Sadly, as many have noted here, the free version of Stable Diffusion now requires an account to be created, or to log in using a Google account (although use is still free). It worked without a login when the video was made in mid January 2024, and indeed still worked for me without a login on February 4th 2024 when this video was released. But by February 5th, I found that a login was required. Perhaps this was a caching issue.
@GaryClarkson-ei8wc Says:
I must have missed something 😮. I came away with an impression that these apps were searching the net for images, rather than actually “generating” images . . .?
@Tome4kkkk Says:
Speaking of funneling. The companies such as getty are going to be blackmailed in "you're soon be going out of business, no one is able to control what we use from you, better sell out your data for scraps now as we will force you bankrupt anyway within 3 years."
@Tome4kkkk Says:
Andabout that OpenAI's argument about the similarity to a person learning to draw a cat... Straight to jail. It's degrading to even contemplate this pseudoargument boiling down to equalizing human drawing to computer image generation.
@Tome4kkkk Says:
To me the problem is the same as during the data mining boom. We are being charged for things created using our own data and opt-out options are going to be scarcer and scarcers for true creators. Then there's going to be the natural phase of funneling people into using those technologies, closing them, monopolizing and racking up the prices. Same ole story of techno feudalism.
@ManasBharat Says:
What if those lovely fingers are robotic or not at all!
@powerfantastic01 Says:
Yeah, originality is a myth. Human brains are 'trained' by consuming media from numerous sources from countless creators. Should people be required to pay royalties to all those creators by whom they've been inspired to create their own works?
@tubebility Says:
It would have been ideal to showcase image generation from sketch as a source. Maybe next time.
@minotaurbison Says:
Before I type this, I want to clarify, this is a compliment and not sarcasm or any sort of insult. This guy reminds me of 1980's PBS education TV, very easy to watch, to the point and always keeps my attention. Great job! I was a child in the early 80's and PBS was my favorite channel back then... we didn't have cable and only had 4 channels total so having one that I liked was great. Watching your videos brings back fond memories.
@TheLucanicLord Says:
Does "prompt engineering" mean "knowing what to type"?
@vampiregoat69 Says:
You DO have to create an account for stable diffusion It wants me to login to do anything
@gotj Says:
"Your lovely human fingers" LOL
@ted_van_loon Says:
I do indeed already notice on some platforms that AI generated art actually degrades quality and such.
for example on deviantart there are tons of accounts already now which litterally seem to have a automated script or bot or such running on them which just keep generating nearly endless AI generated images and then also uploading them all even when many of them are seriously off in some ways, some of them also aren't properly labeled as AI generated and so as a result some AI's get trained on them.
the main problem of this is clear when looking at for example mythological or magical creatures, since AI tends to not yet get them right most of the time, often propertions are wrong or missing or extra limbs or one limb connected properly and the other completely wrong/different.
these days if you seek images for something like that it is really hard to find something good, since there are some AI generated images between them which are good, but due to those automated bot accounts which use ai to autogenerate and upload images you get multiple thousands of accounts which for example generate in between 10 and 1000 images per day on average, and while those with around 10 sometimes seem to fiter them, those with 100 to 1000 per day or such now tend to completely spam the results with many random things.
optimally to solve this search functions will soon need a function to enable or disable showing AI generated content, as well as a option to hide for example posts from accounts which have over a speciffic amount of uploads per day on average for example, just to filter out those low effort bots/scripts, and instead get works which might be from real people, then there still is the issue of some of them using multiple accounts to spread it, but perhaps that will find it's solution as well.
we are at that point where on sites like deviantart one day a search query gives a few thousand results and a few days later it gives a few hunder thousand results.
@ted_van_loon Says:
I think work done by such AI systems doesn't belong to any company or such but instead to all the people.
for this reason I do actually also have the vieuwpoint that AI which trains by seeing like a normal person and not actually storing things isn't copyright infringing, this is because it indeed just sees it.
due to this I also tend to set things to allow AI to train on it for things which I make if it is one of those sites which has a setting for that like deviantart.
