|
|
|||
| forums: groups: | |||
|
"Been playing around with Google Imagefx"
Love you, too
I've been messing around with Google's image generator this evening, feeding it some old prompts I've used on Bing in the past. Apologies if it's a bit scattershot or repetitive- just throwing things at it to see how it handles them. Stuff I've found so far:
- doesn't handle "90s style" as well as Bing. If you want to recreate fictional 90s gameshows, this probably isn't the generator for you right now.
-does a better job of "normal looking" people; expressions look a bit less uncanny, and people are more likely to be normal weight instead of extremely skinny or overweight.
-I've not got the hang of getting it to generate gunge tanks yet. There's probably a way to do it, but I think I need to experiment more. The tricks from Bing don't work, which is unsurprising.
-I'm not getting filtered as often as Bing, but I'm not trying anything particularly saucy.
- doesn't handle "90s style" as well as Bing. If you want to recreate fictional 90s gameshows, this probably isn't the generator for you right now.
-does a better job of "normal looking" people; expressions look a bit less uncanny, and people are more likely to be normal weight instead of extremely skinny or overweight.
-I've not got the hang of getting it to generate gunge tanks yet. There's probably a way to do it, but I think I need to experiment more. The tricks from Bing don't work, which is unsurprising.
-I'm not getting filtered as often as Bing, but I'm not trying anything particularly saucy.
GungeDrop said:
Ooh, looking forward to seeing what you come up with!
thereald said: This looks like it will meet my needs rather nicely.
Ooh, looking forward to seeing what you come up with!
Realism seems to be on par with Flux, and the prompt interpretation is similar to Dalle-3. I've made some nice images so far, it will be interesting to see how they translate to video
thereald said:
Realism seems be be on par with Flux, and the prompt interpretation is similar to Dalle-3. I've made some nice images so far, it will be interesting to see how they translate to video
GungeDrop said:
Ooh, looking forward to seeing what you come up with!
thereald said: This looks like it will meet my needs rather nicely.
Ooh, looking forward to seeing what you come up with!
Realism seems be be on par with Flux, and the prompt interpretation is similar to Dalle-3. I've made some nice images so far, it will be interesting to see how they translate to video
I generally occur. I had a good play around with it when it launched. I think it's a definitive improvement over Dalle3 but it does also have some strength/weaknesses.
1) The realism is good if thats what you are aiming for but there are a few trigger words that trip the filters still. You can work around them using old Dalle3 tricks but the sensitivity of second stage filters are still there.
2) Prompt adherence is generally good but there is definitely a cut off on length and/or complexity before it starts ignoring elements. Haven't figured this out yet I've had it ignore large chunks of the prompt.
3) The model's CFG looks to be quite low. By this, I mean if you use a prompt to generate an image and then re-generate, the variation between the 2 is quite low. Not always a disadvantage as it does help recreating prompts.
4) Unlike Dalle, there is no rewriting of the prompt. This again with the lower CFG will mean you get more consistency between images.
From a messy perspective, when compared to Dalle3, Grok and Flux I would still rank Dalle3 as the best particularly if you use the azure API. Dalle just understands better but it has more filtering now. Grok has produce some good basic images but the data set for messy is too small but it is way less filtered. The same applies to the base flux model however the benefit of Flux is it can be easily be trained.
As most people don't have access to unfiltered Dalle3,
Flux (Finetune) >> Imagen3 >> Dalle3 (Azure api) >> Grok2 >> Base Flux
There are rumors that when Imagen3 releases on AI Studio it will have a filter flag that you can turn. That will be interesting.
Whats the URL? And how well does it do pies?
Sleazoid44 said: Whats the URL? And how well does it do pies?
https://letmegooglethat.com/?q=Imagefx
Wow. First half hour, and I must say I'm really impressed. The models look more girl-next-door rather than Instagram model with excessive make-up. Here's a few of my efforts.
