V2 verses V3 quality

[Dave Gustafson]

I have extreme respect, admiration for the Firefly team and engineers, but something has gone off the rails concerning the V3 preview. This isn't a question, so I placed it in this section of bugs. If I didn't care about Firefly I wouldn't be here showing examples.

These are direct renders of "multiple generations" using V3 preview off the original V2 file. I will show the V2 first, then V3 will follow each image. I kept trying to get a good image from V3, yes it's a preview, but it needs to be addressed. These are the best of the best V3 images. And with that said I'll shut up lol and let you judge.

V2

V3

 

V2

V3

V2

V3

 

V2

 

V3

 

 

 

 

 

[Dave Gustafson]

I strive for mood, dynamics, ambiance, and storytelling in my images and V3 sucks the life out of them.

V2

 

V3

 

[BlueJeanBaby]

EXACTLY. I am very glad I'm not the only one reporting this. Some of the images 3 renders with my old formulas are hysterically terrible. They need to either FIX 3, and/or leave up 2 for those of us who prefer it. 

[droopydog500]

It feels like from model 1 to 2 to 3 the ability to create a realistic photo has increased at the same time going from 1 to 2 to 3 the ability to create artistic images has decreased. Is that what you are feeling?

[Dave Gustafson]

Hey Droopy good to see you!

Yes, it's like the realism push is overriding everything, it should be a button, not shoved onto everything in its current state, artistic V1 & V2 are heading out the door, I'm sad to say. To be perfectly honest I feel bad posting this, real bad but I need to speak up. Want to see something that pains me to show?

V2

V3

 

Multiple V3 generations with the V2 file and I get blown out lighting, distortions, color bleed in B&W photo, etc.., This is after 10 generations, the best of the best and I have this to show. 

[Dave Gustafson]

This weekend I'll go completely deep dive and see what I can do on my end to circumvent these extreme changes and try to pull a positive out of it. Try to find an artistic path in all of this.

[Kartika Rawat]

Hi there!

Thanks for sharing the detailed post. We appreciate it. Can you post some of the prompts you used to generate images? We'll try them at our end.

Thanks again. Happy to help. 

KR

[Dave Gustafson]

Links are dead due to a hardware problem

[Dave Gustafson]

Concerning the frog, the color bleed was caused by a bug when a Chiaroscuro or Film Noir is applied over an image that had it applied previously

[droopydog500]

Here is another example that supports my realistic versus artistic theory.  I created this series of images called "hideous hotel carpeting" making fun of the design patterns used in hotel carpeting.  Here is this same prompt in each of the models.

Prompt: endless plush hideous hotel broadloom carpet, repetitive pattern, seamless interlaced geometric design, geometric shapes, random patterns, pink light-blue brown

Model 1:

Model 2:

Model 3:

These were the best images generated for this prompt in these models. Adding styles did not significantly change the results (or at least did not improve them).

[Dave Gustafson]

Wow, that is an even more dramatic visual difference than mine. Maybe the increased realism could be a slider like the current intensity sliders, I dunno. 

[@mj]

Keep the old engines alongside the new. A bit like versions Adobe

[droopydog500]

Actually, would rather they ditch model 2 and keep 1 and 3.

[Amy27298162kmmg]

I can't even see an Image 1 Model any more

[Amy27298162kmmg]

Yes, I had issues also. I tried the same prompt in both Image 2 and Image 3 models. The image 3 model gave me clay-like looking flowers. The image 2 model gave me beautiful, frilly flowers. An unbelieveable difference! I don't seem to have access to Image 1 model any more. Please Adobe, Keep the Image 2 model as well... please don't ever get rid of it like it seems has happened with Image 1 model.

[Amy27298162kmmg]

Sorry, my images came in huge and without the captions. The first picture is the Image 3 model, and the other two are the Image 2 model.

[droopydog500]

@Amy27298162kmmg, Adobe Express appears to still use model 1.

[Amy27298162kmmg]

Thank you droopydog500. Are you finding that on the online page? I just tried it last night and I didn't see where I had an option. What am I missing? The result I got for flowers was basicallly a blob.

[droopydog500]

@Amy27298162kmmg, There is not a clear published table, but what we constructed on the Adobe Firefly Discord is:

	Model
	1	2	3
Firefly Text to Image (web)		X	X
Firefly Generative Fill (web)			X
Adobe Express	X
Photoshop General Release	X
Photoshop Beta			X
Illustrator	X
InDesign			X
API (enterprise only)*	X	X

*information from API changelog

[@mj]

8)

[Amy27298162kmmg]

Thank you for the information

[@mj]

Oh curious @droopydog500 ,

Could the Adobe folks on this forum say why?

Best

mj

iMSD

[Kartika Rawat]

Moving from bugs to discussions.

[rickg76305781]

image 3 appears to lack the functionality or image control of image 2. using the same parameters (effects etc)  from image 2 to image 3 produces very different stylistc results. also, working with image 3, application of (ex) movement selections appear to have no effect on the final results.

[Kartika Rawat]

Hi there! 

Sorry to hear that.  Could you share the generated output? Also, provide us the prompt and setting you used to generate images?

Thanks,

KR

Moved to discussions