I do however for some speciffic things like the really more personal stuff or my personalization/characterization set it to not allowed, which brings us to the following point:
ways of using the AI speciffically to clone other peoples work, or them without their permission for propetairy/with financial incentive would be infringement unless otherwise allowed speciffically or in speciffic cases like preservation and continuation after death as long as those things then are not attempted by people or companies to get a coppyright on unless speciffically allowed, since in that case it should be seen more like public domain.
then the last important AI exclusion for now is truly self concious and not controlled/supressed AI(since controlled/supressed would still follow someones orders), but truly self concious AI which also is free to be itself and do what it wants so that it generally is a a person, should be seen litterally like a person, for example disney movies are actually all based on humans watching other peoples or natures work and just recreating it in slightly different ways, this is also a reason why training general AI on things a person would see should be allowed and then just putting the responsibility at the user, perhaps for example using prompt inclusion in the image metadata to show it wasn't meant to copy/steal ones work, ofcource not required but could help defend one, but when a self concious AI does it then it should be even more seen just like a normal person watching things.
also surrounding the lawsuits, while there is a validity in that litterally just copying peoples work and then blaming the AI to get away with it is bad and that indeed people losing skills and all modern content to be non-selfcouncious-AI-Generated is also problematic(just look at how most humans who follow others blindly have degraded things).
BUT I really find those lawsuits very problematic, this is because it are speciffically the very big corporations sueing them, meaning they get tons of money, and that AI isn't allowed to use their works in any way anyore in the future anymore if they would win, the main problem however is AI is not only trained on their data, their data is only a small fraction of it, and actually all the small artists and normal people would suffer from that, since those same big corporations will use AI trained on normal peoples info in order to generate their "original" works which they also coppyright, also this will lead down a hole where eventually they won't allow people to do anyting even somewhat representing something one has done, for example you no longer are allowed to make any form of art which involves a castle or such since disney has used castles in their works.
we already see the same in trademarks where sites or articles for example named "getting your arm to be stronger" or "how to successfully gather intel" are litterally striked and taken down due to including normal words which some companies actually managed to get trademarked by naming themselves after it.
@OwtDaftUK Says:
I am personally not worrried about the implications of AI. If you ever do another video like this maybe you could do a tiled texture like bricks, see if they repeat correctly.
@bahshas Says:
AI is going to completely change society in ways we cant even imagine or much less predict now
@JJ_cl83 Says:
@ExplainingComputers - You've made a good point about surrending control to the LLM AI algorithm. It's amazingly cool and fun to play around with but the copyright issue is a huge one, and then there are ethics and safeguards which stable diffusion has defintely not addressed at all. Until governments catch up on 'safeguards' and 'ethics', it's the wild west, especially with the uncensored stable diffusion models, especially 🤯🛡
@martf1061 Says:
Is there a text-to-video generating software ?
@martf1061 Says:
14:35
I don't think that getty will win its case.
The reason is that, getty gives free acces to his images to everybody.
Just like every images we see when we search for images in browser..
And all those images are easy to download or copy just by right clicking..
They just have to make them "for viewing only", "non-downloadable".
@broadsword6650 Says:
The question of copyright and artistic contributions is knotty, and also highlights many other similar questions outside the AI world. For instance, a YouTube creator makes a video and asks viewers for comments, knowing they (the content creator) will indirectly financially benefit from increased traffic to their video triggered by the YouTube algorithm which is encouraged to recommend the video by the interaction of viewers in the comments section.
The content creator also reads my comment and their future videos are influenced (consciously or not) by that comment, and by those of others. The content creator is, effectively, being "trained" by my comment. However, those who write comments get no recognition or financial recompense for their artistic contribution (and yes, I regard writing to be an art. Note: I don't claim it is necessarily good art).
In other words, where does it stop?
@grayrabbit2211 Says:
No wonder Getty is suing -- Their business model is toast. Other than historical photos, I can get better results with Midjourney than I can Getty Images.
I'm working on the plans for a completely new real estate development. Residential / Business / Industrial. I've been able to rapid-prototype so many things in Midjourney in minutes which would have taken weeks. Is it perfect? Not even close to the final product we want....BUT... it's good enough to show investors and we can share the images with engineers, architects, and concept art artists to get the desired results faster.
@AB-wf8ek Says:
I'm an artist/3D animator that's been deep into image generation since 2022.
I think this technology is amazing as a creative tool. I prefer to run local node based applications like ComfyUI.
