I have a group of friends who are professional artists and they are not big fans of generative AI. I've tried my best to explain what it is and isn't and a few of them are interested in taking a more serious course on the subject.
Most of the learning material is either very technical and geared towards programmers or too vague and hand wavy to be practical to someone who wants to incorporate generative AI into their artistic workflow.
Most of them are familiar with text2img models, but ideally I'd like something that explains _how_ diffusion models work, their probabilistic nature, how they're trained on images/text and maybe even a high level description of what the different parts of a setup in, say, ComfyUI are doing (LoRA, VAE, CLIP, UNet models).
I admittedly have a tough time judging the technical expertise of people, but I think some of the resources I've personally found great for explaining the basics (3blue1Brown, Andrew Ngs course, ComputerPhile videos) are still too technical and impractical.
What resources would you point your friends to who both want to learn about generative AI and assuage their fears that AI will make artists obsolete?
Sure, you still use a prompt in those, but it's not really the focus in many cases -- you just use it to give a very rough guideline of what you're trying to do, then start drawing.
Depending on approach you'd either start sketching your astronaut and see the AI turn the sketch into something polished in real time, or you'd generate an image then polish up the details. I think most artists are more likely to prefer the first approach.
This sentiment is reminiscent to the attitudes around photography when it first became practical. People wondered how photography could be considered a form of art when all one had to do was point and press a button. The amount of effort is minuscule compared to what it takes to depict the same subject using traditional media. That being said, there is a lot of skill necessary to capture a photograph properly; lighting, composition, shutter speed, exposure, and so on. I agree that generative AI images will disrupt the market segment for stock photos and "clip art" used for articles, presentations, et cetera.
There will be a need to study and acquire skills related to the use of generative AI image creation. While technology will make simple prompts "good enough" for most outputs, just as our smartphone cameras make taking a picture "good enough", people will still need to study and practice in order to make high quality output.
The best tools in this space are spatial editors and node editors. Text is weak sauce.
Also, you can't use text modalities for film and 3D. It has to be art tooling to get good results.
Artists will always have a job. They'll just be doing way more than they were before.
I think style transfer + pose detection + edge detection gives artists a bunch of new tools to play with.
OP I would look into learning comfyui
if the amount of new offer is not paired by demand, there will be pain.
The reason why this doesn't appear to be clearer, is that this field if filled with tasteless technology-driven engineers that think they can explain or reduce everything to numbers.
Go talk to your friends, and go to their exhibitions. Witness and celebrate their process. They don't need "courses" on such a misplaced and self-entitled technology.
Not everyone finds solace in this strategy. Understanding how the blade is forged as it pierces you may be a mundane final thought. But if you survive I suppose you can use that knowledge to pierce others and destroy their planned futures. Let them pivot or let them starve?
Pivoting to some hybrid of generative AI with human edits is appealing to software developers but soulless and essentially anethema to many others.
Step 1 - get them to sign up for AI image tools.
* Midjourney is best for quick images
* Playground AI is good if they need to modify images but the quality doesn't need to be perfect
* Leonardo AI (now owned by Canva) is a good full suite
* Photoshop AI feature is best if they already work in photoshop
Then show them how to use these tools! That might require you signing up for these tools first and learning yourself.
Step 2 - For learning about how AI image generators work here's my video list.
1) AltexSoft - has a low viewcount but it's a great overview - https://www.youtube.com/watch?v=Rke0V_VkF3c
2) Jay Alammar - it's technical but also visual and he explains it well - https://www.youtube.com/watch?v=MXmacOUJUaw
3) Gonkee - again, technical, but visual, great - https://www.youtube.com/watch?v=sFztPP9qPRc
Workflow example: good for seeing the workflow of SD as of May 2023 - https://www.youtube.com/watch?v=K0ldxCh3cnI
Too technical for what you're looking for: Computerphile, Ari Seff, Jia-Bin Huang
Step 3 - For assuaging their fears about becoming obsolete - I think the following is a great podcast episode. https://80000hours.org/podcast/episodes/michael-webb-ai-jobs...
But their fears might be valid. A test is perhaps: if their boss spent a few days learning to use AI image generators, would they still need them? For some artists the answer would be no, for many the answer would be yes. It'll change over time as the tools get better, but that's a pretty good proxy. If they're doing things that require more iteration, interacting with users and humans and the physical world, nuanced judgement, in-person work, safer. If they're doing things that are contract based, no iteration, get a request and deliver a result, much less safe.
I haven't found any particularly good courses, however I do enjoy this YouTuber, who covers Midjourney and some other generators:
https://www.youtube.com/@WadeMcMaster
The key thing you should try to do is this: make them understand that genAI is a toolset, akin to Photoshop.
