I would like to bring to your attention a recurring issue I've been experiencing with Firefly. Each time I upload my original artwork, I receive the following error:
“Something went wrong. This image produced an error. Please try again. Your reference image violates Firefly user guidelines and was removed.”
My artwork is entirely original, self-created, and does not include any content that would violate community guidelines or infringe on any copyrights. The images I upload are non-cexual, non-offensive, and represent my own creative designs.
I kindly ask to review this issue and provide clarification on why some content is being flagged. My goal is to create detailed drawings and have Firefly generate and fill in the colors and style. While I sometimes succeed, most of the time I receive this error.
Hi @Husef25759865d9sz It is worth posting an example of an image that is giving you the error message. That way people here can try understand what in the image is causing the problem for Firefly.
Agree with @Dean_Utian , examples would allow us to help more. That check is generally not a copyright check, but rather it believes there is something in the image that might not meet the Gen AI guidelines. It does make mistakes in thinking an object looks like something that a human might not agree.
For example: Here are some examples, the flying dragon gave me an error, but the native by the beach sketch worked perfectly. The sketch of the girl didn’t work, but when I used prompts without the sketch, it generated the girl similaly, so it's not an issue of being inappropriate.
@Husef25759865d9sz, it seems to struggle with some artwork. I would note that the analysis to determine "violations" is a different process for prompts than it is for reference images. So it is possible it might generate an image that it refuses to use as a reference image. I have seen other AI image generators do this as well.
Just some observations I have (and others can chime in with their experiences)... Reference images with lots of details seem to be harder for the model to process than ones that are simpler. This can result in rejections or inability to follow the reference. Any reference with a lot of skin showing is more likely to be rejected.
Thank you all for your responses. I've taken note of them and will keep experimenting with AI. My main goal is to learn how to use my sketches or detailed drawings to guide the AI in generating specific results rather than random images. I am testing the machine limits and at the moment, I'm getting mixed results, but in many cases, I can correct them using Photoshop and airbrushing when needed.
7
Replies
AdChoices