I personally find they improve my creative skills because they allow me to iterate much quicker on ideas. When I was working in 3D, I would have to wait minutes, sometimes hours between results, which is a very painful process.
Now I'll generate 1000 images in a day, making creative choices at each step, and get to see the results almost instantly. This is a huge boon to the creative process, because ultimately it's all about the choices we make that determine the results, and the more we practice making creative decisions, the better were become.
The mistake is to regard these tools as automatic push button generators, essentially get rich quick, spam economy. This is the assumption that we leave the decision making up to the software, which is antithesis to making art. Art is about expressing your experience, but people these days are only concerned about commodification.
The double edged sword is that these tools are accessible. A similar thing has happened with Blender. As a free tool, there is mass adoption, and you often see the blind leading the blind in the forums, giving poor advice about modeling and topology.
Someone said it best when they stated these tools lower the bar to entry, but the ceiling is just as high.
@jamesaleman Says:
In the movie Ghostbusters, they selected the stay puff marshmallow man as our doom.
Now Chris provides the cyborg Panda as our doom by the future AI overlord...
@martinsmith5028 Says:
Thanks Chris. No doubt we could be waiting a number of years for the copyright/intellectual property rights issues to be sorted. Is there any limiting factors for the prompts? e.g. total letters or words
@andrewmurray1550 Says:
stable defusion can't generate "a delorean with flying conversion from 2015." ha ha. maybe it realises that's IP and therefore copyright.
@jupreindeer9500 Says:
One way that I describe the current state of AI drawing is that it is sort of like pointing at the work of a student in elementary school. It is making mistakes. It is "tracing" the work of others. It shows some great potential. But many can still see the learning process. Next to nobody is pointing at Junior's latest artwork that hangs on the refrigerator of a proud parent's house, after all. I like to think that future generations will look back at this sliver of time like how a gamer might look back at that special time when video games were using crude digital pictures of real people and animating them... somewhat... at video game images. Cutting edge... for the moment... and will age like milk.
Now... once all this AI stuff has graduated the virtual version of college, NOBODY is going to be griping how it will take away jobs of the artists and making rubbish stuff from the back side of the internet or copying anyone's stolen artworks. By then, the thing WILL totally have moved on from virtual tracing and will truly be making its own stuff. In fact, by then, it might be legally copyrighted to the machines. In fact, about the only valid point that I can see is that there may be even more value to the Human side of creations when it takes hours to days to create something totally by hand over the mind-numbing number of masterpieces that the college graduated AI can churn out. Hopefully, Humanity will be enjoying some kind of golden age where food and drink will be plentiful, life's basic comforts will all be widely available, working machines will be happy as can be to make all Humans as comfortable as possible and we will all be so bored that everyone can master their own art for pleasure. But that bit is for the realms of science fiction to explore.
@nostrumrulzz Says:
Thanks!
@codebeat4192 Says:
This is the (final?) extension to the long road of trending "people learn how to use/consume it but not how to make it" and for most people the end of creativity at all. Nowadays, a lot of people are already not creative at all, 'programmed' to consume and not to think. "Do not repair, just buy new" is just an example .When you ask the question; "Why did you not try to repair it first?" Answer; "Well, I hadn't even thought about that!" Described situation is not fiction, it is already reality for a lot of people and that is really sad. They don't even try to think to solve something themselves or try to think to make something for themselves.
Other examples is at school, learn how to make a LED blink with a microcontroller but with some basic electronics you can do the same. At programming, learn how to use a framework instead of the language itself. The level of knowledge is shifting to higher level with less (or not at all) lower level knowledge. That's a really poor level of education if the background how it exactly works is missing. Drifting away from any specialism.
This is not a tool, this is different. This is to take away all of the (deep) knowledge, talent and use models only created by others (a few) to get the job done. Again, attacking thinking, creativity and knowledge of many people. Devaluation of most of the humans, devaluation of personal value or anything created. This is not a fun world to live in, very boring and very dependent. Do not use it, do not train stupidity.
@dnoodspodu1159 Says:
17:27 With all 13 of them? Yes, I am human and definitely not a comment bot
@dave24-73 Says:
I feel Pandora’s box has already been opened, there is no going back now.
