Review of Magic Canvas - Update

Greetings -

After ‘playing’ with this for a few days I thought I’d make a little review for anyone considering subscribing to Premium to try out the new feature. And becasue I’m just curious about how things work, thought I’d write up some results of my experiments.

So basically, each ‘style’ is formulating a result in a different way. For instance, enter ‘Fox News’ in the color or black and white ‘no art’ styles and it’s doing a mashup of realistic actual photos referencing ‘Fox News’.

But prompt the same in one of the ‘art’ styles and it’s doing a different thing. Mostly it’s taking a more ‘literal’ interpretation of the words you put in.

Honestly, the photo styles are also just plain entertaining as well…

But I’ve focussed on the Super Cute style as it’s a little different. I’ve got quite a variety of outputs from this one, from literal to more of a mashup with ‘real’ elements.

Further, the Super Cute ‘style’ can also do some pretty heavy lifting in adapting imagery and I would say it’s the Magic Canvas’ strong suit.


The heroic portrait style can manifest an image in several different ways. From just slapping a head on a fairly narrow genre of armor -

To a more comic book ‘action’ style image -

To a more artistic interpretation.

I should also include a copy of the Terms of Use, which are surprisingly short -

First off, it says, ‘due to shortcomings in the software, the generated results are at risk of including…’ Laundry list of items. I personally would not call the depiction of those items a ‘shortcoming’. If you value the diversity, equity and inclusion of all elements of life into the creative expression of art, then I would consider those unintended, and in some cases unfortunate, outputs to be the freest expression of Liberty. But mabey that’s just me…

WTF… At any rate, I haven’t seen any ‘hateful’ symbols, although I haven’t prompted any either. But I have seen, with non-associated prompts, some nudity. Including, in super cute style, some images which might be considered inappropriate. I guess I was surprised at first, but upon further reflection I don’t think you can get a ‘clean’ output, without a massive restriction in scope. At which point, what’s the point…

[Obviously can’t show that here, and logged out and back in so they would go away…]

These ‘issues’ are something that are being discussed, along with ‘copyright’ implications, for the technology as a whole. My personal opinion is, just doing a google search can expose people to far worse (real) things. So the fact that the technology (of AI) can produce something we don’t want, or be used in an illegal manner, doesn’t mean we should not allow it. And many things are not inherently ‘wrong’ they’re simply open to ones interpretation.

So now that I’ve triggered a whole bunch of people… :wink: Bottom line is, I’m most interested in seeing how the technology operates, through a variety of outputs based on inputs. However, as a creative tool it can be quite quite a resource. (AI art in general) Especially for people like me who just need a little ‘inspiration’ to get started on an idea, then I can take the ball from there and create something unique.

In closing. I don’t think I would subscribe to Premium just for the magic canvas. Not for the reasons I use my GF. But for some people, as you can see in the contest thread, it may be worth it. Just remember - if you do decide to use it, as with all things in life (as we say around the office) don’t ask the question unless you really want to know the answer.

HL

9 Likes

Love it!! I laughed at the super cute ones. :joy::joy:

2 Likes

I have moral concerns about this new feature. Where are the base images coming from? If they’re just scraped from the web then there is certainly at least one artist somewhere being ripped off for each one…

Greetings -

It’s an issue with a lot of facets, that’s for sure. But I also understand completely the response, yes but, it just doesn’t feel right.. And there were some people who first thought that photographs were stealing our soul. Oh wait, I believe that’s the very definition of Instagram.


We’re only asking it to show us who we appear to be.


And I can’t fault the technology for not liking what we see.

HL

2 Likes

It is not “stealing” Ideas from anyone, but creating an image from impressions just as any artist would. I do see very few puns, which surprises me. Fox news is perhaps unusual in that as news for, by and about foxes, would be my first impression out of context.

You can ask for the weird like a rabbit with a giraffe body and I knows what a giraffe and a rabbit are and puts the two ideas together, It does not “know” or look for someone who has done that. There are weird gaps in its knowledge. You would think that asking for a roadrunner might get either the cartoon or real version but It seems to have no good idea what a roadrunner is beyond being a weird bird.

