wth thick lines and one with thin - and then use a depth pass to have the final image use the thinner lines for features that are relatively far away?

I do that often but composite in PSE and use my bamboo pen on the layers to make the adjustments.. dunno if that can be done within Carrara though, I haven't seen anything..

My Bamboo tablet is my best friend in PSE 

Headwax

 

Stezza said:

how cool is that ... you must share your paint over technique.. awesome 

thanks Stezza!!, it's more a case of action - reaction - so each 'action' you do on an image leads to a different response each time, - as opposed to a 'recipie'.

But  good thing to do is reduce the tonal range with a 'screen' paramter on a 'diffiuse'? layer pass - or one without shadows and plenty of ambient 

 

 

Headwax

Diomede said:

I know that the science video is too long for most people to watch (more than an hour).  But there are a few places that I think people might like because they discuss common features of styles and strokes.  I was really interested in the algorithm dicussion starting at the 24 minute mark and then for the next 10 minutes or so.  I now want to go back and take another look at Carrara's NPR tools and parameters.  For example, can the orientation choices of horizontal, vertical, U, V, Normal, Triangle and Hue allow for some control over cross hatching?  Maybe not.  But it sparked some ideas.  For example, could I render 2 NPR versions, one wth thick lines and one with thin - and then use a depth pass to have the final image use the thinner lines for features that are relatively far away?

 

EDIT - the thickness discussion doesn't start until a little later.

thanks for the heads up on the part to watch :)

thinking aloud here -  traditional eg ink and pen renders (by hand) the  render lines  (often)  follow the contours - I think that there is a clue to be had in a UV pass - but not sure how you would automate it tho.

This image isnt mine 

 

Eg on this etching of a Horse by Lionel Lindsay (Australian)  

 

 

cdordoni

head wax said:

That said, I'm am rendering with lots of ambient etc, so I think doing a relight and using the HDR options would be going against what I am after ???

I was curious what other tools you had used in the Oloneo application. If you are using a lot of ambient lighting, seems line tone mapping or the HDR would be redundant.

Headwax

Yes, that's right. But you can render out a shadow pass as well to use in post :) which is handy because it gives you more control and you get a bit more tonal range back where you want it.

OT: Well I just had the quic kest thre ad del etion ever regard ing a Car rara plu gin.  Da must have certain words their ro bots  se arch for. I'm impressed - it la sted ab out 7 5 seconds.

Stezza

1 9 8 4 with a ro ck et 

Headwax

It calls for a wins ton smith re n de r :)

Stezza

Win Ston

PhilW

Both Luxrender and Octane have tone mapping which converts the internal floating point values to the 256 levels that we tend to use in most common image formats. Both allow you to do this interactively, and come with a range of film type settings, so you can adjust the intensity, brightness, contrast, saturation, gamma, colour tones etc. interactively - so you should never end up with an under or over exposed image, without having to rerender.

Diomede

I went to another forum to discuss some news for that plugin that shall not be named.  

.

 

head wax said:

OT: Well I just had the quic kest thre ad del etion ever regard ing a Car rara plu gin.  Da must have certain words their ro bots  se arch for. I'm impressed - it la sted ab out 7 5 seconds.

 

Vyusur

Phil, is there any way to improve the quality of a final render in octane, to decrease grain and noise, to change sampling and where? Thank you in advance.

cdordoni

PhilW said:

Both Luxrender and Octane have tone mapping which converts the internal floating point values to the 256 levels that we tend to use in most common image formats. Both allow you to do this interactively, and come with a range of film type settings, so you can adjust the intensity, brightness, contrast, saturation, gamma, colour tones etc. interactively - so you should never end up with an under or over exposed image, without having to rerender.

Thanks for posting that, have to check out LuxusCore/Luxrender and look for those settings.

PhilW

Vyusur said:

Phil, is there any way to improve the quality of a final render in octane, to decrease grain and noise, to change sampling and where? Thank you in advance.