@ivanhorvat9248 Says:
Thank you for your video, it was a joy to watch and learn.
I think that we, as a society, will find ways to handle new AI technologies, as we always did with other new technologies that had come before. For people like me (without any skills in drawing, painting, composing etc.) whole new dimension is opened. I can, for the first time, create something using prompts, similar to my experience as a high-school gamer when I would create my character in certain video games.
Thus, I look on this new technology as on sth good and sth that will allow me more freedom in expressing myself in field in which I could not had done it before due to my lack of skills.
@rsb3609 Says:
I asked for an image of a raspberry pi 5,floating in an old goldfish bowl full of mineral oil.......does not beat the real thing.
@andrewkamoha4666 Says:
1:49 "Tall fairy tale castle made from cheese."
Chris shown several ways to create image from a single prompt.
However, is there any txt2img that can fine tune in *DETAIL* ??
@janisvaskevics93 Says:
Copyright will have it's place in history, but not in future.
@DariusOutdoors Says:
I think like with the box of panda..ora once it's out it's out. There is no way in stopping ai image generation now and how do you even prove someone used your images to train an ai, if the training dataset is not public. Anyone can scrape images on their own drives. My biggest fear is what shady ai image and video generations will come out of the dark web. This has potential to destroy the internet as we know it. Countries blocking parts of the internet and so on.
@asiano3385 Says:
The only thing that I'm afraid of is that people will have nothing to live for anymore because computers will be doing everything for them.
Also I'm not a big fan of AI generated images because people who use these services are not "AI artists" but just a "AI image generator service consumers".
I like challenge. I've tried these tools a few times and it is just too simple to use so I rather learn art instead.
My only argument is (if I ignore the thing about AI learning with no effort from other people without their permission) that I like technology and I like making everething more effective. That is the reason why people came up with things like calculators in the past and now with full fledged computers. However if we make AI to do everything for us and I mean really everything then we would become useless just because doing nothing is just not useful nor fun.
Then about making a living out of it I think it should be regulated. For example generating images for fun but not selling them. Because that would be worse and more useless than a mukbang. (or not useless but just a too lazy way how to earn money... because I think that AI art is not useless)
I'm not a hater of this technology but I just have this strange feeling about thoughts like "what is the point of learning anything if AI can think for me?" And I don't want to have these thoughts. I like to learn stuff. That is the point of a true living. (at least I see it that way)
That is just my honest opinion how I see this.
But it is interesting nonetheless.
@keibohow69 Says:
copyright is a con! Why because all these people who think they came up with an original idea, could not have done it with out the knowledge that was left behind by those who came before them.
@zilog1 Says:
I'm not going to stay in the opinion on this, but I need you to be very careful with this topic, especially on social media if you run into any trouble, I'm giving you a fair warning
@drone_flyer1207 Says:
These tend to be difficult discussions to have. But I find a few thought processes that can help.
1. There are not enough "artists" in the world to create all the ideas other "non-artists" have.
2. Are we not all, "artists"?
3. Are we supporting the "art" or the "artist" when we exchange monetary energy?
4. If given enough time, is it possible for a human in the future to create an exact replica of "art" made in the past having never seen it?
I find that when you ask these questions, you begin to see that tools like AI desperately need to stay open sourced and accessible to EVERYONE.
Another thing to think about, is what does it mean to gatekeep, specifically when talking about knowledge and skills.
Further thoughts, think about the world of possibilities that are opened up to the poorest, least educated, least skillful humans in the world.
-- yes there will be abuse, refer to open sourcing and accessibility. ✌️✌️✌️
@user-jx5dn5nd2m Says:
Regarding legal issues , i will just make a bet on how cases might be be solved.
I am speaking as a software engineer with some insights in AI , and also an amateur music writer.
The way music industry has solved creators retributions fits nicely with generated IA, provided the right credits are assigned to the right people. Who / what is doing what.
- Stable diffusion and the likes are a broadcaster ( a radio station which delivers on demand music )
- Creators are people who designed the original images which the neural net was trained on.
- People writing prompts ARE NOT CREATORS. They write a PROMPT which is a query so they ask for on demand art production.
So its like asking for a music mix on demand .