This would for instance be the Krita plugin and InvokeAI. Both heavily cater to an user with little interest of getting into the weeds of VAEs and UNet models, and is much more at home using a drawing tablet.
Krita is closer to being artist-targeted since it's primarily a drawing application with an AI add-on.
InvokeAI is more on the opposite side, an AI tool with primitive but still usable ability to guide the AI by sketching stuff on a canvas.
To OP -- We work with professional artists regularly, and I'm seeing things pick up as more begin to understand the potential for creative control. Artists mainly want to be afforded creative flexibility and control, and need an interface that feels natural for their workflow.
Invoke is OSS, we release continued training/education on a weekly basis (free, on YT) and we'll be releasing a simplified installer soon.
These things hardly matter for artists. Most artists don't even know how a Gaussian blur filter works and they can still produce astonishing artworks.
Invoke AI has incredible open source image editing tooling and a fantastic YouTube and Discord that pair with it.
Curious Refuge (paid) has a great community and paid lessons for AI filmmaking.
Reddit is overflowing with AI art communities, many of them technical in nature.
https://www.reddit.com/r/comfyui/
https://www.reddit.com/r/StableDiffusion/
There are a lot of offshoots of these, and Discord will provide the best resources once you get plugged in.
Whats the goal here? Do your friends express interest in using AI within their practice? Technical or not, an artist doesn't need to understand it to achieve that... Many years ago before I better understood it, myself and other artist used generative AI to create elements of our works and no one knew or cared that we did.
There are a lot of other interesting Computer Generative Art that are not Diffusions. Maybe those a more genuinely interesting to your artist friends.
The details of how it works mostly won't matter unless you're trying to get really specific with how a plugin works. For a filter in Photoshop, you don't find many tutorials on the underlying algorithms, you usually find guidance on when and why to move the slider.
AI will eventually basically make artists obsolete just like every other job. You should focus on the creative possibilities of the present. Things like tools or ComfyUI nodes or whatever that allow you to do things like "auto complete" digital paintings based on live sketches, change a style, generate variations, etc. Focus on the capabilities of the tools in terms of enhancing or producing art or testing ideas. Show how an artist can train a Lora on their art and then use a scribble ControlNet to instantly go from sketch to a manifestation of a finished digital painting in their own style and evolve it with each stroke.
Try making a Jackson Pollack using AI, AI can only effectively make objective images, given it requiring linking similar images via limited text descriptions.
Making headers for newspaper articles or blog posts on the web might be OK if you don't care much. Check out https://www.quantamagazine.org to see how real artists do things. Their header images are really imaginative given the subject matter, something AI today can't do. It might be instructive to take their articles and try to make header images for them.
If you make videos, then its 1080p which can be done by AI. But video has its own issues with video diffusion generators having limits on continuity. They are getting better, but the longer the clip, the less control you have and the longer it takes to fiddle with the prompt(s). It won't ever replace real movies unless you enjoy the SciFi channel or Sharknado. That would require an entire new kind of AI which we don't have today.
No it won't. We've hit a wall and the low hanging fruit are gone. OpenAI hit a wall and is searching for enterprise customers to maintain their valuation, and Google and Facebook are calling them out on it. Meanwhile, text is entirely inadequate for manipulating visual things effectively.
AI will supercharge artists that embrace it. Text-to-art is not good enough and real artists will be working with layers, node editors, and in 3D/video modalities.
The trajectory we are on is for AI to be able to do most intellectual jobs in less than ten years and most physical jobs in less than 20. That is speculation but it is based on many decades of increasing performance and capabilities.
Text does have limitations but AI tools and models are not limited to text to image and new models and systems will be created.
I agree that AI will help artists in the short term. But we should anticipate that humans will eventually just be a bottleneck in the creative process.
ideally I'd like something that explains _how_ diffusion models work, their probabilistic nature, how they're trained on images/text
How they work is probably out of scope of an artist. They’ll figure it out given some knobs, and honestly you cannot explain any of it, you have to get the feeling. Cause it’s not only probabilistic, but island-y. What worked yesterday may noy work today. You have to train yourself on these parameters.
Comfy
Is basically a bash-like visual programming. Cool for developers and repeatable workflows, but overkill and overwhelming for a beginner.
Problems.
The most problems come from python management. Python versions, libraries, plugins breaking the env, etc. you have to prepare some pre-built folders (or cloud images) for them. So that it just works. Then I’d download a few popular checkpoints andshow them the inpaint (mask) tab in a1111. I think that this tab will create a good first impression required for further interest.
