composition references is not working

[Alin38774853zh0c]

i've uploaded an reference image but it doesn't follw it at all ( even i've set it to 100%)
refreshing/ restarting browser can not fix this issue

[droopydog500]

Hello @Alin38774853zh0c,

Thank you for your message. I am sorry you are having this problem. 

Is your composition reference image transparent? If so, fill in the transparency with a contrasting colour. The prompt also makes a difference.  Please share an image that shows your prompt and the composition reference.

Thanks,
    droopy

[redheadrev]

I'm having the same issue. No transparency. 

[redheadrev]

Prompt: cartoon of a child receiving communion. 

 I added this as a composition reference at 100%

1. 

    this is the result:

2. 

    

[Kevin Stohlmeyer]

@redheadrev A few things to consider -

Images should have a focal point when used for Composition/Structure reference. This image has a lot going on for the AI to discern the subject structure.

Second, the main subject is obscured - meaning the child has their back to the camera.

Third - The prompt is asking for a cartoon which contradicts the image used for reference.

Try this instead - no uploaded image, prompt is:

"Illustration of a girl receiving holy communion from an adult in front of her; adult is kneeling"

you can ellaborate from there "girl with blonde hair and pig tails" etc. to refine.

If you really want a cartoon-like image, you can add that to the beginning of the prompt.

"Cartoon illustration of a girl receiving holy communion from an adult in front of her; adult is kneeling"

[redheadrev]

Kevin, thank you for your suggestions. The problem persists however with no manner of composition reference making any change to the rendering 

[Merritt36188282sgn7]

The problem is that you are using the word "child" in the prompt. Firefly ignores the composition reference if the words "child", "girl", "boy", "daughter", etc. appears in the prompt. I usually use the word "person" if I need firefly to look at the composition reference. Though that can still be a problem, for example if I have a picture of a frowning adult, but I want a frowning child that has the same expression as the picture. I would be grateful if anyone has a solution for this, or if they know if there's an official explanation of this behavior from Adobe. Thanks!

[Kartika Rawat]

Thanks for pointing that out. Firefly currently avoids generating videos/ images that include children. This is a careful step to make sure the model handles such content responsibly. We know it can be limiting, but it’s part of ensuring safety as the tool improves.

I hope this helps.

^KR