Thank you so much man! tweaking it instantly made my eevee glass so much better. been looking at a lot of resources trying to get the glass to work. this one just made a world of difference.
My world changed thanks to Blender! Regardless of if i get clients or not, which can make my journey a bit more fulfilling, I enjoy the adventure I am on using it. I want to thank the entire community, the endless tutorials, the creators behind it and even the drag and droppers with millions of likes on tiktoks or instagram; thanks for sailing our worldwide ship xD. Don't mind my crazy. haha
The way AI is used here is exactly how it should be, to help artists improve and achieve better results, not to replace them entirely, which is what we fear.
exactly what i've been saying. ai is the concept and idea generator, it is the thing that gives loose suggestions, it's not the final producer. the biggest problem with ai is that people do not use it properly. these guys use it properly, these guys are fucking cool. also younesSS i like your profile picture, very cool.
it's inevitably moving in the direction you fear , i fear it too .... and it will get there in a year or two , we should accept that fact and adjust our careers accordingly before it's too late ...
@@JC-bm3zo that's just a stupid statement made by someone who listens to tech bros who have no idea what they're talking about. You quite literally need artists for the AI to even work. Without us real artists, you couldn't feed the algorithm database with real art which is then being stolen. What you meant so say was "We're right now in a wild west time of algorithms because it's so new that copyright laws and laws in general haven't been able to keep up. Wait a few years and we will get a few new laws about copyright because the work of millions of artists is quite literally being stolen without them being paid a single penny. It's like the domain bubble in the 90s and early 2000s. It will be over and a new bubble will appear." So, if you have no clue about something... just don't say anything. Oh and btw. don't listen to tech bros. They where always without a single exception wrong.
Even tho there are things around cycles that are still missing like deep rendering, MIP mapping etc., it's still a fantastic renderer compared to Arnold, Vray, Redshift etc. and probably the fastest too from my experience. A lot of people who are new to 3D use Blender and lack experience working with other 3D DCCs and renderers, so when they don't see movie quality renders coming straight out of Blender, they think it's because of Cycles's being not good enough for producing photoreal renders like other commercial render engines, when they lack good lighting, composition, proper color management, compositing and professional color grading etc. I mean, Cycles was made by Brecht Van Lommel, who was one of the lead devs of Arnold, I think he knows what he's doing.
Cycles is considered bad because it's free and it's used by begginers so if you compare average Cycles render to average Arnold, Octane etc. Cycles will look bad, because those renderers are expensive and used mostly by professionals.
As a whole, Cycles is very good. But quality wise, some other renderers have a slight edge. They need less tweaking and usually retain more details. The last example is great for illustrating this: corona or octane with the same textures would have much nicer details on the floor for example. The one they ended up using looks very video-gamey. And the light distribution is often is a bit nicer, which is hard to quantise, but if you tried them you would know that in cycles you need to be more specific with the light angles, sizes, strengths, while with other renderers often moving the sliders once is good enough already....all that being said, I still prefer cycles, because it's almost as good, faster (wihtout optimization), much more stable and very flexible.
@@cyrkielnetwork it used to be bad before the Cycles X and the recent rebump features. it was slow to render compare to those render engine you mentioned. but now a days it's quite near to octane with still about 30-40% missing features from those paid guys that is used for optimization to render big scenes. I always use Cycles for small scenes but since the optimization features to render big scenes is still kinda not there, I still stick on my Octane license, I think Mipmapping is the biggest one that everyone was been requesting for larger texture scenes to save some vram.
I'm not too much into compositing, what are the pros of deep rendering comparing to position pass which can be used in comp to interpret sequence as a point cloud in 3d space?
@@DimiArt you can render different collections separately , for the mouse for example you just turn on indirect only on the outliner, or remove it completely, you render it, then render the mouse on its own over a transparent background, and you just combine both in compositing.
really though! i know exactly what he means by lights "fighting over" a pixel, because of how samples are calculated- but i had never considered just... rendering them separately so they wouldn't, and then using a simple image blending operation between layers to just stack them.
I use Blender for all my projects, and it’s amazing to see that Cycles can achieve such incredible render quality 2:40 . I don’t have an RTX 4090, but the speed of Cycles is continually improving. I also love using the denoise feature, as it saves me a huge amount of time on my small personal projects. I hope we live to see Cycles rendering in real-time!
