Spaces:
Running
on
CPU Upgrade
Error
error
Expecting value: line 1 column 1 (char 0)
?
This is pretty neat... although the generation times are rather slow compared to the original and it only makes one image when it should've made four at a time. Also, you can't randomise the seed either when generating images.
Good start though :)
error
Expecting value: line 1 column 1 (char 0)
?
Can’t you just be patient?
Yeah that's good, very nearly as good as the (currently not) online demo. It's weird, cos I get images that seem to come out looking similar in composition etc, but they look like airbrushed cgi paintings whereas the original online demo they actually looked like photographs. Why would that be?
Online demo should not be errored forever.
Yeah that's good, very nearly as good as the (currently not) online demo. It's weird, cos I get images that seem to come out looking similar in composition etc, but they look like airbrushed cgi paintings whereas the original online demo they actually looked like photographs. Why would that be?
This is exactly how I feel about the Stable diffusion v1.5 and this 2.1 demo version, the results feel more authentic. The smoothed/airbrushed version feels so cheap :(
but they look like airbrushed cgi paintings whereas the original online demo they actually looked like photographs. Why would that be?
The model being used and the options being passed should be the same, so I think there are probably prompts that the official demo is constantly adding behind the scenes.
If I knew what they were, it wouldn't be difficult to add them, but... they're not in a place where I can see them.
Well, it's best to wait for the official version to return.😅
Edit:
I modified params a little for now.
Thanks for your time and effort with this.... you're close, but I do hope the original online demo returns!
Interestingly (perhaps) the demo produces images which download to me as .jpg whilst with your version they come in as .webp and are smaller files
Final thoughts, can I assume the person/people who could fix the online demo have not been made aware of the error? I'm assuming it'd be a quick fix if they knew and wanted to correct it.
Are there other channels we could use to let them know? Without being too impatient!!
I tried making it into a jpg.
Are there other channels we could use to let them know? Without being too impatient!!
If you experience a prolonged error with Hugging Face, sending an email to support may speed up the response. I think this is more about whether or not they notice, rather than a matter of prompting.😅
[email protected]
Thanks again... they are now jpegs and I think there's been another step in the right direction quality wise.
Top man! Nearly there 😉
Thanks!!!
Kinda feels that it's just human skin-tone/detail that is not quite back to where it was now. The CGI, airbrushed feel remains.
I've emailed the address provided above, politely hoping the original can be rebooted.
.... Having used it a little now, this is so close. Great job. Any way to turn up the 'look like a human with human skin and hair' dial and turn down the 'look like an AI CGI image dial'?
That was what was so good about the demo, every other AI image I've seen looked like it was created in a computer; the 2.1 demo images often didn't.
Any way to turn up the 'look like a human' dial and turn down the 'look like an AI CGI image dial'?
There shouldn't be a function like that. It should change depending on the contents of the prompt. For example, you could put ", cg" or ", illustration" in negative_prompt, or put ", photo" in prompt, or so, and it should change like that.
Also, changing the guidance_scale should change the prompt response a little. If you make it smaller, the picture will be more straightforward. If you make it bigger, the picture will be more messy, but it will follow the prompt.
And it might be better to select stabilityai/stable-diffusion-2-1-base. I don't know which model the original used, so I'm using stabilityai/stable-diffusion-2-1 as the default for now.
OK, I'll try using 'base'. I'm using prompts saved from when the online demo was working, so I know they are the same. Appreciate your point re negatives, which I had already factored in. Ditto the guidance scale.
Something is different somewhere in the background.
Once again, thanks so much for your time.
..... no discernable difference when using 'base'..... perhaps, it's even less good.
for now I found and added this friedrichor/stable-diffusion-2-1-realistic
... and also apologies, again, if it was something I could fiddle with I'd happily do so... but I'm just an end user 😀
for now I found and added this friedrichor/stable-diffusion-2-1-realistic
This is not any kind of negative reply, just letting you know 😊... but that's not it either.
I wish I knew more, cos this is quite interesting!!
The only other obvious output difference, now your version is producing jpegs is the file size. I think both are set to 768x768, but previously the images came in at c 100kB and now these are c50kB.
Which kind of marries with the sense that there's is just less detail.
Oh... Maybe the official version was upscaled. If that's the case, it can't be reproduced using the API...
https://huggingface.co/spaces?category=image-upscaling
Funny, that first upscaler actually just seems to add a layer of soft cgi AI -ness.... exactly what I'm trying to avoid.
Oh... Maybe the official version was upscaled. If that's the case, it can't be reproduced using the API...
https://huggingface.co/spaces?category=image-upscaling
Did you mean can't be reproduced? If so, then perhaps we really are back to waiting to see if the original can be fixed. You've done a brilliant job of getting close to it 👌👊👍
Really miss this demo as well
Tried running it locally, other models etc but they all end up looking like "CGI" or obvious AI images. This demo looked real.
Anyone have any idea what models/settings this demo was running?
Can @multimodalart help....? Seems very much linked with the original demo we all miss so much.
Can @multimodalart help....? Seems very much linked with the original demo we all miss so much.
You should ask @multimodalart then.
I can't see any way of direct messaging.
I can't see any way of direct messaging.
I messaged @multimodalart at https://huggingface.co/spaces/multimodalart/flux-lora-the-explorer/discussions/37
The notification will be sent to the other person with @+username. Well, it will also be sent if you simply write it in the other person's Discussion column.
How?