I'm sure its me but my results weren't that good. The just-wet dress was OK (a little dark), the woman throwing a pie in the face of another was good, but a pie flying through the air or thrown through the air was just a pie in the air near the woman; "smashes" got closer. Also, when I had the pie dripping off her face onto her breasts, front or shoulders, it didn't work at all.
Also the size I got was 1408 x 768 and also the DL button produced a file in the .jpg format that wouldn't open. If I changed the extension to .png it did.
Unedited examples attached here.
Also the size I got was 1408 x 768 and also the DL button produced a file in the .jpg format that wouldn't open. If I changed the extension to .png it did.
Unedited examples attached here.
Yes , this just gets better and better , tried the, look up surprised look . and got a great results.
Can not think how you got the boobs though , seems anything suggestive and its , we cannot find what you want.
Can not think how you got the boobs though , seems anything suggestive and its , we cannot find what you want.
Sleazoid44 said: I'm sure its me but my results weren't that good. The just-wet dress was OK (a little dark), the woman throwing a pie in the face of another was good, but a pie flying through the air or thrown through the air was just a pie in the air near the woman; "smashes" got closer. Also, when I had the pie dripping off her face onto her breasts, front or shoulders, it didn't work at all.
Also the size I got was 1408 x 768 and also the DL button produced a file in the .jpg format that wouldn't open. If I changed the extension to .png it did.
Unedited examples attached here.
Also the size I got was 1408 x 768 and also the DL button produced a file in the .jpg format that wouldn't open. If I changed the extension to .png it did.
Unedited examples attached here.
Ooh, interesting. I've not tried pies yet, thanks for giving that a shot. Definitely looks like their aim is a little off
It doesn't currently seem to have a character limit, so writing more detail into the prompt might work? Hope you figure it out, thanks for sharing what you've found so far!A few of my earlier testings. As I say, similar exploits and workarounds that worked for Dalle/Bing still exist in this model currently.
Most of these were just my old Bing prompts applied directly without change. I'm sure with some better prompt engineering you could get better results but a year on, I'm still not seeing the time and effort worth it Vs a Flux tune.
Most of these were just my old Bing prompts applied directly without change. I'm sure with some better prompt engineering you could get better results but a year on, I'm still not seeing the time and effort worth it Vs a Flux tune.
schneider said: Wow. First half hour, and I must say I'm really impressed. The models look more girl-next-door rather than Instagram model with excessive make-up. Here's a few of my efforts.
Would you share the prompt? Would love to try on a tuned Flux model to see how they compare.
noise said:
Would you share the prompt? Would love to try on a tuned Flux model to see how they compare.
schneider said: Wow. First half hour, and I must say I'm really impressed. The models look more girl-next-door rather than Instagram model with excessive make-up. Here's a few of my efforts.
Would you share the prompt? Would love to try on a tuned Flux model to see how they compare.
I vary the prompts slightly, but I go along the lines of
"Woman aged 25 years old , at a nightclub , sitting on chair with a pipe at the top with live audience , light-yellow mayyo and syrup falling from above on head from the pipe , hair enveloped in light-yellow mayyo and syrup, not face , friendly smile , auburn medium length hair , blue dress , no props , people in the background"
messg said: A few of my earlier testings. As I say, similar exploits and workarounds that worked for Dalle/Bing still exist in this model currently.
Most of these were just my old Bing prompts applied directly without change. I'm sure with some better prompt engineering you could get better results but a year on, I'm still not seeing the time and effort worth it Vs a Flux tune.
Most of these were just my old Bing prompts applied directly without change. I'm sure with some better prompt engineering you could get better results but a year on, I'm still not seeing the time and effort worth it Vs a Flux tune.
Would love to know what prompts you used to create these. Would appreciate you sharing.