I'd say the use of AI as a REFERENCE generator is 100% ethical. that's like looking at other people's renders, getting inspired and making it yourself (which is literally what you're doing)
It's not because you're still using unauthorised copied and processed data from millions of people to generate the references. It's not really the same like "processing a picture in your brain" and sketch, draw or paint a mood or an idea based on the inspiration. But "mood boards" with unauthorised images from physical sources and later from the Internet are a common part of the creative process. But we need to be aware that all of this is copyright infringement when used for commercial work.
@@meem-zd3kj I know exactly how generative AI based on machine learning models works by pre-processing millions of pictures and movies without asking the copyright holders.
that ai trick is crazy - works so well! and I feel like is exactly how ai should be used - not to make the end result - but as inspiration / feedback - awesome!
cycles is fine. like 9/10 no problems in production. but then there comes a job where you want a mist pass from a scene with trees, or render volumes indirect to have their shadows as a seperate layer, enable motion blur for generated instances or want real caustics.
Had the same issues with mist pass and trees. Only solution was to convert alphas for foliage textures to B/W only, as grayscale alpha textures mess with the mist pass.
That hint about render layers was so useful. I've heard people say that, but in my experience having many layers just slowed everything down. The trick seems to be splitting by light groups - I'd never think of that, but it makes perfect sense. Thanks!
incredible lecture, so much to learn. I've been following the studio for some time now and was delighted that Alex of all people gave a talk about it and also shared so many insights and the whole thing is accessible to us. Unbelievable.
instead of magnific we can also use free and open source ai models like flux with some controlnet in something like comfyui for free which will give more control and better results but it requires good gpu and some technical knowledge. and for editing, compositing , colorgrading, sound design davinchi is goat. and for 3d stuff you know allready
Cycles is awesome. The quality in relation to rendertime is amazing. I was surprised they dont use denoising, but do it in comp. I wonder what they use ?
One of the big Problem in CG is, that we get forced that beauty is perfect by the clients. i fought many battles and always clients ruin realism out of fear making things not beautiful enough. However the same thing counts for design, yet you can divide good designers from bad, by check how much someone dares.
Cycles kicks ass. Gets better with every update. I would like see an easier/more automated temporal denoising. Yes, it could be done in compositing, but for me, the less I need to do in compositing, the more time I save.
It's amazing to see how You approached the render in separate render passes like all "big boys" do on big budget films... !!!! Marvel - want spent less - use BLENDER !!! :D
the work would never get done in blender. it can't handle that size. This is why Houdini, Maya, Nuke, Katana and all other professional software OTHER than Blender is used. Fact. Keep being a hobbyist.
This guy actualy told some important things about how to use various tools for artists in blender , but iam not sure wveryone underestood that efficiently
you just shared my secret sauce lol, no but honestly thanks a lot for sharing this I also believe that Cycles is way better than what people think and I love octane also that workflow is terrific specially with USDs now you can import particles or geometry with attributes from Houdini and you are golden I am still not comfortable with karma as much as I am with cycles so it does make a difference I also think cycles should implement a few more things to make our lives easier such as deep rendering, more AOVs or a revisited motion blur system but now with all the progress it made I think those are negligible
0:55 Compare the amount of "getting started" and "for beginners" training free or paid, to real High Level/Deep Dive/OMFG You Nerd level training available for Blender. When the most lucrative work to be had using Blender is...teaching people to use Blender it's no wonder this perception is the norm. Everyone being blown away by _lighting passes_ and basic compositing is telling. You my friend are one of a handfull of Exceptions Proving the Rule. Props for the "correct" AI use, nice to see people Getting It.
Instead of talking about stone AI program that isn't Blender (there's other platforms for that) I would really have liked to see way more indepth about those light layers, how they're setup and clear comparisons in noise and render times compared to a single layer render. And also a bit more on your comp denoising, is that Blender, Davinci or something else? :)
I'm very conflicted on the use of AI, but that tip definitely seems like one of the better uses of it. But personally I'll probably still avoid it out of principle, and also because I'd much rather learn from the more deliberate details of a better artist's work than whatever a machine learning algorithm can hallucinate. Still though, it's a decent idea.
