Tarisio images and licensing model

[Riccardo964]

Good morning,

I have searched this forum and I have found no information whatsoever, that's why I decided to create a new topic. Sorry beforehand if this has already been discussed.

So, I have registered to Tarisio's platform for browsing the Cozio archive, and I was awarded one full month access. They state:

Quote

You now have a period of free browsing in the Cozio archive. This gives you access to information about thousands of fine instruments and bows, including photographs, provenance and historical price information.

Then, afterwards, they mention:

Quote

After your free browsing period has ended, you will be invited to subscribe to the archive for $100.

My questions are, if anyone could help:

1. During the first month, all images I download, are those mine to re-use as I wish?
2. Should I acknowledge Tarisio or add as a source whenever I use an image, let's say, in a blog?
3. What is the licensing model that Tarisio uses after all?

Kind regards,
Riccardo964

 

[GeorgeH]

Maybe ask them.

[Riccardo964]

32 minutes ago, GeorgeH said:

Maybe ask them.

From my experience, when it's not explicitly listed (I actually wish it to be a straightforward CC0), then it's *aggressively* proprietary and they "own" it by all means and will take all precautions to protect it.

I might send them an e-mail.

Best,

Riccardo

[deans]

Just curious, how do you want to "re-use" them? I suspect that just having them downloaded on your computer for your own viewing would not be a problem, and hard to imagine they could do anything about it. 

But ask anyway.

[Riccardo964]

56 minutes ago, deans said:

how do you want to "re-use" them?

Well, I would use as input for running some image processing algorithms, which might result in interesting output, which I could "publish", however, if the same input is protected by licensing, I agree with you that this would only serve to my "self-consumption", without any publication whatsoever.

I shall investigate this further - I thought someone else here might have encountered the same issue and would like to share experiences.

Best,

Riccardo964

[Urban Luthier]

45 minutes ago, Riccardo964 said:

Well, I would use as input for running some image processing algorithms, which might result in interesting output, which I could "publish", however, if the same input is protected by licensing, I agree with you that this would only serve to my "self-consumption", without any publication whatsoever.

I shall investigate this further - I thought someone else here might have encountered the same issue and would like to share experiences.

Best,

Riccardo964

I agree, i'd be very careful and check the Tarisio EULA. A lot is changing fast with the introduction of so called AI technology. I suspect IP owners will be actively updating EUALs to protect their IP as the next wave of pc and phones are released.

As far as I'm aware, Tarisio images are a combination of photos they've taken themselves and licensed from others. My guess is their EULA doesn't permit the use of their images (even for educational or research projects) without their prior consent... Let alone a  hypothetical use case of training a LLM with their photos via say NVIDIA AI workbench. The latter would require massive GPU compute beyond the reach of most anyway! 

 

[Riccardo964]

1 hour ago, Urban Luthier said:

My guess is their EULA doesn't permit the use of their images

Yes, I also believe this will be the case.

However, I've found this on https://tarisio.com/cozio-archive/about-cozio/:

Quote

(The Cozio Archive) It was founded by Philip E. Margolis in 2003 to make information about fine instruments available to the wider public.

And (my emphasis):

Quote

Founded in 2003 as a public registry of iconography, provenance and pricing information for historical stringed instruments

Afterwards, they mention "additional" photos that would require requesting to them:

Quote

 

Additional images in the archive

Some Cozio images are not available publicly. These can sometimes, subject to copyright, be made available privately to users with specific research projects. Please submit details of your request to cozio@tarisio.com.

 

Sadly, I can't find any EULA or licensing information whatsoever. I'll keep looking into this. I'll send an email and shall keep us all posted.

Best

[martin swan]

This is a very difficult area which we are all still learning how to navigate.

However it seems clear to me that verty soon we will see some massive lawsuits brought against. AI companies for theft of intellectual property.

iIf your image processing algorithm proved to have any commercial application, then Cozio would surely expect you to pay them for all the raw data …

[Johnny Sun]

I guess you can download as many photos as you want and re-use them privately. 

If you post any photos from Cozio in a forum or blog, you will have to get the permission from Tarisio, and permission from the owner of the photos, if they don't belong to Tarisio. Too complicated, I would avoid doing so.

However I recommend you to subscribe to Cozio. I find it useful and interesting. 

