Training Data and Choosing Ethically
Most large image models were trained on web-scraped material, including a great deal of work by living artists who neither consented nor were paid. This is the field's central ethical problem, it is the subject of most of the 70-plus active lawsuits, and if you make your living from art it is not somebody else's issue.
Where the litigation actually sits: essentially all of it targets model developers, not the people using the tools. As a working artist you are not the defendant — but the outcomes will shape what tools exist and on what terms, and at least one significant appellate ruling is expected soon.
What you can actually control:
- Prefer models with cleaner provenance where the work warrants it. Adobe Firefly (licensed and public-domain training data, with indemnification) and ElevenLabs Music (licensed catalogue from the start) are the clearest current examples.
- Don't prompt living artists by name. It is unreliable, several platforms restrict it, and it is the specific practice that damaged the relationship between this technology and the art community. Describe the mechanics of a style instead — as covered earlier, it works better anyway.
- Train on your own work, not other people's. Custom style training is offered by several platforms; feeding it your own portfolio produces a model that is yours in every sense.
- Be honest about what you did. Passing off generated work as hand-made damages you far more than disclosure ever will, and it is increasingly detectable.
On the anxiety. These tools are very good at producing competent images and consistently poor at knowing which image is worth making. Taste, judgement, art direction, knowing what a client actually needs, and knowing when something is finished — none of that is automated, and all of it is what you were being paid for. The artists doing well with these tools are using them to move faster through the parts that were never the interesting bit.