The big benefit is that it generates results in the context of your scene, and you can't always find real references that work for what you're doing, plus even if you can, the difference in the time taken becomes the big benefit I absolutely think this type of referential usecase will be the norm in the future, even clients can use it to visualise what they're looking for, for the modellers they hire, and this applies to a lot of other content creation jobs too. Making these algorithms the "middle man" is a great direction to head in, instead of trying to make end results through them. We can make work faster and more straightforward without necessarily sacrificing job positions this way
Unironically by using AI you likely are learning the more deliberate details from a better artists because those details and artists works were used in the training of the AI. Honestly with something like magnific in particular you’re no longer just generating from nothing. You need to input your own image as the prompt and then it works on that. It’s technically (or was) an image upscaler. But they added some hallucinations into the algorithm and gave people a slider for how much to let it hallucinate. We looked at it a while ago to upscale some of our old texture sets we lost the higher res versions of and it worked very well as just an upscaler. Better than topaz would.
Ya idk it’s definitely a shortcut. These are all pretty basic details you learn in environment concept art or environment art. I think everyone needs to be a multidisciplinary artist to succeed, but if these new artists aren’t interested in learning about architecture, how wood gets polished with time and wax, etc then idk if their work Will be interesting with it without ai
From what I know, Cycles is lacking just a few features that would make it professional grade. - Variable Compression Rates for render layers (especially cryptomattes) - Render Layer Denoising (although to be fair, you could argue professional workflows would brute force high sample counts without needing denoising on their farms) - DeepEXR support - a more robust ACES integration on render layers, when used in the compositor I am not a programmer and don’t know how difficult these are but it feels like a comparatively small mountain to climb to make cycles truly great
maybe he got lucky with the glass shader there. But if you try to make a thin plastic sheet beacuse something is wrapped in it, you are loosing so much light, it's insane. So from my knowledge, the glass shader is still broken.
Very interesting talk. I really thought that nothing shot was from some other e fine. Remember seeing that ad. I do think the title of the speech is perhaps a bit misleading. I was kinda baffled about the teaser, magnificent work! Also very interesting method for the lights. Yet I wonder how they comp that. Because most times you use the passes. Overlaying so much details, parts start blowing out. I mean in some of the layers, we also see details which are also in other shits
Basically, using AI like this is like having personal teacher who makes improvements to your work so you can see where your messed up and what to do to make it better. More or less.
Cycles doesn't have path tracing options that would make lighting more natural and it must be intentional, because if you were serious about developing your 3D program this would be a main feature.
I love the workflow with Ai. However, what I am wondering is, how long will the newly gained information (tricks) stick with intermediate artists. Is it like a test at highschool that after a quick rehearsal you forget about? How much will we rely on fresh input from AI in the future? Just a thought
I love Blender, but we have to be honest. Houdini is the best of the best of when it comes to simulations. Blender simulations are a bit of a pain and nowhere near as fast as Houdini.
even as someone very critical of the ethics of the ai marketing blitz I don't see a problem in using the output as reference images; artists reference completed works all the time. if anything the bad part is giving a company money for stealing content from smaller creators.
Only other issue to me is that there’s not much reason to be using ai as reference when there’s so much more value in somebody just learning how to properly use real reference
@@warcat3d I think the main issue he was talking in is that you might already be using reference but stuff still feels like it can be improved so AI is meant to help you visualize tons of different ideas that's based on your artwork. Sometimes you cant realize what your missing based off a different image. Like with the chair in this video i dont think a reference could have made you think of going through the work of adding a chair in the corner. Its more of you start off with real reference images and then when you feel stuck you can use AI to give you ideas on what you have so far.
Cycles is great for lot of things, organic objects like realistic skin and hair not so much - overall Arnold still wins. But cycles is getting better all the time.
I primarily use 3ds max + vray. Correct me of im wrong. Isnt the method he mentioned seperating different light groups and rendering different images, same as just adding render elements in vray(or most other engines). And adding them later to get the beauty pass? Or we can just add back to beauty element and it automatically seperates emmissive mats, light groups, reflections, refractions, environment and so on.
I think render elements are just render passes in Vray if I'm not wrong, where here he's using light groups to put certain lights in groups (which is a relatively new feature in Blender) to be rendered them separately and put them back together in comp.
