Firefly is terrible

I fully agree with you even a year later. The engine also regards some prompts as "inappropriate" even though it truly isn't. 

I attempt to generate image to video of godlike visuals and movement, light rays, floating objects, and yet it gives me a "error" as it doesn't fit with their guidelines. Useless tool. Such a limiting and disappointing tool for such a long standing business. 

I'm two seconds away from cancelling my subscription after weeks of trying to create useful results. I'm yet to see any.. at all. 

After a frustrating morning on Firefly I Googled 'Firefly is terrible' and this thread came up. I agree with the above sentiments, it really is an appaling tool.

I submitted a vector of a silouetted river and asked for variations ('variations of river'), and it is returning absolute garbage; completely irrelevant pictures of people or cartoons. It ignores prompts like 'single colour' or shapes (eg 'contain river in a circle'). Just for kicks I tried Firefly 4 Ultra and it decided to add text in what looks like Swedish! What's that all about?!

I'm not a big AI image generator user but on the odd occassion I'm having much better luck with the free services out there. Considering the amount of effort put into promoting Firefly by Adobe, it's embarrassing. 

I agree, firefly actually is almost useless. It just generate shutterstock-like pics but the overall quality is really bad. images are all poor and flat (in the worst way).

I agree, Firefly simply is very bad on it's images outputs, and 4 out of 10 times I gety the "Can't load, we can't display the generated image" Midjourney is BY FAR BETTER and they don't stab you with high costs month in and month out 0 / 10 stars

Yes, the difference can even be worse …

Same silly prompt - two results. One from Firefly, one from ChatGPT. Guess which is which 😉

Hey @OhJakie 

Adding a bit more to what @droopydog500 mentioned.

Start with simple prompts. Build on the image.
Provide style or reference reference images. Even a simple drawing goes a long way.

Reference Image: Firefly will attempt to closely match the image you provided
Style Image: the style you are looking for in the outcome.

Use the image you like as a style or reference image and build on it.
Change the prompt according to the desired outcome.

Cheers

O. Nate

Has it actually got worse? I have just created a picture of three people talking to each other. They have neither arms nor faces. I used a very detailed prompt for this. When I shortened it to just one sentence, I at least got some faces, but the scary kind of faces. It seems to be worse than last year.

You are not alone. I can use the same prompt in ChatGPT an get a great result while Firefly is in "Zombie-Mode" again. People who look so strange that you feel like you're back in the early days of AI. I don't even want to talk about video generation. My “free” attempts were useless. And as long as I can't expect better results, I'm not willing to pay any more money. I hope Adobe brings a noticeable improvement in quality.

Firefly still is so bad. Low, low, low, compared to other engines.
It ignores half (or more) of the prompt, the output is low quality and you have to be extreme lucky to get something you can use in real life.

If you test the same prompt with different macjhines, you see how far Firefly is behind. It's like stoneage and old egypt.