 

 

 

 

[Guido]

These are interesting questions.

Maybe somewhat related: We take it for granted that Tarsio has rights to the photos and info in the archive, especially where they took the photos themselves. But wouldn't or shouldn't the current owner of an instrument have a say?

I.e. could a current owner of an instrument request all info (or at least photos) to be removed from public access?

[IBK]

The Cozio Archive was created by a Mr. Margolis in Switzerland.   Some photos and information were at that time submitted by owners. Two of my violins are listed in the archive and were submitted by me when Margolis owned the archive.  When Tarisio acquired the Cozio Archive, I had to give specific permission to publicly reveal that the instruments were owned by me.  I took the photos that are included. The question who owns the intelect;ual property rights. 

[Geoffrey N]

6 hours ago, Guido said:

These are interesting questions.

Maybe somewhat related: We take it for granted that Tarsio has rights to the photos and info in the archive, especially where they took the photos themselves. But wouldn't or shouldn't the current owner of an instrument have a say?

I.e. could a current owner of an instrument request all info (or at least photos) to be removed from public access?

I'm pretty sure that, at least in the United States, if you take a photograph of a violin you own, you now own the copyright to that photograph unless you choose to assign it to someone else later. Even if you sell that violin, you still own the rights to the photograph you took and so unless you sell those rights with the violin, you have the right to do as you wish with the photograph and the new owner of the violin doesn't have a say.

[Riccardo964]

What I don't understand is this: why Google, Microsoft (OpenAI), Facebook, Twitter, and the likes can download (and use as training sets for AI/ML) a lot of protected data (as mentioned, a lot of venues are actively suing them for doing just that, others are making agreements, ie, selling their user's data altogether), and I, a simple Internet user, cannot?

 

[Geoffrey N]

Insert "I'm not a copyright attorney, this is my personal view not representing my employer, and these things can get tricky pretty quickly" caveats, but I have spent a fair amount of time over the years trying to educate myself in this area so happy to share a perspective.

From what I understand, there are tons of open issues with using other people's copyrighted images for AI/ML training purposes and it will take years for all the issues to settle out in the courts. A lot of the copyright system is about protecting the copyright holder from others reproducing an image/recording without permission (with some exceptions for educational use). Copyright generally doesn't protect you against someone being inspired by your image and doing something similar (as long as it's not sufficiently close to being the same as the original). We've seen high-profile lawsuits recently in this area, including against the pop star Ed Sheeran who was accused of copying the chord progression, rhythm, other elements of Marvin Gaye's "Let's Get It On" in one of his songs, for example. That case went all the way to jury trial and Sheeran was acquitted after the jury found his work to not be infringing. In most of the creative arts people take inspiration from and learn from the artists who preceded them, by becoming familiar with their work and creative process.

So a key question is, if an AI/ML training model that is trained on a large corpus of existing art in a particular area, is that the analog to a human being exposed to the past history of a particular creative art such when you ask it to "create a picture of X in the style of artist Y" (whose works are still under copyright)? If the training model's output constitutes a legitimate creative process the the output should not be a copyright infringement as long as it as sufficiently different from the originals it was trained on -- in fact the creator of the model might even be able to claim copyright on the model's output as being their creative output with the model being their tool, that's a whole other question. But if constructing and then using an AI/ML model isn't deemed creative but merely a type of sophisticated copying, then the owners of the model might be infringing the copyright of the input images. Some copyright holders are trying to get in front of these questions by licensing their art with provisions that say that it cannot be used to train AI/ML models, because so much is unknown and this may help them in the future.

As far as I can see, some people/companies building AI/ML models are being extremely careful and conservative, whereas others are plunging in and taking the risk of potentially finding out that there's a problem later. It's a pretty big mess because this is such a novel area.
 

[martin swan]

Very interesting summary, thanks

[GeorgeH]

8 hours ago, Geoffrey N said:

in fact the creator of the model might even be able to claim copyright on the model's output as being their creative output with the model being their tool, that's a whole other question.

AI output cannot be copyrighted because only works created by a human being can be copyrighted under current law.

[LCF]

2 hours ago, GeorgeH said:

AI output cannot be copyrighted because only works created by a human being can be copyrighted under current law.