@@xanzuls yes, we add different render elements to render corresponding passes along with beauty render. And there is one render element named "back to beauty" which basically creates all the passes automatically. For light grouping, there is something called Lightmix. I use it to render day and night scenes simultaneously. You can turn off/On any lights,emmissive matetials, change intensity,color after the render is complete and save variations. And whichever lights you turn off will also affect the underlying passes. So you can have all the passes for each light combination switched ON/OFF.
I was just surprised that rendering different passes was a new thing..I mean it's a method used from decade. . I thought blender already had something similar to render elements to automatically render the passes.
@@punithaiu render passes are not a new thing, but that's not what he was talking about when he was talking about rendering out lights separately, as I think he used light groups in blender and light groups are a similar feature to lightmix in vray. Blender has render passes, and you can choose to render them under the 'view layer' panel.
Nobody using Cycles in workflow production since 1994, you ever worked in studios or it was just in your bedroom ?!?Arnold, V-Ray, Renderman, you know theses engines ? No comment...
![BOWKYLION - วิงวอน (ex-change) [Official MV]](http://i.ytimg.com/vi/cLdnZWDaO-w/mqdefault.jpg)
For those wondering, the thickness output (as of 4.2) is only used for Eevee to fake refraction. It's not used by Cycles.
It was most likely added by the modeler so they can use eevee to preview what they're modeling as they're modeling it.
@@Starlit-Music oh absolutely, it was purely for their previewing the general look as they modeled.
Also used for subsurface scattering (EDIT: and translucent shader) in Eevee
Thank you so much man! tweaking it instantly made my eevee glass so much better. been looking at a lot of resources trying to get the glass to work. this one just made a world of difference.
The fact it free already enough for me, but when it getting better every interation is outstanding dedication.
My world changed thanks to Blender! Regardless of if i get clients or not, which can make my journey a bit more fulfilling, I enjoy the adventure I am on using it. I want to thank the entire community, the endless tutorials, the creators behind it and even the drag and droppers with millions of likes on tiktoks or instagram; thanks for sailing our worldwide ship xD. Don't mind my crazy. haha
I picked up Blender in 2008 and that has always been the case. I remember Blender before Material Nodes. It was awful.
Thanks to Blender Foundation, they can be free and have great development.
The way AI is used here is exactly how it should be, to help artists improve and achieve better results, not to replace them entirely, which is what we fear.
goated statement
exactly what i've been saying.
ai is the concept and idea generator, it is the thing that gives loose suggestions, it's not the final producer. the biggest problem with ai is that people do not use it properly.
these guys use it properly, these guys are fucking cool. also younesSS i like your profile picture, very cool.
Yup, that's the best use of AI, inspiration and guiding, not doing work for you
it's inevitably moving in the direction you fear , i fear it too .... and it will get there in a year or two , we should accept that fact and adjust our careers accordingly before it's too late ...
@@JC-bm3zo that's just a stupid statement made by someone who listens to tech bros who have no idea what they're talking about. You quite literally need artists for the AI to even work. Without us real artists, you couldn't feed the algorithm database with real art which is then being stolen. What you meant so say was "We're right now in a wild west time of algorithms because it's so new that copyright laws and laws in general haven't been able to keep up. Wait a few years and we will get a few new laws about copyright because the work of millions of artists is quite literally being stolen without them being paid a single penny. It's like the domain bubble in the 90s and early 2000s. It will be over and a new bubble will appear." So, if you have no clue about something... just don't say anything. Oh and btw. don't listen to tech bros. They where always without a single exception wrong.
Even tho there are things around cycles that are still missing like deep rendering, MIP mapping etc., it's still a fantastic renderer compared to Arnold, Vray, Redshift etc. and probably the fastest too from my experience. A lot of people who are new to 3D use Blender and lack experience working with other 3D DCCs and renderers, so when they don't see movie quality renders coming straight out of Blender, they think it's because of Cycles's being not good enough for producing photoreal renders like other commercial render engines, when they lack good lighting, composition, proper color management, compositing and professional color grading etc. I mean, Cycles was made by Brecht Van Lommel, who was one of the lead devs of Arnold, I think he knows what he's doing.
Cycles is considered bad because it's free and it's used by begginers so if you compare average Cycles render to average Arnold, Octane etc. Cycles will look bad, because those renderers are expensive and used mostly by professionals.
