P: Photo realistic humans not optimal

[sabareeshsabu]

Whenever i try human photos, there is some wierd manioulation of fingures, faces, perspective etc which make the picture looks little odd. I hope the team is still working on fine-tuning it. I know it is to avoid copyright/content authenticity issue but the images are not looking great due to that.

[WilliamOesch, AEL]

Yes, I've noticed some issues with fingers and hands. I did have some success with regenerating the photos. I've come to accept it in the early stages of a beta. 

[sabareeshsabu]

yes, agree

[happie_97]

I notice that as well.

[Tracy.H]

The quality of our models will improve with time, and generated content will better respect the prompt. (Merging all similar threads here so the team can access all requests on this issue)

[Oz24019493dktj]

Many of the times when generating faces they come out terrifying.  It needs to do a better job with generating faces.  

[Prakash291078359qjs]

Hey, Photo realistic is not generating properly, especially I am trying to generate the people... Even I tried using negative prompt to get the output (e.g., no artisitic, no illustration, no CGI etc.) 

[Manu29114066tn2u]

I used "Ultra realistic male Yogi meditating in himalayas" and limbs are not coming out clearly

[Tracy.H]

Thank you for your feedback. The quality of our models will improve with time. Right now, the model doesn't understand negative prompts, so adding it in your prompt will add it to your image.

[Jishuang27092557thfg]

Need more training on girl and cute

[SjonU]

Seems to have a little problem with faces and limb lengths and ends. I am a 3d compositor and video editor. I am anxiously waiting for AI to be able to handle moving images. Especially if you could add a timeline and keyframes.

[Tracy.H]

@SjonU I have also noted your request about video support.Thanks for the feedback!

[biagiodistefano]

The definition of people's faces is not well defined when "photo" is requested as a typology. Often the eyes, nose and mouth, or other small details can appear unrealistic.

[li291573946ve4]

The hand of the generated character is weird.

[vickimartineloader]

Same problem encountered when stipulating graphic or vector. Facial problems and too many fingers

[Jay Kat.]

The AGI cannot illustrate human face.

A lot of distortion in the face and perspective is off in the composition.

https://firefly.adobe.com/generate/images?prompt=hyperealistic+portrait+congo+family&seed=68523&seed...

https://firefly.adobe.com/generate/images?prompt=realistic+portrait+Canadian+family&style=closeup&se...

[JoshCatchur]

I don't work for Adobe, I'm just another user. But part of this is just an issue with most AIs when trying to get more than one person in an image. However, experimenting can help. For the Congo family image, just changing the word congo to "Congolese" produces better results. For the Canadian one, removing the composition setting helped get better results. But it defeats the purpose of having those pre-made settings if you can't use them. 

[MrFreddie90]

First of all I would like to thank you for giving me the opportunity to test Adobe firefly, it is a real pleasure to experiment with this function, as the title suggests, I have seen that the female and male figures still suffer from many processing imperfections, especially in the images where a figure is portrayed in full, while the results of the portraits are excellent.

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

[danielmdq]

Hola!

No se si este es el lugar para esto, pero lo dejo

Noto que en otros sistemas de AI, y también en Firefly, las manos de las personas se muestran siempre con formas extrañas, deformadas, con multitud de dedos... Mi consulta es si esto es prodecto de un bug o bien el sistema necesita más tiempo para mejorar estos detalles

Muchas gracias!

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

[René Emmanuel29549601xbqq]

The tool Text to Image had some problems designing faces and hands, I think it can improve.
Apart from this, the platform is amazing.

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

[tiffbutler]

Version: Beta

Platform: Mac, Chrome, OS Monterey

When I entered Hawaiian vacation to Kauai tourists

I wanted to see what it would do for people/illustrations of people. The people are warped and distorted.

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

[NASER29618953wxfh]

In 90% of images that Firefly generates that have humans or representations of humans (regardless of which content type is chosen), the hands are distorted.  They are either misshapen, missing digits, deformed, missing altogether or more than two, and never appear right.

Similarly, cows, bulls, and horses, almost always have extra legs or hooves.

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

[Helal uddin]

Actually, I love the environment of FireFly

I've exported a few images, and illustrations. But the problem is finger isn't generated properly. which is very ugly and ridiculous. So plz focus on this issue.

Thanks. 

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

[Greg24008437om3x]

Firefly is lacking two major features that many other AI generators have implemented to deal with this particular issue. The first is negative prompts - basically telling the AI all the things you don't want in your image. Firefly does allow you include the "[avoid=xxx]" parameter, where XXX is replaced by something you don't want, but it isn't a powerful as propper negative prompting. The second missing feature is including, either optionally or automatically, face correction via one of the dedicated models buitl specifically to do this. Effectively, after generating your initial image it passes the image though the second model to rebuild the face without the distortions. Some of the models are extraordinary and produce amazing faces. Perhaps this will come as part of the inpainting model that is still "in experimentation".

[wingki298364197i4u]

especially eyes, finger, arms are mess

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

[Gray299429652jwg]

Take a look at the new Midjourny 5.1 You need similar to render humans convincingly.  The tech exists.  Grab it.

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