Houdini 20

   27409   66   5
User Avatar
Member
119 posts
Joined: 8月 2015
Offline
Even before AI the amount of "art", be it concept art, 3D sculpts, digital "natural" media paintings etc. you could find on the internet was immensely inflated - don`t get me wrong, the skill displayed was and is marvelous, only that the style already seemed to mostly have a heavy bias towards recurring themes and motives.

Midjourney and co. in my opinion, take these images and bloat the amount now to epic proportions. A lot of them might look quite nice, but very soon (I guess it`s already happening) nobody will appreciate them and, even worse, nobody will appreciate the real artist`s works.
Edited by Sygnum - 2023年8月31日 04:55:38
User Avatar
Member
313 posts
Joined: 10月 2016
Offline
Sygnum
Midjourney and co. in my opinion, take these images and bloat the amount now to epic proportions. A lot of them might look quite nice, but very soon (I guess it`s already happening) nobody will appreciate them and, even worse, nobody will appreciate the real artist`s works.

There's an off topic warning here, but here in the rather relaxed Lounge I'll let you be the judge of that. There are so many aspects of this. In fields like automotive industrial design or architecture I have a hard time to accept the idea of any AI artist replacing the human or even assisting the production, maybe someone know something about that? My reason for this is that with the most skilled artists the artwork is secondary, and the actual ideation and refinement process is key. Simply compare this with a book author, where actually the writing process is not really the effort. Replacing the skill of a great artist with prompts and sentences seem to me as quite a frustrating experience with simply the wrong tool.

However, for more or less inspirational images where cliches are welcome and there is simply no budget for real concept art processes probably there is a market space for "cheap" prompt based machine generated images.

Frankly I've not followed the development closely but believe your are probably right. When things become cheap and anyone can do it the value basically drops. So, while on one hand maybe a lot of upcoming artists get discouraged to continue (I'm imagining this to be likely as a former art student myself). Maybe the real case is opposite. I have a feeling that the need for original artwork and real concept artists might actually increase when people get sick and tired of the cliches in AI art.

Is not the uncanny valley effect related to when machine generated images has a certain style? (i.e. being too close to realism, but not quite there)

Edit: Once again I took a look at the Midjourney showcase [www.midjourney.com]. This technology look a bit like advanced matte painting (been there for a while). My guess is that they made an algorithm that can basically reverse engineer input images, analyze the components, and then reassemble the images while changing things such as the background, the setting the style, colour and materials etc. When looking at some of those creepy hands it look a bit like two or more images have been masked together in a clumsy way. Seeing how bad that looks my only conclusion is that the AI does not even understand how hands should look like. That said I can't believe for a second that the input data does not come from millions of resources from copyrighted contents. There is a lot of "quality" art behind the images such as composition, lighting and style, photography and much more and those things are not for free and there is no way any AI can figure those things out without simply being fed a lot of such data. It makes me kind of upset simply watching those things, because it just look like a lot of people are getting their rights messed with by the use of technology. These are just the things that legislation is supposed to protect us from, and those researchers simply found a way of how to use machines ability to manage tremendous amount of data and get around copyright limitations.
Edited by SWest - 2023年9月1日 17:11:58

Attachments:
creepy-hand.png (66.2 KB)

Interested in character concepts, modeling, rigging, and animation. Related tool dev with Py and VEX.
User Avatar
Member
111 posts
Joined: 1月 2018
Offline
as for AI in Houdini i think theres another use.
This time with this huge Tab menu with all nodes. id love that AI will pick 3 or 5 nodes that could be used as first (on top of menu) (like with chess algoriyhm prddicts next move)
all this based on my work (AI model will be individually trained)
so fe. when i do destruction work i pick NEW model and start recording (training) so AI model with time gets better to prompt (suggestion) my next node so i can pick it with one click and not śearching whole menu... or evrn worse memorizing and typing node name (its too geek for me xD)

so yeah i would like that AI learn my workflow and have good suggestions to speed up working process by greater UI/UX. (this time in Tab node menu) but this can apply in search too (this new option in 19.5 that i havnt tryed yet xd)

but i'd like to have full controll over those models. like: create, delete, merge and shere (with studio members or community) basicly all what open source stable diffusion offers in terms of managing models itsels - like Lora, embedings etc.

But all this only for Tab menu where i spend most time and it slows me terribble.
AI COULD ALSO SUGGEST ICONS/ACTIONS FROM SHELF TOO!