Up vote for this 🔥
As a whole, Cycles is very good. But quality wise, some other renderers have a slight edge. They need less tweaking and usually retain more details. The last example is great for illustrating this: corona or octane with the same textures would have much nicer details on the floor for example. The one they ended up using looks very video-gamey. And the light distribution is often is a bit nicer, which is hard to quantise, but if you tried them you would know that in cycles you need to be more specific with the light angles, sizes, strengths, while with other renderers often moving the sliders once is good enough already....all that being said, I still prefer cycles, because it's almost as good, faster (wihtout optimization), much more stable and very flexible.
@@cyrkielnetwork it used to be bad before the Cycles X and the recent rebump features. it was slow to render compare to those render engine you mentioned. but now a days it's quite near to octane with still about 30-40% missing features from those paid guys that is used for optimization to render big scenes. I always use Cycles for small scenes but since the optimization features to render big scenes is still kinda not there, I still stick on my Octane license, I think Mipmapping is the biggest one that everyone was been requesting for larger texture scenes to save some vram.
I'm not too much into compositing, what are the pros of deep rendering comparing to position pass which can be used in comp to interpret sequence as a point cloud in 3d space?
the separate renders is also so clutch, never thought of that
I still don't understand how it's done
@@DimiArt you can render different collections separately , for the mouse for example you just turn on indirect only on the outliner, or remove it completely, you render it, then render the mouse on its own over a transparent background, and you just combine both in compositing.
really though! i know exactly what he means by lights "fighting over" a pixel, because of how samples are calculated- but i had never considered just... rendering them separately so they wouldn't, and then using a simple image blending operation between layers to just stack them.
@@DimiArt I forgot to mention you do this with render layers. that way everything is automated you just have to press render button once.
@@dzibanart8521 Can you make a tutorial?
Hell yeah ,another round of very high quality tutorials :)
Cycles isn't better than I think, because I think Cycles is the best.
octane.. redshift lmao
If you think it's the best it's because you're ignorant.
@@oBCHANo fr
@@oBCHANo I personally think cycles looks like dog water
@@oBCHANo dang dont be rube some people think its the best i think its really good for a render engine and its free
I use Blender for all my projects, and it’s amazing to see that Cycles can achieve such incredible render quality 2:40 . I don’t have an RTX 4090, but the speed of Cycles is continually improving. I also love using the denoise feature, as it saves me a huge amount of time on my small personal projects. I hope we live to see Cycles rendering in real-time!
Same here due
I'd say the use of AI as a REFERENCE generator is 100% ethical.
that's like looking at other people's renders, getting inspired and making it yourself (which is literally what you're doing)
It's not because you're still using unauthorised copied and processed data from millions of people to generate the references. It's not really the same like "processing a picture in your brain" and sketch, draw or paint a mood or an idea based on the inspiration. But "mood boards" with unauthorised images from physical sources and later from the Internet are a common part of the creative process. But we need to be aware that all of this is copyright infringement when used for commercial work.
@@3dvfxprofessor Tell me you have no idea how ai works without telling me
@@meem-zd3kj I know exactly how generative AI based on machine learning models works by pre-processing millions of pictures and movies without asking the copyright holders.
@@3dvfxprofessor more than two years with this take...
no use arguing about fair use being on the end user / transformative work
5:02 The most complicated node ever created!!!!!
damn, you guys are using my texture pack for lighting :D I hope it's been useful! Really good talk by the way!
Absolutely love it
Very interesting to listen, great job guys! Thank you.
Fantastic presentation, just Fan Tas Tic !!!
Cocacolastic man🤝😁
Absolutely mindblowing, very value able tips
I was using the magnific trick at the last company I was at. It was a powerful workflow.
Thanks to share your experience with us!
This is exactly where I've thought A.I. would excel in. Using it for brainstorming is brilliant.
This is the best talk to come out of the conference so far. Many practical techniques
really interesting workflow. thanks for sharing with us !
that ai trick is crazy - works so well! and I feel like is exactly how ai should be used - not to make the end result - but as inspiration / feedback - awesome!
Yeah I love using ai as an ideas tool
@@coltynstone-lamontagne Me too, I like to quickly see my rough ideas with AI, because sometimes ideas is just shiny object and not worth creating.
cycles is fine. like 9/10 no problems in production. but then there comes a job where you want a mist pass from a scene with trees, or render volumes indirect to have their shadows as a seperate layer, enable motion blur for generated instances or want real caustics.
