Thank you for using our service PIXTA.
Pursuant to the Article 2 of PIXTA Terms of Use,
we hereby notify you of updates of our Terms of Use
which will take effect on August 29, 2023.
The updates this time mainly include those to cope with environmental
or legislative changes in Japan related to the Content such as compliance
for the Content with the drones, our policy for generative AI,
and compliance for the Qualified Invoice system in Japan
commencing in October 2023.
1.What will be mainly updated in our Terms of Use:
What will be mainly updated in our Terms of Use are as follows:
Terms of Use – License Agreement
Items xvi (newly added) of Paragraph 13 of Article 16
For the purpose of proper management of the Content license in our service,
we have added to the prohibited uses of the Content “Use for AI learning purpose
including any generative AI models”.
As we have mentioned, any users who are interested in the license
for the machine learning purpose should contact us by the following form separately:
https://www.pixtastock.com/contacts
Terms of Use – Contributor Agreement
Paragraphs 2 of Article 23
Paragraphs 12 (newly added) of Article 23
Paragraphs 4 of Article 24
item vii (newly added ) of Paragraphs 5 of Article 24
Paragraphs 4 of Article 25
As we expect that needs of the Content for machine learning purposes
will increase more and more, we would like to focus on the area
in order to enhance the profit of the Contributor Members.
The updates this time are intended that we explain our policy
for the machine learning so that the Contributor Member could understand
especially based on the recent discussions on generative AI.
You can refer to the following FAQ in more details about our handling the Content
for machine learning purposes.
Also we have updated our Terms of Use by improving each provision,
or each term, and adding explanations etc.,
aiming to make the Terms of Use more understandable for all the users.
You can check the updated Terms of Use in the following URLs.
https://pixta.co.jp/images/pdf/terms_en_20230829.pdf
We will always try our best in order to make progress and further improve our service
so that more and more people can use it comfortably.
2.FAQ
Q. How is the “machine learning training data” used?
Is the Content such as photographs, illustrations or footage distributed in PIXTA ever used as the training data for image generative AI?
A. “Machine learning” is a data analysis technology that first ingests huge amounts of data into a computer which learns certain regularity inherent therein, and that enables prediction, judgment, or classification etc. for given unknown data by applying learning results. Today, “machine learning” is used in various situations or applications in the forms of image recognition, phonetic recognition or other forms. The recently much-talked about “image generative AI” is one example of machine learning.
We, PIXTA Inc., work on marketing the photographs as the dataset for machine learning, assuming that they may be used as training data mainly for image recognition such as face recognition and object detection.
Examples of image recognition or pattern recognition, in general, may include the use of AI to detect dangerous movements by humans in any places such as plants etc. for safety management, or AI to inspect products or goods, or AI cash registers to automatically recognize products and their prices in a shopping cart. The need is increasing more and more to use as training data for such photographs as available in PIXTA which depict various scenes of ordinary or business or more, and also safely usable because the attribution who uploaded the photographs is apparent.
In case of training data for image recognition models, the photographs are analyzed and, in principle, they are not displayed like existing usages such as advertisements. For example, the programs of AI cash registers must have learned and analyzed a lot of data of product images, however, usually they would not have to display any of such learned training data for the customers at AI cash registers.
Therefore, considering that such use as the training data does not impair the interests of both contributors and models, and that, as mentioned above, the need and sales for machine learning training data will increase, which should result in the contributors’ interests, we find datasets for machine learning worth appealing to the market. For your information, the terms and conditions for us to grant a license of dataset for machine learning prohibit, in principle, “to include the dataset in any products or deliverables from machine learning or display for the use” or “to use as the training data for any image generation model”.
As AI and machine learning develop, usage as the training data should be diversified beyond application in image recognition models.
We believe that it should remain a mainstream for a certain time to license as training data for image recognition models, however, considering the rapid development of generative AI in recent years, the co-existence with generative AI should be inevitable. Also we should keep our eyes on the trend of laws and regulations etc. for generative AI to be enacted in the near future.
If there is any concern that it may affect the contributors’ interests when the Content is used as training data for image generative AI, we will take utmost care to deal with such cases, and we will work on with transparency to explain to the contributor members whenever we consider necessary.
Q. Are the illustrations included in the training dataset for machine learning?
A. We, PIXTA Inc., work on marketing the photographs as the dataset for machine learning, assuming that they may be used as training data mainly for image recognition such as face recognition and object detection. In nature of image recognition models, we have never granted a license to use the illustrations as the training dataset. However, as AI and machine learning develop, usage as the training data should be diversified, and the need for illustrations should be expected to arise for the future.
We believe that it should remain a mainstream for a certain time to license as training data for image recognition models, however, considering the rapid development of generative AI in recent years, the co-existence with generative AI should be inevitable. Also we should keep our eyes on the trend of laws and regulations etc. for generative AI to be enacted in the near future.
If there is any concern that it may affect the contributors’ interests when the Content is used as training data for image generative AI, we will take utmost care to deal with such cases, and we will work on with transparency to explain to the contributor members whenever we consider necessary.