As a sort of a poor artist myself, it was amazing to see how SD can take two sprites and inpaint them together across a masked edge. Or replace a segment with a completely new content. (Although for them it may look meh)
Txt2img is not that wow, cause everyone is already tired of it.
Can’t point to resources, sorry, I learned from the internet by googling all my questions.
I can draw well. But I don't associate my identity with it. So I am not an artist. I write well but I am not a writer. I am happy to delegate these forms of work to AI, the same way I'll delegate laundry to washing machines.
For someone who calls themselves an artist, their identity is centered around their ability to draw.
Photographers will use editing tools. Most don't mind tools. A photo editor enhances the photo, the core still comes from composition, lighting, etc.
I would start here. Show that it could be useful first. How can AI white balance dozens of photos? A favorite use of AI among artists is doing different resolutions, expanding a small image into a capsule or cover image.
I reckon there are some hurdles to overcome before seeing a lot of non-technical generative AI imaging uptake among advanced and professional artists. A) most of these tools UIs aren’t useful to traditional artists in any part of the direct creation process yet, B) the prompt-based workflow, conceptually, doesn’t fit into traditional art processes so it’s essentially starting from scratch, and C) some vocal denizens of the DIY NN scene have done a fine job of making it as unappealing as possible to traditional artists.
I know plenty of artists— both commercial artists and professional fine artists— that use simpler generative tools like midjourney for mood boards, reference, etc. but don’t know a single one that directly uses the output to make art.
One big problem is that the UIs of all local stable diffusion front ends are not designed for people to make art— they’re designed for people to operate a neural network image generator. From a technical perspective, that’s the same thing, but from an interface design perspective, it is a very different goal, and it shows. For people that are enamored with the technology, or for people with a technical background that are used to wielding a bunch of abstract acronyms and don’t have to build up a bunch of base knowledge, that’s great. For people that just want to express themselves and already have another way to do so, it’s a huge, annoying impediment. (And while it’s done with the best of intentions (usually,) assuming non-technical artists will be as enriched by the technical knowledge and tools much as technical people without art expertise comes across as pretty conceited.)
Another problem is that getting an image generator to create things in a way that makes sense in a normal artistic process is difficult. Art, generally, is about building things from broad stroke base to the finest finishing details with direct, granular control, and deciding/discovering what that finished piece will look like while you go, with all of the kismet and happy accidents that go with that process. You don’t usually make huge changes after you’ve polished it up because you probably wouldn’t have gotten that far with an element that didn’t work. When you start with something that has all of the “finished product” detail implemented already, that entire process is turned upside down. It’s distracting. Think about the way Bob Ross worked his way through a painting— do you think he would even want a workflow where he used words to describe a completed painting and then decided things he wanted to change about it? Not being an artist, encountering that tool is incredibly freeing. When you’ve put a bunch of time and practice into figuring out how to wield that artistic perspective and then use the tools you prefer to build and develop your visions from raw ingredients, it’s restrictive, disorienting. The thing you’ve been organically growing into for years that makes your art yours— the way you make basic shapes, flicks of the pencils or brush, ways that you might subconsciously separate tiny little background elements from each other, all of which are equally applicable with digital tools— is missing from this piece and has been replaced with an amalgam of many other artists hands. That’s the reason many people like generative image stuff: you don’t have to understand all of the tiny little components that make a piece what it is and just think about the objects and comparatively very-broad-stroke style. You’ve lost the verbs and adverbs from your process and are left only with nouns and adjectives.
Also, while obviously not representative of the user base on a whole, the active, vocal community of stans that can’t distinguish between Reddit and the rest of their interactions have done a great job of alienating traditional artists. Their gleefully predicting the obsolescence of traditional artists— using tools built with their work without permission, no less — makes it tough to tap into that excitement. A lot of us have mouths to feed, mortgages to pay, and cancer to treat, and while many disciplines are not as vulnerable as many think, people like concept artists are actively being screwed over because tech money decided they want to get paid for our hard-won skills instead of artists. This alone makes it a tougher sell than it should be.
Specifically the reading and tutorials sections may be of interest to you.
Edit: honest question, why the downvotes? Am I not contributing to the conversation?
There are some real dangers here that need to be navigated.
Junior artists who get their start doing commissions are not going to have that path anymore.
Reddit's job boards for junior artists are a literal desert now.
In a sense, the activation energy gradient to producing good results has lowered such that the economic reward for up and coming artists has deteriorated.
> Why don’t you let the artists maintain their own practice, rather than pressuring your friends to join you in yours?