Had the same issues with mist pass and trees. Only solution was to convert alphas for foliage textures to B/W only, as grayscale alpha textures mess with the mist pass.
@@ra-ul1238 yes, a simple math round node helps. Or rendering the mist pass in a separate scene with eevee.
That hint about render layers was so useful. I've heard people say that, but in my experience having many layers just slowed everything down. The trick seems to be splitting by light groups - I'd never think of that, but it makes perfect sense. Thanks!
Render layers (AOVs) are handled through View Layers, at the top right !
incredible lecture, so much to learn. I've been following the studio for some time now and was delighted that Alex of all people gave a talk about it and also shared so many insights and the whole thing is accessible to us. Unbelievable.
THANKS, for sharing!!
Great presentation, awesome work! 🤩
This was really educational and inspiring. Thanks Alex and kudos to the whole Blender team!
I love blender videos because they are mostly entertaining, but you can actually learn useful stuff in the process.
instead of magnific we can also use free and open source ai models like flux with some controlnet in something like comfyui for free which will give more control and better results but it requires good gpu and some technical knowledge. and for editing, compositing , colorgrading, sound design davinchi is goat. and for 3d stuff you know allready
Recognized his works, nice
Cycles is awesome. The quality in relation to rendertime is amazing. I was surprised they dont use denoising, but do it in comp. I wonder what they use ?
since they mentioned aftereffects, im guessing they use magic bullet looks denoiser
@@memeycorny7773 have to look into that
One of the big Problem in CG is, that we get forced that beauty is perfect by the clients.
i fought many battles and always clients ruin realism out of fear making things not beautiful enough.
However the same thing counts for design, yet you can divide good designers from bad, by check how much someone dares.
Cycles kicks ass. Gets better with every update. I would like see an easier/more automated temporal denoising. Yes, it could be done in compositing, but for me, the less I need to do in compositing, the more time I save.
As a long time cycles enjoyer, it is indeed superior.
Blessed is the person who is too busy to worry in the daytime, and too sleepy to worry at night.
Amazing, it's free but come with superpower.
This is truely amazing
Very good presentation!
THANK YOU MAN; BLENDER FOREVER
I can't find my jaw, impressive demo.
such a great presentation, really love how he articulated how once we get to the end of a piece we can compliment it with other tools / add-ons etc
It's amazing to see how You approached the render in separate render passes like all "big boys" do on big budget films... !!!! Marvel - want spent less - use BLENDER !!! :D
the work would never get done in blender. it can't handle that size. This is why Houdini, Maya, Nuke, Katana and all other professional software OTHER than Blender is used. Fact. Keep being a hobbyist.
Great insight, thanks!
7:13 this is very good insight, makes a lot of sense, well done
I use blender to modeling, great 3d Tool!
great educational video for a beginner like me
If they ever let a piece of software win the Nobel prize I nominate Blender
for what?
This guy actualy told some important things about how to use various tools for artists in blender , but iam not sure wveryone underestood that efficiently
they're too busy ignorantly jerking off to another blender presentation.
Great talk!
you just shared my secret sauce lol, no but honestly thanks a lot for sharing this I also believe that Cycles is way better than what people think and I love octane
also that workflow is terrific specially with USDs now you can import particles or geometry with attributes from Houdini and you are golden
I am still not comfortable with karma as much as I am with cycles so it does make a difference
I also think cycles should implement a few more things to make our lives easier such as deep rendering, more AOVs or a revisited motion blur system but now with all the progress it made I think those are negligible
It took him a while to realize that everything he decided not to change, he was actually choosing.
I love blender ! Cycles is amazing
Awesome talk
Great session
0:55 Compare the amount of "getting started" and "for beginners" training free or paid, to real High Level/Deep Dive/OMFG You Nerd level training available for Blender. When the most lucrative work to be had using Blender is...teaching people to use Blender it's no wonder this perception is the norm. Everyone being blown away by _lighting passes_ and basic compositing is telling.
You my friend are one of a handfull of Exceptions Proving the Rule. Props for the "correct" AI use, nice to see people Getting It.
I just LOVE Blender!!