Have you got the latest update - it adds Adaptive Sampling, which speeds up getting noiseless renders. You can turn it on in the Edit / Render Target Settings and set a threshold - when the noise in a certain area meets the threshold value, it stops rendering that area, focusing the rendering on other areas (I think iRay does this by default), so if you have an image with a difficult area, it focuses the processing on that. This means that you can set a higher Samples per Pixel target without getting silly render times.

Lighting also plays an important part - ensure that you have enough lighting in your scene. That doesn't necessarily mean having lots of light sources, but watch out for areas in an image that will only be lit by reflected (bounced) light, particularly if the light has to bounce more than once to get to it. Most of my images have a pretty simple lighting set-up - an HDRI for environmental lighting and then maybe a key light and a rim light or opposing light (on the opposite side to the key light). There are exceptions, but the fewer lights you have, in general the easier it is to control to get the effect that you want. Every light that you add should be there for a specific reason.

I hope this helps.

UnifiedBrain

Diomede said:

I know that the science video is too long for most people to watch (more than an hour).  But there are a few places that I think people might like because they discuss common features of styles and strokes.  I was really interested in the algorithm dicussion starting at the 24 minute mark and then for the next 10 minutes or so.  I now want to go back and take another look at Carrara's NPR tools and parameters.  For example, can the orientation choices of horizontal, vertical, U, V, Normal, Triangle and Hue allow for some control over cross hatching?  Maybe not.  But it sparked some ideas.  For example, could I render 2 NPR versions, one wth thick lines and one with thin - and then use a depth pass to have the final image use the thinner lines for features that are relatively far away?

Thanks for posting.  I watched it, as I am currently interested in the same thing.  NPR does some pretty interesting crosshatching.  You can definitely change the direction and look of it of it with horizontal, vertical, U, V, etc.  My issue is more about how objects are treated in NPR.  In a single scene, you can have some great crosshatching on some items, and something different on others, with seemingly no rhyme or reason for the difference.  There is apparently something in certain objects that the NPR engine reacts to.  If I could figure it out, it would be a lot easier.

 

 

head wax said:

thanks for the heads up on the part to watch :)

thinking aloud here -  traditional eg ink and pen renders (by hand) the  render lines  (often)  follow the contours - I think that there is a clue to be had in a UV pass - but not sure how you would automate it tho.

This image isnt mine 

 

Eg on this etching of a Horse by Lionel Lindsay (Australian)  

 

 

It does appear that the Carrara NPR engine follows contours as well.  Here is a partial render where it mostly worked as expected:

 

 

But when I add the Dinokandas, the crosshatching gets very weak or disappears entirely.  I can make them light or dark, but the lack of crosshatch shading is the issue.  The Uncle Humerus character is also a challenge.  Still trying to figure out at least a workaround without having to resort to postwork.

 

Diomede

Thanks for bringing this up, UB.  The following is for people who are not familiar with your point - that there does not seem to be enough information to know how an NPR of two seemingly similar objects will turn out.  There is a function, so we ought to be able to identify it with trial an error, if nothng else.  But I haven't been able to thus far, although I've picked up a few patterns.  Definitely would like a discussion of the NPR functions by someone familar with development, but that isn't going to happen.   

 

I do know that NPR is affected by object properties.  One clear example is mesh density. Therefore, two seemingly similar objects can render differently.  Here is a quick example.  I have two cubes of identical proportions and identical shaders. The cube on the left has the minimum number of polygons (6).  The cube on the right has been subdivided several times to increase mesh density.  It has 1536 polygons.  They both have identical uvmaps (simple box face).  The shader for both is the same checkered mixer.  I have put them both in the same scene and used the default PR and NPR settings. A default photoreal render will look identical (or very, very, very similar).  Not so for NPR.   As you can see, the default NPR renders the low density cube with much less definition than the high density cube, even though they are the same shape, the same dimensions, and the same shader.

 

Cubes photoreal default render

 

Cubes non photoreal default render

 

Headwax

Stezza said:

Win Ston

 

classic S tezza  :) John hurt will be impressed.