Artists who get their start in AI early can build recognition and clout within the community. They'll potentially be at a big advantage.
a16z is investing in non-technical artists to bootstrap production and community. There's actually a lot of interesting stuff happening when you dive in to AI art.
It’s just market forces being unsure about the future of hiring artists.
Companies are like lemmings, they just follow one another.
Artists already know that gen AI is just another tool in the toolbox.
We’re just waiting for MBAs to also realize this.
GenAI making artists obsolete is a dumb idea. No MBA is going to be trying to coax a GenAI model into making an image.
The non-artists community? The "I don't have anything I can show I own the copyright to" community? The "I made this with genAI so now the company can't even protect it's copyright?" community? The "tech-bro NFT fraud" community? The "do it for the exposure" community?
Which community?
So disgusting.
> activation energy gradient
> producing good results
> economic reward
> build recognition and clout
> big advantage
> a16z is investing
> bootstrap production
you're saying a lot of things about commodification and mass production, which are completely unnecessary to the expression of the human condition.
I dont think anybody who has taken up an artistic practice, professional or otherwise, should or would spend much time basing their life decisions around an online job board.
You’re missing the point.
I notice a lot of non-professional artists running to the rescue of artists. I suspect this is in part because there are a lot of young people who have pursued a degree in art and have been frustrated by the mismatch of their expectations in the commercial world. There is a lot of romanticism.
I myself went to art school (MFA). Actually I was really grossed out by the big Art gallery scene. It professes to be open minded, anti establishment and non commercial but it is extremely narrow in politics, hugely elitist and deeply oriented around big money. That’s ok, but I wouldn’t claim that Art has any moral superiority.
I dont think any aspiring artist, professional or otherwise, was ever deeply concerned about their future prospects in creating corporate graphics or mechanical pop songs. It was a fast buck, that’s all.
Attempting to seduce them into committing ever more of their creative efforts in the service of a machine that only seeks to churn out more artifacts misses the entire point of ever starting, and especially continuing, an artistic practice of any kind.
The entire purpose is in the act of creating and the efforts involved in honing a craft that entirely relies on your commitment.
It doesn’t get any simpler or less pretentious.
The road to mastery is the entire bloody point.
Trying to encourage your own friends to short circuit that only serves in getting you to the artifact faster, and in the process neglects anything of any value from the entire process. Is that really being a friend?
Unless you think the only thing that matters here is the artifact. Then you’re lost.
Oh, you didn't read the whole post. I'd recommend you do, it'd help with understanding and replying.
> , rather than pressuring your friends to join you in yours?
The OP is literally saying they're asking.
There are more constructive ways of engaging.
Let's try something simple.
Have you ever used a pre-existing piece as part of your composition? Would it matter if that was not created manually?
What's the least creative part of what you do, are there any bland things you could fill in (e.g. paint these clouds or grass for me).
So if you make your art for "you", why should I care about it? If this is how artists feel then bring on the AI.
It seems like your post is mostly based on assumptions about the poster, and a misreading(?) of this one word.
There’s no hubris in wanting to assuage people’s fears that their skills might become obsolete.
> The important part of art is the practice and the transformation that occurs in the creator.
They claim to be engaging (talking to friends) to the extent that they can be without already being a working artist. Maybe they are lying, but all we have is a post on the Internet, so if we assume the author of the post is a liar, we don’t have much to go by.
> You will never understand this unless you engage. Trying to circumvent it again and again will only result in circumambulating the point and will miss it every time. No matter how shiny the artifact that results.
Do you actually work in art? I don’t. But what do you mean by engage? I’ve taken some art classes. It was fun. But this was a cultivated experience, I was paying the instructor to have fun and explore ideas.
On the other hand, you can poke around online and find people who will complain about the day-to-day bullshit of their art jobs. I worry that, almost by definition, the easy paths to engagement are almost by-definition not going to give a good view of what the day-to-day bullshit experience looks like.
People romanticize engineering and programming as well. The beauty of getting a tangible solution, sprung from your mind, that impacts the world. At the end of the day, a lot of houses and dinners were bought by the need to throw together boring assed corporate inventory management systems.
Most work is the sort of stuff that nobody wants to do, after all. That’s why they pay you for it, rather than going the other way.
I suspect we could drill down into what you are trying to say for a whole thread but the result is just going to be a specification for pretension.
Why don’t you shelve the pretentious corporate rhetoric yourself and re-read what I’ve suggested.
Get off the computer and pick up a pencil, or mound of clay, or whatever, and give it a few years of regular practice. Theres nothing pretentious about that. It will do everything and I will need to say nothing.
