My team is planning to use Dall-e and create a web app.
Our need is that, user should not be able to use inappropriate content (e.g. adult, religious, hateful) while generating an image. How can we control this?
Does Dall-e use copyrighted images while generating image?
You can use Dall-e now via the API for experimentation. Some have made the observation that the API image generator is both less expensive and of lower quality than the consumer version at labs.openai.com. The API may be getting a much improved version next month.
The API model already has a rather strict filter that is oriented towards denying objectionable images, such as nudity or violence. You can also ensure that policy-violating input prompts are not sent to it by using the moderations endpoint to check the phrasing (although this is more to block language models). Religious imagery may not be objectionable to either filter, so you would have to analyze the prompts yourself if you wish to oppress people’s religious freedom.
The training data sources are not documented. It is proprietary data. We can see in other products that OpenAI doesn’t deny themselves use of everything that is still under copyright since 1927 due to terrible IP laws in the US, but the image maker is only “inspired” by those.
The images and other content generated by AI don’t have the status of “natural person” required to hold copyright, so they are instantly public domain. OpenAI gives you the right to use and “own” their generations to your input.
Hi and welcome to the Developer Forum!
Adult themes are not allowed with OpenAI products unfortunately.
For Question 1
- Any adult and hateful content, there were none or at least from my experience, you will not be able to even produce anything resembling one. For religious one, I don’t know… unless you specify it, I don’t think the AI will produce religious content. Dall-E3 is totally safe and you don’t need to control anything.
2.As what the answer above said, there is no documentation regarding their image sources.
There is a Moderation endpoint that you can send over your prompt and receive back if it violates OpenAI policies or not I think. So you could filter out an inappropriate request.
Moderation - OpenAI API