|
I feel like I've been to this market in hong kong
|
#
?
Jun 25, 2023 19:12
|
|
|
|
| # ? May 30, 2024 19:01 |
|
|
all those jeezii are made of edible wax
|
|
#
?
Jun 25, 2023 19:21
|
|
|
Ghaz posted:why doesn't it ever give me poo poo this good when I try garden of earthly delights That was MJ 5.2 with one level of zoom out. Putting Bosch's name and 'oil painting' into the prompt were the secret words for me
|
|
#
?
Jun 25, 2023 22:22
|
|
|
https://hugelolcdn.com/v/894371.mp4
|
|
#
?
Jun 26, 2023 03:42
|
|
|
This is going to be a built in feature of gimp and every open source art thing in ten years. People will look back and think "how were they able to charge for that? And why is his foot all hosed up?"
|
|
#
?
Jun 26, 2023 03:48
|
|
|
is there no way to prompt the fill? seems a lot less useful that way
|
|
#
?
Jun 26, 2023 03:53
|
|
|
There is, but you can choose to leave it blank. It's incredibly useful and powerful.
|
|
#
?
Jun 26, 2023 04:21
|
|
|
ahh nice
|
|
#
?
Jun 26, 2023 04:22
|
|
|
Strike while the gimmick is hot.
|
|
#
?
Jun 26, 2023 09:52
|
|
|
Usable DragGAN https://github.com/XingangPan/DragGAN https://huggingface.co/spaces/radames/DragGan 
|
|
#
?
Jun 26, 2023 16:32
|
|
|
Megazver posted:Usable DragGAN Looks wild, but is there any practical way to use it yet? I think I read you need a model trained for each type of subject and it only works on GAN images. The Github link mentions a way to GAN-ify an image (I think?) but it didn't look straightforward to set up.
|
|
#
?
Jun 27, 2023 00:11
|
|
|
I also can't imagine a lot of use cases for it. Neat trick, but I'd rather use controlnet to position the subject. Animation? Maybe
|
|
#
?
Jun 27, 2023 07:30
|
|
|
$1.50 hot dog vacuum from Costco    $1.50 hotdog national TV presidential candidate from Costco in the style of H.R. Giger 
Mercury_Storm fucked around with this message at 09:35 on Jun 27, 2023 |
|
#
?
Jun 27, 2023 09:12
|
|
|
KinkyJohn posted:I also can't imagine a lot of use cases for it. Neat trick, but I'd rather use controlnet to position the subject. Animation? Maybe Fixing family photos, especially with children who never look at the camera A lot of people use that Instagram filter that makes your eyes imperceptibly larger Tons of nefarious purposes. Make some lady's dress extra low cut, turn the politicians head so he's looking at her, make him look surprised with his mouth open, make her look at him angerly. Whatever. Take a normal photo, make two people turn and look at each other and change their expression, totally changing the context of the photo. You could do all this before but was a real pain in the rear end
|
|
#
?
Jun 27, 2023 12:07
|
|
|
 pink plastic el camino in restaurant drive-through style of barbie advertisement highly detailed
|
|
#
?
Jun 27, 2023 12:10
|
|
|
MJ 5.2 seems to be much better at creating fake tech products. Earlier versions had weird perspective issues, hosed up instrumentation placement etc.        
|
|
#
?
Jun 27, 2023 12:28
|
|
|
Mercury_Storm posted:$1.50 hot dog vacuum from Costco don't doxx me mods???????????????
|
|
#
?
Jun 27, 2023 12:33
|
|
|
The zoom out feature for Midjourney is a lot of fun. Between the zoom out and piling on remix prompts / changing aspect ratios, I'm finding that I'm spending more time on refining and exploring individual concepts. Making an image with a tall aspect ratio then remixing with a custom zoom to fill out the sides of the image seems like a useful technique. lovecore joypunk hippie tumblr, 50mm cinematic, wearing rose-colored glasses and wandering through a melting psychedelic iridescent field of wildflowers, photograph, photography  From the initial prompt, I chose #4 and modified the zoom and aspect ratio slightly to get a taller image, applied a 'mild' vary to try to clarify some of the elements of the image:     At this point, I modified the prompt: lovecore joypunk hippie tumblr, 50mm cinematic, wearing rose-colored glasses and happily wandering through a melting psychedelic iridescent field of glowing mushrooms at night, photograph, photography  I took #2 from the above grid, did a 'strong' vary to emphasize the mushrooms and followed #2 and #3 from the new grid:  Path from #2:    Path from #3:    Same idea, different prompts: Eyeball on a pike, Neversoft logo    darkcore horror violence tumblr, 50mm cinematic, braces made of barbed wire on teeth    
Soulhunter fucked around with this message at 15:48 on Jun 27, 2023 |
|
#
?
Jun 27, 2023 15:42
|
|
|
this is so rad
|
|
#
?
Jun 27, 2023 15:46
|
|
|
Turns out you can revive old jobs and play with the zoom level by using the /show [job id] command to make the zoom out buttons appear. This is so cool! I can't wait for more inpainting/outpainting type features to roll out for Midjourney.      
|
|
#
?
Jun 27, 2023 16:54
|
|
|
Soulhunter posted:Turns out you can revive old jobs and play with the zoom level by using the /show [job id] command to make the zoom out buttons appear. There's always a bigger foot
|
|
#
?
