Soften AI security filters

Forum|Forum|1 day ago
September 21, 2026
3 replies
27 views

I’m an architectural visualiser.

I often use the generative AI feature to bring realism to CGI’s. I use it to replace people’s faces/hair, plants, and so on. When it works, it’s superb.

It’s not uncommon for me to be working on a swimming pool, or spa shot, where the security filters kick in and prevent me from using the AI in these scenarios (as if I’m trying to make pornography or something). The security literally prevents me from using the tool as it was designed to be used.

There have also been cases with planting schemes where security features have inexplicably kicked in also - you can’t prompt for a hebe rakaiensis, for example?

Gord@APL

Inspiring

Rather than Latin names, try describing the shrub, eg. “a compact, dome-shaped evergreen shrub with bright green foliage” (adapted from a description on the Royal Horticultural Society website). May not be exact, but may be close enough for your purposes.

c.pfaffenbichler

Community Expert

Aside from @Sameer K ’s comments please consider that Adobe works for a lot of markets and thus has a lot of sensibilities (of the legal kind) to accommodate …

Could you post examples that provided stumbling blocks for you?

What is »hebe rakaiensis« exactly – a plant species or a gardening term or …? Are you confident it is a common enough term/concept that it should be meaningfully represented in Firefly’s training material? Do the partner models fail on the term, too, or is it immediately flagged as offensive/violating guidelines/…?