Copy link to clipboard
Copied
These are just some examples of many prompts / images that have been rejected for absolutely no reason that I can think of. Firefly seems INSANELY restrictive, and does not offer any explanation or guidence on how to avoid rejection. It seems totally arbitrary. I've made many videos that really push the line in terms of being scary and disturbing, but for some reason, seemingly innocuous prompts get rejected.
Here's one that was rejected no matter what I prompted. "camera slow push in on a weary kneeling sci-fi character wearing a red cape in the rain in the middle of a cyberpunk city with red neon lighting"
Here's another. An old image of my Grandmother from the 1950s. I thought it would be interesting to animate gramma moving around from this age. Rejected.
"Glamorous woman from the 1950s smiling at the camera"
Copy link to clipboard
Copied
Hello @silvershanks,
Thank you for your message. I am sorry you are having this problem.
I am presuming you are using text to video.
Your first image errs as a reference image for me, also. As you noted, it does not tell you why, but i suspect the model might possibly be interpreting the red on the cape as blood. It will not generate using images it thinks reflect violence (or the result of violence), which includes things it thinks are bloody.
For the second image, that also erred for me. The beta model will not use a reference image if it thinks it might be a child. I think the model might think that image depicts someone under the age of 18. If I swap your image with one of a clearly elderly lady in a similar pose, it generates it.
I would note, this is a beta. It is possible the restrictions are over-conservative during the beta.
My best,
droopy
Copy link to clipboard
Copied
I did figure out one work-around to red colors being rejected in firefly. You can simply invert the color hue of your source image, feed it through the video generator, then just reverse the hue sift back to normal in the video color grade. Easy.