“rabbit in giraffe’s body eating cauliflower trees”

4 Likes

I use Premium for the free use of the designs in the catalog and the dashboard features like text, etc. I have played very, very little with the Magic Canvas, mainly because I simply am not interested in it. I played with it a little and I have more trouble trying to decide what to type in and what style to use, and figure I really wouldn’t be using it for actually doing anything with it. It’s just not a feature I’ll be using.

2 Likes

It’s an interesting thought experiment. What’s the difference between an AI and a human artist who’s capable of mimicking another artist’s style? That’s allowed, as long as it’s not used in fraudulent ways.

How do artists learn their craft? They build on the experience of others. Art school packs your head full of a catalog of preexisting styles, techniques, and works. Can you truly then create something unique not based on things you’ve seen, learned, and experienced? Is there nothing new under the sun?

The main difference then is speed. It’s a threat because of speed and quality, not because it learns from others.

9 Likes

Love mouse soup! Maybe I’m just bent a bit but that’s a keeper for sure :blush:

2 Likes

Here’s my issue with AI, and it’s probably an outlier belief…How much satisfaction do we actually get from telling a machine to create something for us? (It doesn’t actually help us to grow our skills, which come through trial and error, failure, more practice, and eventual understanding and mastery, although it does help the AI to become more powerful. And this one appears to be very well done, so kudos to whoever created it.)

In the end, I’m afraid that letting an artificial intelligence do it for us will cause us to walk away feeling less empowered and creative than if we try it ourselves the old fashioned way, and actually overcome the learning curve for ourselves. That process…trying it, making mistakes, and trying again has a phenomenal growth value for us as individuals.

Having said that…I do enjoy seeing the beautiful artwork that the artificial intelligence is creating, and it is a lot better than I can do, so I probably wouldn’t mind hanging it on my wall. But it would hold much less value for me than if I created it. (Kind of like fake diamonds, YKWIM? They’re just as pretty as those that nature creates, but hold a lot less value by virtue of the fact that the natural kind is so much rarer/harder to get.)

Still, I’m sure it will be a great selling point, so don’t mind me. :smile:

8 Likes

As a follow up - regarding foreign languages - ie, non-english

In color / B&W (No Art) styles where it’s just scraping images and mashing them up, foreign languages work pretty well. As you would expect-

Proper names work as well, however, not knowing any foreign languages myself, I can’t test the veracity of its algorithm. It’s probably doing a google translate (as was I) so the results may not be exactly as expected.

The last one being open to interpretation as to whether it ‘worked’ or not… :slight_smile: This makes it fairly universal in application. ‘Boat’ in Russian for example -

The ‘art styles’ also work with non-English - “Fishing on a Boat” in French then Spanish. Although tougher ones like Russian just kept turning out non-associated images.

So that’s pretty interesting, and something not totally expected. There are limitations but, given the relatively narrow scope of its intended function, I’d say it’s capabilities are fairly broad.

HL

3 Likes

Greetings Jules -

It’s a valid concern. In a practical sense, a more likely scenario is the flooding of AI’s ‘interpretive’ visual representations. With those images being incorporated to interpret yet again. After only a few iterations it’s like inbreeding, you end up with a result which is infertile and ultimately unrecognizable from our fundamental understanding.

Some might say the exploration is a benefit as it expands our understanding of who and what we are.

But there are usually costs to that exploration we’re not aware of at the time.

But that’s what we Humans do, it’s both our strength and our weakness to explore the limits of human existence. Though usually we only find a limit after it has been crossed -

With that cost inevitably being extraordinary…

HL

This feels like it’s in the neighborhood. As with a lot of technology, this might be less about what it is doing than about who it is doing it to and who it is doing it for. Context matters. There would be less concern about driverless cars if they didn’t have to operate around frangible humans, for example.

Cory Doctorow posted a pretty good essay about this a couple days ago (on his ad-free, tracker-free site):

https://pluralistic.net/2023/02/09/ai-monkeys-paw/#bullied-schoolkids

5 Likes

You make a good point, but as with just about anything, the user should be cognizant of what they are producing. This tool opens a great possibility that someone could create something that would violate not only copyrights but ethical artistry.
I tried it out and it pulled several photos that I know were copyrighted. Altering a copyrighted picture does not eliminate the copyright law and if the alteration does not keep the intended ethical value of the original, it can be subject to lawsuits.
I would encourage everyone who uses this tool to be aware of what they are using and follow good ethical rules.

