The reason this sentence exists is because Adobe wants to create the impression among readers that it owns the output and that it's Adobe's choice how creators use those images. But under current copyright interpretation, Adobe doesn't own those images. So it's nice that it's giving permission, but that's not Adobe's permission to give -- so thanks but also heck off Adobe, nobody needed to ask you for permission in the first place. You can use any AI image commercially because AI images are not under copyright.
Of course, Adobe would love to have a world where most art is generated algorithmically and Adobe is in charge of deciding how that art gets used and what gets generated because it controls the tool. So it's in Adobe's best interest to pretend that it's granting artists a permissive right, rather than recognizing that it doesn't have any real legal argument to make that artwork generated through Firefly is owned by Adobe (or by anyone for that matter).
And that's good! It's not anti-AI to say that, because what you have to realize is that what companies want from AI image generation is a model where every single artist goes through them in order to build or generate anything. They want a model where creative tools are a service, for the same reason why Adobe wants its tools to all be subscription based. No SaaS company is getting into generative AI with the goal of increasing accessibility of art. They are (Adobe especially) interested in closing down that accessibility. They are all drooling at the opportunity to turn your workflow into a SaaS business that can only be run on extremely expensive hardware clusters.
So yes, the denial of copyright for AI-generated images does make it trickier to monetize those images, but denying that copyright has the much more important effect of making harder for these companies to lock out competitors and build services where they control/monopolize an entire creative market. You can still use AI during a creative process and end up with a thing that can be copyrighted. But Adobe can't release a tool and later on start to argue that nobody else can train competing generators using that tool, or that the tool can only be used in a particular way, or that everything the tool can generate is owned by Adobe. That matters.
It means that competitors can use Adobe Firefly output to train their own models (including locally run models like stable diffusion). It means that there's a limiting factor in place that keeps Adobe from making lazy grabs to assert ownership over large numbers of images. It means that you can pull images generated by Firefly into other pipelines without asking Adobe permission.
You can see the same thing playing out with ChatGPT. OpenAI's TOS states that you're not allowed to use OpenAI to help build something that competes with OpenAI. That's going largely unquestioned, but my strong suspicion is that it's only a TOS violation to break that rule, because again, OpenAI does not own the copyright on anything that GPT generates. So if you're not signing that EULA, it's not clear to me that OpenAI has any legal right at all to restrict you from using output that you find online as training data. As far as I can tell, current copyright consensus in the US is that the text that comes out of ChatGPT is public domain. But that's not what OpenAI wants, because if anyone can build anything using ChatGPT's output, then how is OpenAI going to build a moat around their service to block competitors? How are they going to eventually turn profitable by closing off access and raising prices once people start to rely on their service? So just like Adobe, they stick the language in and hope nobody calls them out on it.