Our Beliefs

We believe that as AI image generation exists longer and longer within the public eye, there are very clear, well-understood ethical criticisms about its existence. Though, like the Greek myth of Pandora's Box, it's simply too late to close the lid on AI image gen.

Though, once again like the myth of Pandora's Box, hope is still inside.

With the silly metaphors out of the way, let's discuss the issues with AI image generation, the positives of AI image generation, and how we can combat the negatives.

1. Training data & consent

Many AI image generation models have reportedly been trained on images scraped from the internet, often without clear permission from the artists, photographers, or original rights holders for those images to be used as training material.

This is one of the most well-understood critiques of AI image generation, especially when it comes to AI “art”.

Why legislation alone isn't enough

Unfortunately, we believe that it is difficult to combat this issue through legislation alone, because training datasets, image ownership, licensing status, and consent can be difficult to track and verify at scale.

The companies and research labs that trained these models may not always have a clear, public data point like: “we took this image, from this specific artist, from this website.”

While some information about image origins may exist, such as the websites or datasets involved, enforcing rights across large-scale training datasets is complex, especially as AI adoption continues to grow and become well-established.

How Silumin responds

To combat this, we'll be donating 10% of our profits to select charities, targeting institutions aiding in accessible arts education and creative mentorship.

Currently, we'll be donating to VIBE Arts, which “provides a wide variety of hands-on, high quality, accessible arts programs for children and youth aged 3-29.”

Our starting goal is $100. You can track our progress here.

2. Deepfakes, misinformation & misuse

As AI image generation becomes more accessible, deepfakes can be generated at mass scale, potentially spreading misinformation or creating non-consensual sexual content involving public figures or private individuals.

This is another incredibly unfortunate reality of AI image generation, and one we personally detest with a passion. The internet is already plagued with misinformation and unethical sexual content, and irresponsible AI use only continues fueling the fire, damaging the landscape known as the Internet.

How Silumin combats misuse

Silumin combats this through strict rules relating to deepfakes, impersonation, and non-consensual identity misuse:

• All prompts containing a public figure's name will be automatically flagged by our system, and the prompt will not be generated. Due to our privacy policy, your prompt will never be automatically published or permanently stored in our database. Prompt checks may happen locally in your browser and/or through temporary moderation systems used only to enforce our safety rules. See our privacy policy here.
• No LoRAs trained on public figures will be allowed on the website.
• Any image-to-image generation involving explicit material, non-consensual identity use, impersonation, or prompts requesting that type of content will not be allowed.

3. Child sexual abuse material (CSAM)

This is absolutely the most serious, detestable issue within the landscape of AI image generation, and one that we believe deserves clear, careful attention.

A common argument in favor of this type of generated content is that “the content is fake, therefore it has no real consequence.”

We strongly reject that argument.

Even when an image is artificially generated, sexualized depictions of children are harmful, exploitative, and unacceptable. They normalize abuse, create demand for abusive material, and can be used to harass, exploit, or endanger real people. Additionally, how do you think AI models know what naked children look like?

AI image generation models are trained on large datasets, and when those datasets are gathered at scale, there is always a risk that harmful, abusive, or improperly labeled material may be included if strict filtering and moderation are not applied. We are not accusing any specific company, researcher, or model provider of knowingly training on illegal material, but we believe platforms using AI image generation have a responsibility to take this risk seriously.

How Silumin addresses CSAM

Silumin does not allow CSAM, sexualized depictions of minors, age-ambiguous sexual content, or prompts attempting to create, imply, or disguise this type of material.

• Any prompt, image-to-image request, LoRA, or uploaded reference material involving minors in sexual, exploitative, or suggestive contexts will be blocked. See our privacy policy here.
• Accounts attempting to bypass these rules will be permanently banned.

This is not an area where we believe there is room for compromise.