That sounds like yet another quagmire. For instance if I give instructions or programming commands and gestures  to a software tool with some design goal in mind I think of it as my own work, within the bounds of the usual caveats. One implication of what you're suggesting might be that autocorrect invalidates the copyright of an author. 

[AaronS76]

On 8/24/2024 at 4:15 AM, Riccardo964 said:

Well, I would use as input for running some image processing algorithms, which might result in interesting output, which I could "publish", however, if the same input is protected by licensing, I agree with you that this would only serve to my "self-consumption", without any publication whatsoever.

I shall investigate this further - I thought someone else here might have encountered the same issue and would like to share experiences.

Best,

Riccardo964

I don’t do image process modelling but other forms of modelling as my day job. If this was a project I was setting up then those images would be considered background IP and the images could only be used with permission of the owner of that IP. That IP would be included in an IP register that was stored on a secure server in line with any legal requirements. There would be a data sharing agreement between parties to formalise the use of the IP. As an owner of background IP they may or may not be entitled to IP in the developed product. 

[Riccardo964]

14 hours ago, Geoffrey N said:

As far as I can see, some people/companies building AI/ML models are being extremely careful and conservative, whereas others are plunging in and taking the risk of potentially finding out that there's a problem later. It's a pretty big mess because this is such a novel area.

All this copyright made me remember about a Noam Chomsky quote that discussed AI and Language Models (my emphasis):

Quote

(LLM is) plagiarism software because it doesn’t create anything, but copies existing works of existing artists modifying them enough to escape copyright laws."

Here is the full article's link (paywalled, of course ): https://www.nytimes.com/2023/03/08/opinion/noam-chomsky-chatgpt-ai.html dated March 8th, 2023.

-------------

A further "investigation" into the image's metadata, informed this:

xmpmeta:  
RDF:  
Description:  
UsageTerms:  
Alt:  
li: NO RIGHTS GRANTED
DerivedFrom: None
rights:  
Alt:  
li: © TARISIO
creator:  
Seq:  
li: Daniel Lane
CreatorContactInfo: None

So, the only information there is that it has copyright and that the creator was Daniel Lane.

[HoGo]

The big problem is that the AI companies market their product like it was a living creature, not just another SW tool.

For me these AI bots and models (whatever thay try to call them) are still just SW tools and if you buy a license to use SW tool then whatever you create using that tool is your result and not of the company. There has been AI parts implemented within commonly used image/sound/video manipulating SW for decades (starting with OCR mnay decades ago) and noone claims that edited pic is no longer yours.

I'd put it to analogy with CAD/CNC. CAD/CNC producers cannot claim any IP on your product that you designed/produced suing their tools. Before CNC only very skilled workers could get those resuts, now many amateurs can produce things unimaginable before. The AI is similar in that it for example allows a complete novice/hack write code or "create" illustration or manipulate images etc.

The AI companies of course see that in this case to these tools do produce results with absolutely minimal effort/skill on side of the user so they want to profit on that. The big companies don't allways follow the rules and try to bend them instead if there is vision of (huge) profit. I don't think there is kinda grey zone in law regarding AI, more like missing interpretation of existing laws to current state of reality. I think we can expect big lawsuits about intellectual property of the source database used in the learning phases of AI that will settle these questions.

Bask to OP: I think fair use is what could appply to the original questions.

read more here: https://guides.nyu.edu/fairuse

 

[Jeny Mahon]

On 8/24/2024 at 8:55 PM, Riccardo964 said:

All this copyright made me remember about a Noam Chomsky quote that discussed AI and Language Models (my emphasis):

Here is the full article's link (paywalled, of course ): https://www.nytimes.com/2023/03/08/opinion/noam-chomsky-chatgpt-ai.html dated March 8th, 2023.

-------------

A further "investigation" into the image's metadata, informed this:

xmpmeta:  
RDF:  
Description:  
UsageTerms:  
Alt:  
li: NO RIGHTS GRANTED
DerivedFrom: None
rights:  
Alt:  
li: © TARISIO
creator:  
Seq:  
li: Daniel Lane
CreatorContactInfo: None

So, the only information there is that it has copyright and that the creator was Daniel Lane.