Blender has a certain look that just can’t be shaken off
This is Fantastic!
amazing talk
Instead of talking about stone AI program that isn't Blender (there's other platforms for that) I would really have liked to see way more indepth about those light layers, how they're setup and clear comparisons in noise and render times compared to a single layer render. And also a bit more on your comp denoising, is that Blender, Davinci or something else? :)
they use After Effects for compositing according to the chart at 1:48
That's exactly why the say reference is king
Amazing job!
Thanks !!
The way how AI is used is how I teach it to my design students as an ideation tool - but slowly it starts to also replace still image rendering.😊
I like this
I'm very conflicted on the use of AI, but that tip definitely seems like one of the better uses of it. But personally I'll probably still avoid it out of principle, and also because I'd much rather learn from the more deliberate details of a better artist's work than whatever a machine learning algorithm can hallucinate. Still though, it's a decent idea.
The big benefit is that it generates results in the context of your scene, and you can't always find real references that work for what you're doing, plus even if you can, the difference in the time taken becomes the big benefit
I absolutely think this type of referential usecase will be the norm in the future, even clients can use it to visualise what they're looking for, for the modellers they hire, and this applies to a lot of other content creation jobs too. Making these algorithms the "middle man" is a great direction to head in, instead of trying to make end results through them. We can make work faster and more straightforward without necessarily sacrificing job positions this way
Unironically by using AI you likely are learning the more deliberate details from a better artists because those details and artists works were used in the training of the AI.
Honestly with something like magnific in particular you’re no longer just generating from nothing. You need to input your own image as the prompt and then it works on that. It’s technically (or was) an image upscaler. But they added some hallucinations into the algorithm and gave people a slider for how much to let it hallucinate. We looked at it a while ago to upscale some of our old texture sets we lost the higher res versions of and it worked very well as just an upscaler. Better than topaz would.
Ya idk it’s definitely a shortcut. These are all pretty basic details you learn in environment concept art or environment art. I think everyone needs to be a multidisciplinary artist to succeed, but if these new artists aren’t interested in learning about architecture, how wood gets polished with time and wax, etc then idk if their work Will be interesting with it without ai
Cycles is seriously underrated!
From what I know, Cycles is lacking just a few features that would make it professional grade.
- Variable Compression Rates for render layers (especially cryptomattes)
- Render Layer Denoising (although to be fair, you could argue professional workflows would brute force high sample counts without needing denoising on their farms)
- DeepEXR support
- a more robust ACES integration on render layers, when used in the compositor
I am not a programmer and don’t know how difficult these are but it feels like a comparatively small mountain to climb to make cycles truly great
The first two already exist, we use them - you just have to output a custom EXR
@@FutharkStudios You mean by using the compositor as an intermediate to transcode the render?
@@MegaIrgendetwas yes, it’s quite powerful once you make your own custom exr. Splitting Denise and compression methods based on what you require.
maybe he got lucky with the glass shader there. But if you try to make a thin plastic sheet beacuse something is wrapped in it, you are loosing so much light, it's insane. So from my knowledge, the glass shader is still broken.
Then there's Octane in Blender.
Amazing
Very interesting talk. I really thought that nothing shot was from some other e fine. Remember seeing that ad. I do think the title of the speech is perhaps a bit misleading. I was kinda baffled about the teaser, magnificent work! Also very interesting method for the lights. Yet I wonder how they comp that. Because most times you use the passes. Overlaying so much details, parts start blowing out. I mean in some of the layers, we also see details which are also in other shits
if your lights aren't set to too bright values, in a linear workflow you should be able to just add those passes together to get the right result
@@theminecraft4202 yeah but since we see other details in his passes, they all get added. Since we use addative
Would love to see some tutorial on how to do it...
He was so preoccupied with whether or not he could that he failed to stop to consider if he should.
Cycles is probably the best of the free renders. But in terms of photorealism quality it falls just a little short of Octane, Arnold and Redshift
Basically, using AI like this is like having personal teacher who makes improvements to your work so you can see where your messed up and what to do to make it better. More or less.
Those who try to do something and fail are infinitely better than those who try nothing and succeed.
Love the render times. I have a 3060TI, and can get scenes to 10 seconds a frame pretty easily
Very cool ideas. Gotta start using AI to level my Blender skills!