Another area where Houdini could learn my behaviour and preferences could be Animation.
FE. for start i will train custom character that is alreafy rigged with spline control objest (one that u actually animate) so for start i could use some mocaps to train basic moves, i could pick one that reflect character objectives. like this character knows kung fu . then i,ll set my custom poses using "spline controll objects" recording/training AI model. AND finally animate character using those trained models as helper. (firts i teach AI, then it do what i wanted and speed my work.

Damn. actually any object in Houdini 3D space could have such XYZ AI atatched to train its movement in 3D space, and then use it for any purpose to speed up any work.

Even nodes themself could be traind and instead of default settings and presets they could learn our preferences (with avareness of whole node structure) so FE. it will setup big render resolution if we are making stuff for print fe. etc. This could speed up tiedious node setting too.

I WANT AI IN HOUDINI ALL OVER PLACE NOW!

and in setting nodes in materials too!

and AI that is pretrained and can change or duplicate materials when i change renderer
fe i make materials in redshift that is fast and then i want render is in Arnold on farm.
so AI makes new Arnold materials based on redshit! cool AF

ups sorry for long post but ideas just pop in my heqd and sorry for bad english.
Edited by oldteapot7 - 2023年9月1日 22:07:46
User Avatar
Member
313 posts
Joined: 10月 2016
Offline
Okay but just a short reply concerning your amazing Houdini+AI dream :-)

For such things to work a lot of data has to be fed to those models. There is actually no real "AI" (hear me now) just a bunch of data and an algorithm that selects from that data. A CPU has actually no idea whatsoever what is going on (it just crunches numbers).

So the best quality AI would probably only be possible if Houdini experts and gurus are willing to open up their workflows and share how they do things. Do you really know many who would welcome such a "free and open" world. My guess it would rather be those who want to learn who would welcome such a thing rather than those who did the long end demanding journey to become experts in their field.

I'm not really convinced about having AI monitor a single user's methods and then suggest uses from that, because my guess is that the amount of data is too little. Furthermore, in my experience, I don't have very much benefit of "learning from myself" but the best is when you learn from someone who have more knowledge and better experience than you do.

That said, I do also believe there are many possibilities to improve the user experience with maybe some kind of "AI-ish" suggestions. Although Houdini has a lot of power and allow access to as much low level as a TD or artist can want, I also find this to be rather time consuming at times. As mentioned, templates can be used once something has been done (referring to shelf features now and node presets).

However, what is probably feasible is to have some machine learning dig through the various learning materials that is already made available (for free) and combine that will all the available documentation. Based on that it could be possible to do simple searches for tasks and generic solutions and quickly find suitable guidelines. Maybe even Houdini automatically could recreate such a thing, meaning create basic setups, as a starting point for further customization. This would simply just be the next level of what is already done with the shelves and the examples in the docs.

Cheers! (keep em coming)
Edited by SWest - 2023年9月3日 04:18:13
Interested in character concepts, modeling, rigging, and animation. Related tool dev with Py and VEX.
User Avatar
Member
396 posts
Joined: 4月 2018
Online
Edited by eikonoklastes - 2023年9月2日 01:47:38
User Avatar
Member
30 posts
Joined: 12月 2016
Offline
"I WANT AI IN HOUDINI ALL OVER PLACE NOW!"
This would catapult houdini into the next paradigm shift in graphics software.
In reality things have changed so much since this last evolution of a.i. That now 3d companies need to design for a.i instead of users if they want to stay relevant. Houdini has the advantage that it is very programmable and you can do almost anything. So having an internal a.i that simplifies the software by 90% will make houdini the default winner. I think thats the only way it will survive in a world where creating simulations, videos, with prompting will just get exponentially better the only advantage a 3d package will have is if you can use it to prompt and adjust things with its tools.
Edited by tsiwt - 2023年9月2日 23:56:04
User Avatar
Member
17 posts
Joined: 4月 2020
Offline
in my personal opinion i dont wish for AI to evolove that much , it must be limited to some extent , never a bot will replace artists and their long learning journey ,its just unfair
User Avatar
Member
313 posts
Joined: 10月 2016
Offline
d3dworld
in my personal opinion i dont wish for AI to evolove that much , it must be limited to some extent , never a bot will replace artists and their long learning journey ,its just unfair

Well I hear you. It is a scary thought and in probably 10 or 20 years we will know how much humanity will allow this new syntheric organism to take over. Unfortunately history has show that humanity learn its lessons after things go to far. So I’m expecting the actual ”rise of the machines” and then the collapse will come and we start over again with a new lesson.

The thing with self-learning technology is that there is probably no limit. It is a machine made to keep redefining the data. It is an illusion of a living organism just like an actor is playing a role to be alive. It seems alive but it is just data and an algorithm.

