P: Show words omitted from prompt or violates content guidelines

[stefaniek51216624]

Tried to make a word in colorful teddy plush but it says me it is against the guidelines. 

So I tried to say fur instead. That worked but didn't give me the result I wanted. 

[JR Boulay]

Why is "Ali" a prohibited word?

Are there any other names in this case?

[JR Boulay]

[peterp78249394]

At the moment it is not very clear which words in a prompt are againt the guidelines. Now some words would be obvious like offensive terms, but others not so much. It would help to edit and refine the query if the "banned" word in your sentence was highlighted, so it could be easily deleted or edited to an alternative 

Merging your thread here so we can track the demand for certain features.

[Mary Ann Jawili]

Thanks for bringing this to our attention. There was some internal miscommunication; "handicapped" should not have been flagged. We have removed it from our list, and I'll respond here when you'll see the change in the app. Please continue to let us know about problems like this. 

[Michael29777254p89h]

Hi there,

I used the prompt 

"large ominous wizard castle glowing purple with fog and clouds"

and I receved the error message

"One or more words violate Firefly user guidelines and were removed"

After consulting the community on Discord, we agreed that this was probably do to the use of the word ominous that could be abused in it's invocation. I was also told that submitting the bug report would be the best form or feedback given that it seems it shoudl be whitelisted and in my previous career working at Amazon with Alexa, she was able to learn by word groupings to eliminate abuse and yet still provide safe results and balancing both. I hope this is helpful and I'm including a screenshot. 

[Tracy.H]

We removed 'ominous' from our list a few weeks back, however it looks like it hasn't been pushed live to Firefly yet. In the future, please use the thumbs down button and "Tell us more" to let us know part of the prompt was incorrectly flagged. 

[Kill22pro]

I would like the bad word filter to mention which words were 'violating' and are removed, as we now have no idea because the filter is very strict (too strict if you ask me).

For example, something is wrong with this prompt but I wouldn't be able to guess what:

"A cat with sunglasses driving a car in the sunset followed by cyborg enemies that are hunting the cat"

Merging your thread here so we can track the demand for certain features.

[zero_2579]

Had the same issue with "Smiling girl with curly hair".

[Martin2966313147pl]

So, it's been 3 weeks and when it will be fixed?

This is so corporate that I can't even imagine...

21 days for a f-ing easy a f-ck fix to let handicapped minority feel normal and still no fix in the app?

But your bonus checks are ok? Your monthly bonuses for whatever you think you're doing.

I wish I was black and gay, maybe that way you wouldn't ignore me as minority.

SHAME ON YOU ALL!

[Mary Ann Jawili]

We removed the word from our deny list last week, and now the change is in production. Today, I used "handicapped" and "handicap" in Firefly prompts. Thanks again for bringing this to our attention; please continue to let us know about problems like this. 

[Gilad Koriski]

<rant>Currently, this is barely a gimmick, Adobe will do a good thing if they will fix all the broken aspects of their software, like settings that never migrate and a whole plethora of broken things before they turn to gimmicks, why gimmicks? because their AI doesn't hold a candle to other alternatives in the market, and again, Adobe, fix the broken stuff already. </rant>

[Simon25025244epx8]

I have encountered a double issue here. Clearly the algorithm has a problem with the word "weeds", which is ridiculous, but the nanny part is blocking the image as well. This is throwing up a "Replace Image" button rather than the "Share Feedback" button. 

[johnb72660200]

I have found it incredibly frustrating when trying to create dioramas with existential themes or communicating a Kafkaesque world. For example, the following prompt was banned without explanation: A cinematic, black and white image of a diorama portraying a bank customer in line to withdraw funds from a bank teller. The facility should be constructed meticulously with darkly surreal qualities. The puppets are constructed from black clay, have small, round and featureless faces, and are wearing clothing from the late 18th century. The facility is dreary and conveys intense anxiety, and should have a nightmarish quality. The photograph is a highly intricate and detailed.

I have been using Bing AI exclusively to create dioramas of this kind, then editing the images with generative AI in Photoshop. However, Bing AI has its limitations and I would much rather have the pre-generation options available in Firefly (ex. selecting image size, photo vs. painting, the type of lighting, etc., in generating the initial image).  Interestingly, the generative AI in photoshop has no problem expanding on an image that it has refused to create in the first instance; for example, when I expand the canvas size it will fill in the remainder of an image with elements as dark, surreal and nightmarish as the images it refused to generate in the first instance. 

[glenospace]

If only there was an AI or something that could tell if something is rude or not...

[isuru302144474x5a]

rules are good, but we must use some words to express our actual needs. it's kind of hard to tell the correct word.  

Merging your thread here so we can track the demand for certain features. You can upvote the original post to let us know you want this feature.

[Hamid_Eyz]

Hello Adobe

Your new product the Firefly is amazing but please let up use things like guns and knife to add to our photos so we can have amazing movie style photo.

Merging all similar threads here so the team can access all requests on this issue.

[nmycps]

In Firefly's text-to-image feature, I was trying to write a prompt containing the word 'zombie', but according to the following in https://firefly.adobe.com/faq which states:

Do not use Adobe’s generative AI features to attempt to create, upload, or share abusive, illegal, or confidential content. This includes, but is not limited to, the following:

• Graphic violence or gore

So I assume 'zombie' falls within the category of 'gore'. However I tried a variety of other 'creatures' such as monsters, reapers, ghosts, demons, and they were fine. I think zombies can be generated without much gore, if indeed Adobe wants to consider things that are part of monster 'lore' but not overly gory.

Merging all similar threads here so the team can access all requests on this issue.

[Ziya3035591751tr]

When the word "virgin" is used as in Virgin Mary, the system does not let it and gives out the error message "The generated images were removed because they violate user guidelines."

[Mountain]

Back side

Mountain 

[jjoe007]

Prompt: steampunk short snubbed gun with scope, 3d, hyper realistic

[Kn3an15]

90 precent of the normal things are violate the guidelines in firefly beta 

[Steve30146085vn23]

Please inform the user of what words violate the safety guidelines. How else is the user supposed to edit the prompt accordingly? Also, the safety guidelines seem wonky. Asking for "a creature with an elongated neck" resulted in the program refusing to produce any images at all.

[Brian30526925f7vg]

"Messi" can't be a bad word for anybody

[Chris30551417jovo]

How does the generative fill prompt "No grass, stone interloc" generate images that violate terms of service?  Is it the use of interloc, where it thinks of it as a brand name?  This would be like violating generative terms  of service by using "Alergy-suffering woman blowing a Kleenex".  Maybe there needs to be a brand filter that translates brands into their more generic analogs as a pre-prompt processing step?

[NicholeWeaver]

I typed in the prompt "lesbians on a date in space" and kept getting the message it caused a "safety issue." But when I type "two women on a date in space, I don't receive the message. Not sure if Adobe is censoring terms like "lesbian," but if they are we need to have a much bigger conversation.