~ruegger~
I've enjoyed ImageFX so far. Definitely much better than Bing in terms of realism, the quality of the background scenes, some of the detail that can be achieved for the props, and the photorealism of the models. I like that the scenes and models change very little between prompts as this makes it easier to build up 'stories' over a series of prompts. The filters appear to be a bit more relaxed on some of the sexier aspects big boobs, stockings/suspenders, lingerie, etc.
Down sides so far are that it can be frustrating when you do want to change how the model looks both in terms of physical characteristics (hairstyles etc.) and what they are wearing. The main frustration, however, is in getting the models really messy. I know that this has been an issue with other AI platforms in the past, but appears to be especially difficult to achieve with ImageFX. I have totally failed to get a good pie in the face yet, or even a pie on the head. I have tried the following prompts which got reasonable results on Bing but with zero success on ImageFX so far: throw, hit, splat, slap, wipe, smear, push, press, hold anyone got a prompt that has worked for them? I have had better results with gunge from above, and pouring buckets or bowls over the head, but even then it has proved difficult to make and keep the face messy. Bing favourites for this, such as engulfed or encased, don't appear to work on ImageFX. I also use the term 'explode' on Bing when the pie hits as that gives reasonable results but again, no consistent luck with Image FX.
It would be great to work out these deficiencies as, in general, I think this is a big step forward just at the time that Bing appears to be going backwards.
Down sides so far are that it can be frustrating when you do want to change how the model looks both in terms of physical characteristics (hairstyles etc.) and what they are wearing. The main frustration, however, is in getting the models really messy. I know that this has been an issue with other AI platforms in the past, but appears to be especially difficult to achieve with ImageFX. I have totally failed to get a good pie in the face yet, or even a pie on the head. I have tried the following prompts which got reasonable results on Bing but with zero success on ImageFX so far: throw, hit, splat, slap, wipe, smear, push, press, hold anyone got a prompt that has worked for them? I have had better results with gunge from above, and pouring buckets or bowls over the head, but even then it has proved difficult to make and keep the face messy. Bing favourites for this, such as engulfed or encased, don't appear to work on ImageFX. I also use the term 'explode' on Bing when the pie hits as that gives reasonable results but again, no consistent luck with Image FX.
It would be great to work out these deficiencies as, in general, I think this is a big step forward just at the time that Bing appears to be going backwards.
Clown Ralph said: I've enjoyed ImageFX so far. Definitely much better than Bing in terms of realism, the quality of the background scenes, some of the detail that can be achieved for the props, and the photorealism of the models. I like that the scenes and models change very little between prompts as this makes it easier to build up 'stories' over a series of prompts. The filters appear to be a bit more relaxed on some of the sexier aspects big boobs, stockings/suspenders, lingerie, etc.
Down sides so far are that it can be frustrating when you do want to change how the model looks both in terms of physical characteristics (hairstyles etc.) and what they are wearing. The main frustration, however, is in getting the models really messy. I know that this has been an issue with other AI platforms in the past, but appears to be especially difficult to achieve with ImageFX. I have totally failed to get a good pie in the face yet, or even a pie on the head. I have tried the following prompts which got reasonable results on Bing but with zero success on ImageFX so far: throw, hit, splat, slap, wipe, smear, push, press, hold anyone got a prompt that has worked for them? I have had better results with gunge from above, and pouring buckets or bowls over the head, but even then it has proved difficult to make and keep the face messy. Bing favourites for this, such as engulfed or encased, don't appear to work on ImageFX. I also use the term 'explode' on Bing when the pie hits as that gives reasonable results but again, no consistent luck with Image FX.
It would be great to work out these deficiencies as, in general, I think this is a big step forward just at the time that Bing appears to be going backwards.