Anyway, although technology can evolve fast it is we as humans that can be stubborn, slow down or limit those things. We need to recognize that it is in the interest of corporations and capitalism that machine learning should take over more and more market shares. Humanity will need to ask itself what kind of life do we want for ourselves and our children?

Anyway, enjoy your life as it is as long you can.

Cheers!
Interested in character concepts, modeling, rigging, and animation. Related tool dev with Py and VEX.
User Avatar
Member
327 posts
Joined: 5月 2016
Offline
SWest
d3dworld
in my personal opinion i dont wish for AI to evolove that much , it must be limited to some extent , never a bot will replace artists and their long learning journey ,its just unfair

Well I hear you. It is a scary thought and in probably 10 or 20 years we will know how much humanity will allow this new syntheric organism to take over. Unfortunately history has show that humanity learn its lessons after things go to far. So I’m expecting the actual ”rise of the machines” and then the collapse will come and we start over again with a new lesson.

The thing with self-learning technology is that there is probably no limit. It is a machine made to keep redefining the data. It is an illusion of a living organism just like an actor is playing a role to be alive. It seems alive but it is just data and an algorithm.

Anyway, although technology can evolve fast it is we as humans that can be stubborn, slow down or limit those things. We need to recognize that it is in the interest of corporations and capitalism that machine learning should take over more and more market shares. Humanity will need to ask itself what kind of life do we want for ourselves and our children?

Anyway, enjoy your life as it is as long you can.

Cheers!
not to derail this topic, but watch Colossus, the Forbin Project. it came out in 1970
User Avatar
Member
374 posts
Joined: 3月 2009
Offline
Can we please stay on topic and stop opinionaing on the world with AI?
User Avatar
Member
313 posts
Joined: 10月 2016
Offline
LukeP
Can we please stay on topic and stop opinionaing on the world with AI?

Did you mean, let’s not discuss long term implications of AI? Or are you asking for a scientific or authorative statements? What world do you live in? Opinions are what drives a free society.

On the contrary, there is an ethical side to the implications of machine learning. I don’t know if you are aware but there are both smart and highly educated people who are concerned.

In the field of trading of stocks machines has already taken over the intense part. I can assure you that in areas where speed and vast information is key the machine learning algorithms will easily be more effective than human ability. Now, problem solving is already implemented and research seem to be on the way with simulations.

Some predict (I recently watched a documentary in my native language) that content creation will be done with AI.

Whether we like it or not machine learning is here to stay the question is probably only about how and where we want it.

Cheers all
Interested in character concepts, modeling, rigging, and animation. Related tool dev with Py and VEX.
User Avatar
Member
119 posts
Joined: 8月 2015
Offline
LukeP
Can we please stay on topic and stop opinionaing on the world with AI?

Of course, so we can further discuss a topic which will be soon have no point when H20 is presented...

Regarding the more interesting offtopic - as of in this moment in time, the implications of AI will primarily be, that the amount of output a solo artist will be able to create will literally explode. In my very own case, in the area of compositing, I`m capable now to do 95% of rotoscoping fully automatically and speed up a lot of retouches, clean plate generation etc. by ten to a hundred times.

So in my opinion, to even have a miniscule chance against generative ai we urgently need this technology to get a bit of ease at least in mind numbing areas like retopo, uv unwrapping, rigging, certain areas of animation and so on. Because, let`s face it, 3D has gotten so insanely complex it`s literally impossible to even have a halfways decent grasp on the numerous areas of a cg software.
Edited by Sygnum - 2023年9月5日 02:19:17
User Avatar
Member
111 posts
Joined: 1月 2018
Offline
First you cant stop evolution of technology. i think its even scientificly proven that this beast just accelerate and theres no way to artificially suppress or delay it.

If SideFX from any reason wants to delay introducing AI in Houdini its stupid and i want to know it before ill invest more time into learning the software.

If SideFX fe. wants to save industry from changes so they could sell more licenses or something like that its very bad thinking and an dead end.

Its a race as always so if not SideFX then fe. realtime 3D software that using nodes will introduce AI. even Blender could do that. or some 3rd party software company will introduce some one button AI sollution plugin for Houdini that will use prompts to make any explosion you like

I agree that real battle is betwean 2D generative AI models and old school, time consuming, expensive 3D. and 3D one needs AI acceleration ASAP.

Imagine that 2D generative AI could do any explosion you like without even touching 3D.

Anyway i still think that Houdini is the only software that is ideal for AI thanks to its node base aproach. And thats why iam learning it becouse it have future.

