Participant

Restricting generation and giving errors

Forum|Forum|9 days ago
July 21, 2026
2 replies
12 views

I have to be honest and say that Firefly is a very useless tool, because a) so many innocuous words seem to be banned and b) we’re not even allowed to see what this list is. Furthermore, I've never been able to generate a decent, realistic or graphic novel-esque image through any of the Fireflys, from 3-5. I generally generate my images through the Gemini models. They’re much less restrictive, don't treat us like children, and the results are always better. Even my ChatGPT membership produces better images than the Firefly models, and without so many restrictions. Currently tried generating a character who was slightly overweight, a bit chubby, and the "overweight” and “chubby” words seem to be the offending ones. I mean, people are overweight and chubby - I certainly am. Any chance the Adobe Firefly people will eventually develop a version of this app for grown-ups?

A

alan26gh

Participant

I agree with your sentiments. In general, Chat GPT is better than Firefly, but Gemini is ahead of both in responding to prompts The key is English grammar. Alternate words are often necessary to allow AI to generate what you would like, and even then it’s sometimes a hit or a miss. The problem is obviously the limit/bar that has been set as companies are probably anxious about offending people…blame the Woke!!

daniellei4510

Community Expert

They aren’t anxious about offending people. They’re anxious about being taken to court. I’ll admit Firefly does seem to have more restrictions than many other models, but Gemini can be over-cautious as well. Gemini refused to generate “A child screaming as he rides down a slide.” and “a young boy being chased by a lion,” since it “does not produce images that depict children in danger.” Or something to that effect. Another one Gemini refused to generate was a “a woman emerges from a flower” (I think that’s what I prompted), but Gemini refused on the grounds that it depicted women in a subordinate manner. Again, something to that effect.

The most relaxed model in my experience is Midjourney. I haven’t used it in a while, but even if the AI interpreted a prompt as questionable, one could “appeal” the decision and apparently send the prompt to a more discerning AI model (or perhaps a moderator?) and the prompt was usually accepted upon review.

Adobe Community Expert | If you aren't submitting your assets in sRGB, you probably didn't read the rules.

Sign up

To post, reply, or follow discussions, please sign in with your Adobe ID.

Sign in to Adobe Community

To post, reply, or follow discussions, please sign in with your Adobe ID.