Down sides so far are that it can be frustrating when you do want to change how the model looks both in terms of physical characteristics (hairstyles etc.) and what they are wearing. The main frustration, however, is in getting the models really messy. I know that this has been an issue with other AI platforms in the past, but appears to be especially difficult to achieve with ImageFX. I have totally failed to get a good pie in the face yet, or even a pie on the head. I have tried the following prompts which got reasonable results on Bing but with zero success on ImageFX so far: throw, hit, splat, slap, wipe, smear, push, press, hold anyone got a prompt that has worked for them? I have had better results with gunge from above, and pouring buckets or bowls over the head, but even then it has proved difficult to make and keep the face messy. Bing favourites for this, such as engulfed or encased, don't appear to work on ImageFX. I also use the term 'explode' on Bing when the pie hits as that gives reasonable results but again, no consistent luck with Image FX.
It would be great to work out these deficiencies as, in general, I think this is a big step forward just at the time that Bing appears to be going backwards.
This is with gunge/slime for the most part, but I've had some success by specifying "completely covered in", or "almost unrecognisable under". It's pretty fiddly though, I do agree. It's not got the character limit that Bing has, so try just repeating the message a few times- reiterate how messy they are, in a few different ways.
Clown Ralph said: I've enjoyed ImageFX so far. Definitely much better than Bing in terms of realism, the quality of the background scenes, some of the detail that can be achieved for the props, and the photorealism of the models. I like that the scenes and models change very little between prompts as this makes it easier to build up 'stories' over a series of prompts. The filters appear to be a bit more relaxed on some of the sexier aspects big boobs, stockings/suspenders, lingerie, etc.
Down sides so far are that it can be frustrating when you do want to change how the model looks both in terms of physical characteristics (hairstyles etc.) and what they are wearing. The main frustration, however, is in getting the models really messy. I know that this has been an issue with other AI platforms in the past, but appears to be especially difficult to achieve with ImageFX. I have totally failed to get a good pie in the face yet, or even a pie on the head. I have tried the following prompts which got reasonable results on Bing but with zero success on ImageFX so far: throw, hit, splat, slap, wipe, smear, push, press, hold anyone got a prompt that has worked for them? I have had better results with gunge from above, and pouring buckets or bowls over the head, but even then it has proved difficult to make and keep the face messy. Bing favourites for this, such as engulfed or encased, don't appear to work on ImageFX. I also use the term 'explode' on Bing when the pie hits as that gives reasonable results but again, no consistent luck with Image FX.
It would be great to work out these deficiencies as, in general, I think this is a big step forward just at the time that Bing appears to be going backwards.
Down sides so far are that it can be frustrating when you do want to change how the model looks both in terms of physical characteristics (hairstyles etc.) and what they are wearing. The main frustration, however, is in getting the models really messy. I know that this has been an issue with other AI platforms in the past, but appears to be especially difficult to achieve with ImageFX. I have totally failed to get a good pie in the face yet, or even a pie on the head. I have tried the following prompts which got reasonable results on Bing but with zero success on ImageFX so far: throw, hit, splat, slap, wipe, smear, push, press, hold anyone got a prompt that has worked for them? I have had better results with gunge from above, and pouring buckets or bowls over the head, but even then it has proved difficult to make and keep the face messy. Bing favourites for this, such as engulfed or encased, don't appear to work on ImageFX. I also use the term 'explode' on Bing when the pie hits as that gives reasonable results but again, no consistent luck with Image FX.
It would be great to work out these deficiencies as, in general, I think this is a big step forward just at the time that Bing appears to be going backwards.
All the tricks I used for Dalle3 for coverage also work for Imagefx - but like Gungedrop, these are for slime and not pies.
My advice would be to describe the effect, rather than the action. So maybe something like "dollops of cream are splattered on their face" rather than "pie pushed into face" for example. I'm not a pie person so don't have anything I know works for sure.
Agree I'm finding it's great for everything apart from certain physical characteristics of the subject - it's like beyond about 3 or 4 characteristics it loses track, and it seems to disregard any description of height. But at least you don't have to use euphemisms.
Two main criticism are the faces tend to be too similar from photo to photo, although you can yuse terms like oval face, diamond face, high cheekbones, small nose, wide eyes etc.
