Often when I try to expand a portrait to include a whole face (starting with a partial one), I get a border around the portrait instead of more face. Any way to prevent this? I'm using the online version of Firefly in Chrome on Windows 10.
I've been having a problem with Firefly and started a conversation with Droopydog 500. However, I've been unable to continue because 1. A file I attempted to upload as an example failed to upload and 2. Further communication led to the following error message: Your message couldn't be delivered to the recipient because you don't have permission to send to it.
Ask the recipient's email admin to add you to the accept list for the recipient.
Maybe I can upload the file here. The problem is that when I choose expand in Firefly, too often it simply puts a border or something irrelevant like a tile around the image I want expanded instead of adding to the image itself. Please see the attached file for an example. Am I doing something wrong, or is it a bug? Thanks
Thanks for posting the image. I consolidated all the posts back into this thread.
Ok, this is definitely a problem. I am experiencing the same thing. I will note that when I try to expand the image on all four sides at the same time, this problem happens with every generation. If I try to expand it on two sides at a time:
I had more success, although some of the generations still had this effect. Two side expansion:
I also took this image into photoshop and tried to expand all sides at the same time and it did not have the problem, so it appears to be isolated to the web version of generative expand. If you have access to Photoshop, you might want to use that. Otherwise, on images affected by this problem, do not expand on all four sides at the same time (as a work around for this problem).
Appreciate your following up. I used almost 20% of my generative credits with this problem, so it's more than just annoying. How can this reported as a bug to the developers?
Just as a note on Generative Expand, I consistently see the best results (the only usable results, in my opinion) using it one edge at a time, and being careful to limit the number of pixels being added with any single application to below to 2K in any dimension. I get best quality if no one dimension exceeds 2048 px.
Requirements vary, of course, but I have never seen a usable result from an attempt to expand two, three, or four edges at once. This may be because the aggregate dimensions are too great. (I'm assuming that a 2048x2048 image expanded by 100 px on all sides requires calculating a total image space of 2248 x 2248, which may be incorrect, but quality always seems to suffer in these situations.)
I've noticed Lisa Carney goes further in her work, in which final output quality must be pristine, by avoiding Generative Expand almost entirely and using only Generative Fill.
Hmm...on the web version, generative expand is one of 3 options of generative fill. The other two are insert and remove. So I'm not sure what you mean by using generative fill instead of generative expand. Thanks.
Sorry not to be clear. Generative Fill applies to smaller sections, mostly (but not always) inpainting, as the name implies. Generative Expand means extending one or more entire side(s) of an image, making it larger than it was.
In Photoshop, which is where I use these tools 99% of the time, these operations are clearly differentiated. Generative Expand is initiated with the Crop tool, whereas Generative Fill uses a selection of some kind (marquee, lasso, etc.).
I think the terminology in the web app is a bit misleading, tbh. It probably saves on UI coding.
ok, got it, thanks. Using one side at a time to expand generally gives good results and [sometimes] using two at a time works as well. I'm also using Midjourney and expanding on all 4 sides works almost all the time. Problem with Midjourney is that it won't work on an outside image, only images it creates.