I hope that SideFX understands that too
Edited by oldteapot7 - 2023年9月5日 05:39:18
User Avatar
Member
313 posts
Joined: 10月 2016
Offline
There is no chance that the staff has missed the train with bells and whistles. However, with great power comes what? There is also the competitive side, like you mentioned and most likely they can not talk about it before it is released. So I just encourage you to wait for it. It is also possible to develop things right after the competitors and do it better.

Cheers all!
Interested in character concepts, modeling, rigging, and animation. Related tool dev with Py and VEX.
User Avatar
Member
279 posts
Joined: 6月 2016
Offline
There is already AI software to help 3D artists, like Cascadeur, or Audio2Face, but you can't expect a magical button that will do it all for you, AI is repetitive, and it has problems with hands for characters, I have an FB page for artists, and I get tons of post requests with AI images, I decline a lot of those because AI tend to produce very similar images, to the point that many looks basically the same, I feel the same thing would happen with AI VFX!
User Avatar
Member
130 posts
Joined: 6月 2019
Offline
AI hype is over, thankfully.
I doubt we'll see anything in H20 about ML. It doesn't seem to be the focus right now and sidefx is relatively small company to pour resources into rnd.

I think is going to cool down in upcoming years, revolution is cancelled imo in two-three years you won't be blown away. It'd be still clunky, barely controllable assistance that you can use very efficiently in some parts of pipeline.

The most interesting what we'll see in H20 realistically
  • New rigging based on compiled blocks is basically confirmed
  • That means continuing abstracting away SOPs as data operators. So we should be prepared to embrace completely new graph editor and "metagraphs" I guess Graphs that build graphs. Could be fun and actually this is something that I've never seen in node based apps
  • There is a mention in illume presentation that husk procedurals won't require license. I really hope we'll see some sort of new lightweight engine capable of invoking cached compiled blocks. Like Houdini Player so no license if you don't want to author graphs (similar to Substance). I guess it could be alternative to OpenExec
  • COPs 3.0 is probably not ready, but hopefully the foundation will be in H20. So CPU/GPU wrangles for volumes. Personally I'd love to see HLSL for that, but it could be still vex. Though I'm not sure if it's technically possible to introduce gpu target for vex. Anyways, hopefully H20 introduces new compute backend
User Avatar
Member
111 posts
Joined: 1月 2018
Offline
Well.. AI hype isnt over, it just starting. its not true that AI cant handle hands correctly, its common missunderstanding how ML works. Basicly AI hallucinate when it dosnt have enough data. its easilly fixable just by adding more training data. and hands were fixed that way. Also its not true that AI isnt contolable. thers many ways to controll AI output now. Just catch up guys

Now its hype about video generative AI and it is actual threat to 3D VFX.

i dont think that in H20 will be ML but H20.5 is a must.

something like Nukes CopyCat would be great
User Avatar
Member
119 posts
Joined: 8月 2015
Offline
elovikov
AI hype is over, thankfully.
I doubt we'll see anything in H20 about ML.
I think is going to cool down in upcoming years, revolution is cancelled imo in two-three years you won't be blown away. It'd be still clunky, barely controllable assistance that you can use very efficiently in some parts of pipeline.

As critical as I am of numerous aspects of AI, I think this is an ignorant statement. We´ve barely just seen the very tip of the ice berg. The bean counters will make sure to exploit the technology as much as possible. If the hype was indeed over, then you`ve ignored the WGA and SAG-AFTRA strike which is going on for months - partially also because of unresolved issues involving A.I. usage. We´re seeing court cases and rulings everywhere on the planet due to this "cancelled, cooling down" revolution...
Edited by Sygnum - 2023年9月5日 11:12:28
User Avatar
Member
313 posts
Joined: 10月 2016
Offline
Edit: For ethical reasons I’ve decided to censor my post here. I just don’t feel comfortable about it.
Edited by SWest - 2023年9月5日 14:58:04
Interested in character concepts, modeling, rigging, and animation. Related tool dev with Py and VEX.
User Avatar
Member
279 posts
Joined: 6月 2016
Offline
Like I said, I have different peoples using different AI software posting images to my FB page every day, and I have yet to see proper hands, and the fact remains, AI produces very predictable results, very repetitive, it is very easy to spot an AI image, AI has a long way to go before replacing humans, if ever, some peoples have been waiting for years for that magic button that will do the work for them, while others prefers to actually do the work!

Attachments:
365171958_134908266330924_6233780722364796949_n.jpg (121.4 KB)

  • Quick Links