3 Likes

Are you sure about that?

If the “impression” is indistinguishable from the original, then it’s a copy. And I think we’ll see more of this problem — inadvertently or intentionally — as the technology evolves.

5 Likes
That is one of my old bailiwicks and so have a very much more complicated outlook.

#1 Diamonds are a monopolized resource that works very hard to bend your mind to their benefit and profit. They are quite brutal on the folks that do the actual labor, but if a country is overthrown, and the new masters are brutal to the workers then those are “blood diamonds” and you should not buy them.

#2 leaded glass will disperse color more than diamonds and zircons a bit less (though zircons stand up better at high temps) but nothing natural will scratch a diamond so it is easier to push the others as “fake diamonds”.

#3 Actual real diamonds can be produced that even have the same structure (unlike synthetic ruby that does not) and quite indistinguishable, but industrial diamonds take less work and are easier to sell, and gem diamonds are not worth the effort as prices have dropped. The crazy price of gold has driven gold rings to be melted down in massive numbers, and those buying have jars full of small diamonds. and falling prices of the medium normal sizes.

Diamonds are not actually “rare” but the monopoly (DeBeers) works very strongly to control the volume sold thus keeping the price high. Russia has a huge volume on hand as there are large deposits from the Russian Traps and threaten to dump them into the market more than actually doing so. This is in the back of my mind watching the current war.

As for the resulting artwork, I see it as a tool and not imagination itself. You can make jewelry with a lump of metal and a hammer but are quite limited as to where you go from there. I was often criticised for making jewelry that was cast and not made from hand tools, but in casting I had much more freedom of design.
Architectural drafters hated Autocad as it could do perfectly in hours what took them days by hand, and worse it replaced them as the skills did not transfer well.

This Ai (that is not really hard Ai) is just another tool with a different set of skills needed, and sketch artists that had to spend years getting and honing their skills are feeling the same as the old drafters.

3 Likes

Then the original was not very original. That has been an issue for a very long time. Imagine having a copyright on a computer that worked by using ones and zeros. Or as really happened having an icon of a place to throw trash that was a trashcan.

The problem of “ownership” has always been a legal one and not a logical one. If I have a photo of a brick wall and you “steal it” are you stealing from me or the guy that built that wall?

This is even more problematic…


That design was plastered on a wall several hundred years ago, but my legal liability is to the person who took the photo even though I had to do a lot of work to make it work as a relief carving which is what the original was. Where is the “copying” there? Who copied what?

2 Likes

Indeed, AI image generators have opened a huge can of worms that our copyright law (and maybe even our entire concept of copyright) just isn’t prepared to handle.

But as the research above showed, it’s going to become easier and easier to create faithful reproductions of copyrighted art, and that’s already happening to a limited degree. I think accidental copyright violations are a credible risk to users of these tools.

Personally, I love playing with AI images as art therapy, but I’d be hesitant to use them in anything commercial. Even though the risk of legal action is minuscule, I don’t want that liability, or to be in the position of inadvertently ripping off an artist who makes a living from their creations.

3 Likes

That problem has been endemic from the start and made hugely worse first by the internet as almost every piece of art is “copied” to the internet, and most can be copied from there. Watermark as you will, most of those can be bypassed. or “fixed”. Just watch Youtubes of the Ukrainian war. Even folk who are “stealing” from others will have lots of moving watermarks, and the watermark of who they have it from blurred out, and then a threat at the end that they will vigorously enforce their copyrights.

In Fashion they have tried banning copyrights, so folks are copyrighting their logo and trying to make that more important than the actual design.

It is a very deep and endemic flaw that keeps getting worse as technology evolves, but if it cracks much of the social and legal system cracks with it.

2 Likes

If you know the issue then it is not inadvertent. If you have a mouse running a steamboat and it looks like Steamboat Willie then you will know. If it resembles the African Queen only the heroes have rat faces then you will know it is not.

1 Like

Humans spit out copied art and photos even more than AI, so maybe we should be the ones who are banned.

3 Likes