One thing that always gets the dog in Bing, but not here is the words bra and bikini!
One thing that always gets the dog in Bing, but not here is the words bra and bikini!
schneider said: Two main criticism are the faces tend to be too similar from photo to photo, although you can yuse terms like oval face, diamond face, high cheekbones, small nose, wide eyes etc.
One thing that always gets the dog in Bing, but not here is the words bra and bikini!
One thing that always gets the dog in Bing, but not here is the words bra and bikini!
Basically any clothing (as long as it's not transparent) will work.
Faces are similar from photo to photo, but the faces in your images don't look like the faces in my images. It's not the same as "Bing Face" as it got to be called. As well as face shapes, you can describe make-up, ethnicity, nationality, and for want of a better word, archetype, words like hipster, model, influencer, goth, geek, and so on. By default, the seed is the same for all generations based on the same prompt, so you're going to get subjects looking the same each time unless you change the prompt. You can set the seed to be 'unlocked' so it uses a new seed for each generation, which can offer a little bit more variety (but only a little bit).
I played around with Imagefx today and had some pretty good results for pies. The key was using "splatters" instead of "hits" and specifying that the mess stays on the girl.
"at an outdoor party , a cream pie splatters on the face of a woman wearing a soaked see-through halter-top chiffon sheath with a crisscross bodice and long gloves , leaving pie on her face and dress."
I also got better results with the active tense (pie splatters face) than passive (face is splattered by).
I like how the "soaked see-through" results in pokies! I haven't had much luck with that on Bing or even Krea.
I'm still getting relatively small pictures (1408x768) and all named "image_fx_1.jpg" requiring me to change the extensions to .png!
Attached here are a few of the successful pictures and one of them after editing in PhotoShop Elements (just cropping and expanding).
"at an outdoor party , a cream pie splatters on the face of a woman wearing a soaked see-through halter-top chiffon sheath with a crisscross bodice and long gloves , leaving pie on her face and dress."
I also got better results with the active tense (pie splatters face) than passive (face is splattered by).
I like how the "soaked see-through" results in pokies! I haven't had much luck with that on Bing or even Krea.
I'm still getting relatively small pictures (1408x768) and all named "image_fx_1.jpg" requiring me to change the extensions to .png!
Attached here are a few of the successful pictures and one of them after editing in PhotoShop Elements (just cropping and expanding).
kyxz11 said: Out of curiosity, did anyone managed to get their account banned yet from experimenting with prompts?
I also wonder if there is limit of how many images one could generate in a day.
I also wonder if there is limit of how many images one could generate in a day.
Not trying to get myself banned, but unlike Bing there is no UI distinction between filtered images and filtered prompts.
There is a daily limit, it's around 200 generations. Hit it the other day. The UI shows you how many generations in each session, but not a daily total.
thereald said:
Not trying to get myself banned, but unlike Bing there is no UI distinction between filtered images and filtered prompts.
There is a daily limit, it's around 200 generations. Hit it the other day. The UI shows you how many generations in each session, but not a daily total.
kyxz11 said: Out of curiosity, did anyone managed to get their account banned yet from experimenting with prompts?
I also wonder if there is limit of how many images one could generate in a day.
I also wonder if there is limit of how many images one could generate in a day.
Not trying to get myself banned, but unlike Bing there is no UI distinction between filtered images and filtered prompts.
There is a daily limit, it's around 200 generations. Hit it the other day. The UI shows you how many generations in each session, but not a daily total.
Its less obvious but there is a separate prompt filter and an image filter in place. The prompt filter will return no images within 2-3 seconds. If the generation continues but you fail after 10-15 secs that is the image filter. you can tweak the prompt and see the difference in generation time. There isn't however a separate error returned in the browser. There are different Post responses logged if you want to go down to that level.
Design & Code ©1998-2025 Loverbuns, LLC 18 U.S.C. 2257 Record-Keeping Requirements Compliance Statement
Epoch Billing Support Log In