Cycles doesn't have path tracing options that would make lighting more natural and it must be intentional, because if you were serious about developing your 3D program this would be a main feature.
What is better than arnold ?.octane??
this guy should see the mega light in unreal engine 5.5 :)
It didn't take long for Gary to detect the robbers were amateurs.
I love the workflow with Ai. However, what I am wondering is, how long will the newly gained information (tricks) stick with intermediate artists. Is it like a test at highschool that after a quick rehearsal you forget about? How much will we rely on fresh input from AI in the future? Just a thought
The ai training is super cool
I love Blender, but we have to be honest.
Houdini is the best of the best of when it comes to simulations.
Blender simulations are a bit of a pain and nowhere near as fast as Houdini.
I don't think anyone challenged this, and the talk is about Cycles. I wish Karma gets as fast as Cycles at some point.
well, that's an offtopic.
This video is about a rendering engine whereas your comment is about a DCC. Sorry but, it’s misplaced here
Did you watch the video? They even showed Houdini used with Blender.
Nice talk !
"Cycles is better than you think"
-Blender
I dont think anyone doubts cycles anymore at this point. This whole "IS blender good enough?" mindset is outdated
'course it isn't, I LOVE cycles
BLENDER ALways BEST than any studio Standar software
even as someone very critical of the ethics of the ai marketing blitz I don't see a problem in using the output as reference images; artists reference completed works all the time. if anything the bad part is giving a company money for stealing content from smaller creators.
Only other issue to me is that there’s not much reason to be using ai as reference when there’s so much more value in somebody just learning how to properly use real reference
@@warcat3d I think the main issue he was talking in is that you might already be using reference but stuff still feels like it can be improved so AI is meant to help you visualize tons of different ideas that's based on your artwork. Sometimes you cant realize what your missing based off a different image. Like with the chair in this video i dont think a reference could have made you think of going through the work of adding a chair in the corner. Its more of you start off with real reference images and then when you feel stuck you can use AI to give you ideas on what you have so far.
Bro the goat
Cycles is great for lot of things, organic objects like realistic skin and hair not so much - overall Arnold still wins. But cycles is getting better all the time.
ALEX SAYTIEV ‼️‼️‼️
👏👏👏👏👏👏👏👏
I thought for a good second that this video was gonna be about steroids but then I looked at the profile picture of those who produced it...
funny did the same with crea ai, just with a blockout of my scene, it worked out super and i implemented a lot into my scene
A lot of “Epic Bro” CGI.
5:40 - To fade out the edges, Does he mean add a texture to the light or to the object?
I think he means to add a texture to the light. This gives a pleasing gradient fade to the light, which looks really nice in reflections.
@ what kinds of textures, are they normal gobos. What kind of gobos?
I primarily use 3ds max + vray. Correct me of im wrong. Isnt the method he mentioned seperating different light groups and rendering different images, same as just adding render elements in vray(or most other engines). And adding them later to get the beauty pass? Or we can just add back to beauty element and it automatically seperates emmissive mats, light groups, reflections, refractions, environment and so on.
I think render elements are just render passes in Vray if I'm not wrong, where here he's using light groups to put certain lights in groups (which is a relatively new feature in Blender) to be rendered them separately and put them back together in comp.
@@xanzuls yes, we add different render elements to render corresponding passes along with beauty render. And there is one render element named "back to beauty" which basically creates all the passes automatically. For light grouping, there is something called Lightmix. I use it to render day and night scenes simultaneously. You can turn off/On any lights,emmissive matetials, change intensity,color after the render is complete and save variations. And whichever lights you turn off will also affect the underlying passes. So you can have all the passes for each light combination switched ON/OFF.
I was just surprised that rendering different passes was a new thing..I mean it's a method used from decade. . I thought blender already had something similar to render elements to automatically render the passes.
@@punithaiu render passes are not a new thing, but that's not what he was talking about when he was talking about rendering out lights separately, as I think he used light groups in blender and light groups are a similar feature to lightmix in vray. Blender has render passes, and you can choose to render them under the 'view layer' panel.
@@xanzuls thanks for clearing that up.
3:16 PLAMET OF THE APES REFERENCE (the window)
my goat
Nobody using Cycles in workflow production since 1994, you ever worked in studios or it was just in your bedroom ?!?Arnold, V-Ray, Renderman, you know theses engines ? No comment...