 

 

Headwax

Diomede said:

Thanks for bringing this up, UB.  The following is for people who are not familiar with your point - that there does not seem to be enough information to know how an NPR of two seemingly similar objects will turn out.  There is a function, so we ought to be able to identify it with trial an error, if nothng else.  But I haven't been able to thus far, although I've picked up a few patterns.  Definitely would like a discussion of the NPR functions by someone familar with development, but that isn't going to happen.   

 

I do know that NPR is affected by object properties.  One clear example is mesh density. Therefore, two seemingly similar objects can render differently.  Here is a quick example.  I have two cubes of identical proportions and identical shaders. The cube on the left has the minimum number of polygons (6).  The cube on the right has been subdivided several times to increase mesh density.  It has 1536 polygons.  They both have identical uvmaps (simple box face).  The shader for both is the same checkered mixer.  I have put them both in the same scene and used the default PR and NPR settings. A default photoreal render will look identical (or very, very, very similar).  Not so for NPR.   As you can see, the default NPR renders the low density cube with much less definition than the high density cube, even though they are the same shape, the same dimensions, and the same shader.

 

Cubes photoreal default render

 

 

Cubes non photoreal default render

 

now we are getting somwhere - impressive work

 

Headwax

UB wrote

But when I add the Dinokandas, the crosshatching gets very weak or disappears entirely.  I can make them light or dark, but the lack of crosshatch shading is the issue.  The Uncle Humerus character is also a challenge.  Still trying to figure out at least a workaround without having to resort to postwork.

that big gorilla thing in the back looks had etched - thanks for pointing that out  

Headwax

Diomede said:

I went to another forum to discuss some news for that plugin that shall not be named.  

.

 

head wax said:

OT: Well I just had the quic kest thre ad del etion ever regard ing a Car rara plu gin.  Da must have certain words their ro bots  se arch for. I'm impressed - it la sted ab out 7 5 seconds.

 

thanks Diomede

 

Headwax

 

 

Continuing with a random 'Pirate a Day". Initially I was going to name this chap "Blunderbuss Pete". But being a lazy fellow I looked for a random web based Pirate Name Generator. So this be:

Owen "Gull Basher" Carabis .... AKA "The Poison Claw of Palm Atoll!" - who never cut his fingernails, not ever, not even to go to church.... and whose Blunderbuss (named Ralph) was always going off at embarrassing moments.

Carrara render passes post worked  with Topaz Simplify, Topaz Impressions, Rons Brushes, Olone for Fine tuning, Fotosketcher

I used a shadow pass on a plane beneath the pirate and cannon balls and then did another render of the ship and canon and composited them all together

wgdjohn

Diomede... thanks... that is a huge difference.

Vyusur

PhilW said:

Vyusur said:

Phil, is there any way to improve the quality of a final render in octane, to decrease grain and noise, to change sampling and where? Thank you in advance.

Have you got the latest update - it adds Adaptive Sampling, which speeds up getting noiseless renders. You can turn it on in the Edit / Render Target Settings and set a threshold - when the noise in a certain area meets the threshold value, it stops rendering that area, focusing the rendering on other areas (I think iRay does this by default), so if you have an image with a difficult area, it focuses the processing on that. This means that you can set a higher Samples per Pixel target without getting silly render times.

Lighting also plays an important part - ensure that you have enough lighting in your scene. That doesn't necessarily mean having lots of light sources, but watch out for areas in an image that will only be lit by reflected (bounced) light, particularly if the light has to bounce more than once to get to it. Most of my images have a pretty simple lighting set-up - an HDRI for environmental lighting and then maybe a key light and a rim light or opposing light (on the opposite side to the key light). There are exceptions, but the fewer lights you have, in general the easier it is to control to get the effect that you want. Every light that you add should be there for a specific reason.

I hope this helps.

 

Phil, thank you for the response and detailed explanation! My octane version is 3.20.20, I guess it's fresh. I already know where is «target» menu, but I couldn't find any «threshold» parameter. I only increased the Max.Samples parameter, but I don't know if it has effect on anything.