viel zu viele Fehler

[jamiea93619187]

Während Adobe Firefly anfangs noch durchaus respektable Ergebnisse geliefert hat, hat sich die Qualität inzwischen stark verschlechtert. Prompts werden oft ignoriert – statt einer Ansicht von hinten sind die Personen plötzlich von vorne zu sehen, und ähnliches. Gleichzeitig treten gravierende Fehler auf: Auf Gruppenbildern tauchen plötzlich Zwerge auf, Menschen haben keine Gesichter mehr usw.

Der Dienst ist in diesem Zustand kaum noch brauchbar – und es gibt nicht einmal eine Möglichkeit, solche Fehler sinnvoll zu melden. So viele Daumen runter kann man gar nicht anklicken, wie nötig wären.

Wann wird der Dienst endlich überarbeitet und verbessert? ChatGPT funktioniert im Vergleich dazu tausendmal besser – obwohl ich dafür kein kostenpflichtiges Abo habe, das ich wie adobe cloud teuer bezahle. Bitte fixt das!

[Kartika Rawat]

Hi jamiea93619187!

Thanks for your feedback, and we’re sorry Firefly hasn’t been meeting your expectations recently. Could you share some generated images and the prompt with us?

Your input is incredibly valuable, and we appreciate your patience as we continue to enhance Firefly.

Let us know.

KR

[jamiea93619187]

Unfortunately, I didn't download the incorrect results. There's no point in keeping them. I'll re-enter the programs to see if I can reproduce the results.

[jamiea93619187]

[jamiea93619187]

[jamiea93619187]

and thats just a quick try.

[jamiea93619187]

i can show you a lot more.  There are women, dwarves, and children instead of men. Hands and arms are depicted completely incorrectly, legs are missing? The baklen suddenly disappear into the sky. Instead of looking forward, they look backward, etc.

[Monika Gause]

In your Prompt you did not mention that you are viewing from behind. You also did not mention grown up men.

[jamiea93619187]

I don't think it changes the fact that Firefly  often don't understand promts correctly. Below is an example. The same prompt is used by othe ai image tools. 

promt: A woman is in a huge cave, looking down at a mining town inside the cave. Above is a hole in the cave, and through it you can see the sky. Everything is done in a fantasy anime style! It's dark; light only comes through the hole in the cave. The woman is in a cave! There are no trees or plants in a cave.

[Monika Gause]

Firefly doesn't understand negative prompting. If you do not want something, do not mention it.

[jamiea93619187]

i tried it both ways. ( without negative promt) doesnt make it better:

[Monika Gause]

I think you might need to be more precise. This actually is a hole in the cave, but if you want a small hole at the top, you need to tell it.

[jamiea93619187]

i tried no chance for me to get the image i wanted

[AKS666]

Truth is, Firefly is BY FAR the worst of all generative AIs out there. 

[droopydog500]

So, I used model 3 rather than 4. I have found when you want multiple things in an image, it can help to make a base image of the background and then use generative fill to add things in (or sometimes subtract things out).

I took the one on the right to generative fill and added the woman:

resulting in:

 

My best,

    droopy