Here's the article, I have a subscription and can give 10 a month

https://www.nytimes.com/2023/03/08/opinion/noam-chomsky-chatgpt-ai.html?unlocked_article_code=1.GE4.Qmcs.plHYjl7Y881B&smid=url-share

Also, I wasn't aware that photos on the Cozio archive were download-able? Right-clicking gets me nowhere!  I just assumed that was because they were protected.  If I really want an image I just take a screen shot, I only use them for my own reference.  

[Riccardo964]

1 hour ago, Jeny Mahon said:

I wasn't aware that photos on the Cozio archive were download-able

Perhaps they aren't, indeed, one could alternatively use other programmatic approaches for collecting photos. Also, one could always Print-Screen anything showing on monitors, however, this is cumbersome when there are about 21k photos in a repository.

PS: thanks for the link on nytimes!

[HoGo]

Everything you view IS actually downloaded into your computer/phone cache. The web  creators just make it more annoying to get to the source file by using the library scripts that don't react to right click. Each web browses has the tools to show all contents that is part of the web you're wiewing including all pictures at all resolutions offered in the view. No need to printscreen and assemble pics needed.

[Riccardo964]

After a lot of back-and-forth with Tarisio's Team, explaining what I would like to do, etc, they didn't budge and stated:

Quote

Unfortunately, we do not license our photos nor do we give use permission. Thank you for your understanding.

And also:

Quote

Thank you for your reply. Unfortunately, we cannot grant licensing for the photos.

Which is basic the same response - whatever - this venue is dead.

 

So, I went to ChatGPT to ask things (I'm only highlighting a few things, the full "dialogue" is here, for anyone to check, if you have interest - you could also ask it - https://chatgpt.com/share/8d4fccbb-c5fd-486f-b5f6-5691c73f17b6), for instance, what was exactly fair use and gray areas.

Quote

Practical Examples of Fair Use and When It Doesn’t Apply

• When Fair Use Applies:

  • Quoting a few lines from a book in a book review.
  • Using a screenshot of a movie for critical commentary in a blog post.
  • Creating a meme that alters the original image significantly in a humorous or critical way.

• When Fair Use Likely Does Not Apply:

  • Using a copyrighted photo in a blog post without alteration just to make the post look nicer.
  • Downloading an image and using it as-is for commercial purposes, like in marketing materials or as a blog feature image.

I also asked why they can get copyrighted images and I, a simple user, couldn't.

Question: "why openai can use copyrighted images to train LLM algorithms and I as a simple user cannot?"

Answer:

Quote

One of the primary justifications for using copyrighted content in AI training is the legal concept of fair use, particularly in jurisdictions like the United States. Fair use allows for limited use of copyrighted material without permission from the rights holders, under certain conditions, particularly if the use is considered transformative and does not harm the market for the original work.

• Transformative Use: Training an AI model is often considered transformative because it does not directly reproduce or distribute the original works in a consumable form. Instead, the content is used to learn patterns, language structures, or visual data, creating something fundamentally new (the AI's capabilities), rather than repurposing the original content for direct consumption.

• Purpose and Nature: AI training is generally seen as non-commercial or research-oriented, especially if the training process itself does not involve selling the copyrighted works or directly competing with the rights holder’s market.

It's basically saying that their models "transform" images altogether, so they can.

The "dialogue" is more nuanced than that, but that's the take away, in general.

IF I can manage to "transform" the images to exhaustion, then PERHAPS I won't be liable. But I have to take my chances.

AND, I bet Tarisio has much more money than I to pursue legal action, so I guess I'll do "my research" for my amusement only, regrettably.

 

Thanks for reading thus far,

Best,

Riccardo964

 

[bick]

1 hour ago, Riccardo964 said:

It's basically saying that their models "transform" images altogether, so they can.

This is a hot topic in the industry. Using copyrighted material for AI training is a grey area currently as others have stated.

https://www.reuters.com/legal/litigation/ai-companies-lose-bid-dismiss-parts-visual-artists-copyright-case-2024-08-13/

I wonder if you could establish a contract to use the images for your purpose, as money is likely the big issue.

I am not a digital copyright lawyer. But this guy is. This case discusses using an actors voice to generate AI voice files of a specific persons voice, eliminating them from the revenue stream of that recording in the process. There is a similar case where someone uses input is used for "research", but was actually used for-profit. Not knowing your specific case, (in my opinion) if there is money on the line, companies will do everything they can to get their "cut", just as a regular person would do the same if their art was being stolen, as in this case I linked.