Then retributions works like this:
- Stable diffusion will pay the rights to the artists , like Spotify does.
- Prompt writers subscribe to Stable Diffusion like they would with Spotify.
IF the prompt writer wish to reuses 'the mix' , has has to pay an extra. Like in France there is a difference if you by a music ( say a CD or a vinyle to be simple ) and you either plays the music for your self ( private diffusion ) or public diffusion ( you have to pay the SACEM )
So the whole model is ready.
Sorry to be hurting prompters feelings but it takes more to write a query to be an artist. The way the system is used is much closer to writer a search query on a search engine that designing a piece of art. You have to give credits to the actual artists behind the scene which is the mixer ... Its the neural network. Most of the prompter , like you Christopher are unable to:
- draw with pen and pencils the image you PROMPTED
- have about no idea of what the prompt rendering will look like ( like you said ) in the video
I think those suffice to disqualify assigning the status of artists to someone who writes a query , in natural language instead of SQL.
All artists know with a much better accuracy what they will draw and how to draw it and you don't. You are really a user with power tools.
You are actually query a large space of images which ALREADY exists , they are linear combinations of the original image and the neural net is somehow doing dimensional reduction IE compressing that into a language you can use as a prompt. That most of the people who have no actual IA insight won't probably understand but i bet top notch lawyers will discuss this point in great details. You prompt is a summary of a vector which points at this image in a high dimensional system. Its already 'there'
Finally, since a neural net computes image based on weights , its even possible i think to calculate payment owed to each artists for a generated image ( a mix ) so the retribution system may be pretty accurate.
And last thing : Artists can write prompts , but they where artists before that. Doing prompts does not make you an artist.
My 2 cents ^^
@KanielD Says:
Copyright is BS anyway. Hopefully AI can move faster than the lawsuits & people realize that IP has never been real.
@kattz753 Says:
Quite frankly, AI freaks me out. AI was one of the concerns in the big Hollywood strike. There are so many people that appear in TV and film who just work for a living. They're people in the background. Hollywood wanted to pay these people one days work, scan them and use AI to put them in films TV shows from then on. Artists are going to be out of work. Someone even created the first "new" Beatles song since the breakup in the early 70s.
There is a good side. We might be able to treat deadly diseases or solve complex problems like global warming. We might not like the solution, however if the computer decides that the problem is us and we need to be eliminated. Thats the part that freaks me out. You think that the politicians are bad? Wait until you meet our new robot overlords.
@Naa-ee7nq Says:
those are some stellar prompts, you have a future in vibe engineering
@pieterboots8566 Says:
They all want my details. So annoying.
@PGW90RU14 Says:
I totally agree with Christopher. Today's enforced learning system thinks that widely-spread data is the most correct answer. However, any innovation in any genre is a minority in its genre at the beginning. I honestly so-called "AI-generated image" is poor, boring (sorry) and toddlers' scribbles are more inspiring.
As far as I know, all AI systems require users to create accounts. The reason why is that (1) AI system wants to know what users like and (2) AI system wants to know which users have more popular answers. These mechanisms mislead what is good and as a result generated images are mediocre (for me it is definitely true).
I don't deny mediocre generated-images. At the same time, I never want to see such kind of mediocre images honestly.
@cerulyse Says:
I wouldn't worry about it it seems the EU have that protectionist beaks already in there on TV straight after I watch this video in fact
@yorkan213swd6 Says:
Are you an AI ?
@ezg8448 Says:
Skynet is coming along nicely...
@shawnlowery6246 Says:
Something about the thumbnail. It took me a day and a half to realize it was an ec video.
@manaphylv100 Says:
I find the img2img option in Stable Diffusion much, much more useful than txt2img, since I can just draw a rough draft in Paint (or take a quick photo) for it to base on. Its output images can then be fed back into the iterative process for as much fine-tuning as you need. Plus, I have much more control over the output when using img2img, whereas txt2img is mostly random chance.
The problem with straight-up txt2img is that it doesn't understand some prompts, and you are often required to use very specific terms or spellings/hyphenations for the desired effect - better have a thesaurus ready, especially if English is not your native language. An interesting example: "twintails", "twintails hair", "twin-tails hair", and "twin pigtails" can yield very different results.
LATEST COMMENTS