Jun 27, 2023 16:56
|
|
|
Soulhunter posted:Turns out you can revive old jobs and play with the zoom level by using the /show [job id] command to make the zoom out buttons appear. Perhaps one day even.... not on Discord.
|
|
#
?
Jun 27, 2023 17:20
|
|
|
With how many $30/mo paid people are in there it's kind of wild to me that the Midjourney people haven't paid a contractor the like 2 weeks it would take to make a web frontend for the process
|
|
#
?
Jun 27, 2023 17:30
|
|
|
AARD VARKMAN posted:With how many $30/mo paid people are in there it's kind of wild to me that the Midjourney people haven't paid a contractor the like 2 weeks it would take to make a web frontend for the process you can't print money if you're spending money.
|
|
#
?
Jun 27, 2023 17:36
|
|
|
the more convenient the interface the more people will run out of fast hours and pay for a higher tier. keep making it better and easier to use and ensure the tiers are profitable. Of course, that's not how finance works, but it'd nice if it did 
|
|
#
?
Jun 27, 2023 17:41
|
|
|
speaking of convenient interface, you can do animations with diffusion bee real easily now i'm just making some nice stock photo scenes for the old corporate website https://i.imgur.com/yUWVTgf.mp4
|
|
#
?
Jun 27, 2023 17:55
|
|
|
Earwicker posted:speaking of convenient interface, you can do animations with diffusion bee real easily now Diffusion Bee is Mac only? Probably only a matter of time before it or something like it is ported to win / linux. Like the animation function assuming you have fine control and it's not "I'm going to let the AI go for a walk" which this still looks really smooth compared to what I was last playing with.
|
|
#
?
Jun 27, 2023 17:59
|
|
|
pixaal posted:Diffusion Bee is Mac only? Probably only a matter of time before it or something like it is ported to win / linux. Like the animation function assuming you have fine control and it's not "I'm going to let the AI go for a walk" which this still looks really smooth compared to what I was last playing with. the animation function is pretty new and there are two different modes. with this one you basically give the ai four different prompts, all of which will use the same seed and set of standard diffusion paramaters (guidance scale etc), and a number of frames, and it builds the transition over them. the animation above is 100 frames, and each frame had 25 steps. it took about 45 minutes to complete (on a macbook pro m2). the other elements you can control are the rate of zoom, rotation, and movement along the x and y axis. there is a second mode where it will build a transition between two prompts which each can have their own seeds and different parameters. ive only played with it a little but its a bit rougher. Earwicker fucked around with this message at 18:15 on Jun 27, 2023 |
|
#
?
Jun 27, 2023 18:07
|
|
|
   prompt was "dinner aint dinner without the beans"
|
|
#
?
Jun 27, 2023 18:31
|
|
|
Lol the cat at the table meme represented
|
|
#
?
Jun 27, 2023 18:35
|
|
|
 Basically a clipped prompt but a real fun one, SD Dreamshaper 6.31
|
|
#
?
Jun 27, 2023 22:30
|
|
|
AARD VARKMAN posted:With how many $30/mo paid people are in there it's kind of wild to me that the Midjourney people haven't paid a contractor the like 2 weeks it would take to make a web frontend for the process They're probably at/near capacity If they had a web interface I would probably cough up $30 but I am not going to gently caress around with discord as an interface
|
|
#
?
Jun 27, 2023 23:32
|
|
|
Amazing. What software and prompt?
|
|
#
?
Jun 27, 2023 23:47
|
|
|
comparing two methods of animated stable diffusion. same prompts, same seeds. 100 frames, 25 steps each. the top one took about 4x longer than the bottom one. Interpolator: https://i.imgur.com/Iq61Ysb.mp4 Deforum Video: https://i.imgur.com/sL9THCb.mp4
|
|
#
?
Jun 28, 2023 02:07
|
|
|
null_pointer posted:Amazing. What software and prompt? "1988 ektachrome photograph of a compact station wagon with vaporwave aesthetic with pink blue and white, parked in front of a convenience store with the beach in the background --q 2 --v 5 --ar 16:9"  Gave me this slightly cropped version, perfect time to try out that new zoom out feature, did it at 1.5:  While we're on the color subject:                EDIT lmao this one is literally "1976 ektachrome photo of buff lifeguard, dumb, vaporwave aesthetics with pink, blue and white --ar 4:3"  You'll shoot your eye out kid KakerMix fucked around with this message at 07:06 on Jun 28, 2023 |
|
#
?
Jun 28, 2023 06:35
|
|
|
So a new text->video & video->video model came out recently called "zeroscope", in two flavors: zeroscope_v2_576w and zeroscope_v2_XL. Looks like the only (?) intended use case right now is with this Auto1111 extension. It's pretty clunky use and the the XL model can have steep VRAM requirements (note: uses an impossible qty of memory unless you're running with xformers), but the output's pretty promising! Before:  After: https://i.imgur.com/DNbv5zt.mp4 (Album w/ params) Before:  After: https://i.imgur.com/IrSmxu3.mp4 (Album) Super fun to play with, if kinda unpredictable. There's some chatter about managing to make longer videos by using the extension's image inpainter drive the video gen off the last frame of a previous video, but all that managed to make for me was wild scribbly lines and artifacts 
|
|
#
?
Jun 28, 2023 10:00
|
|
|
Earwicker posted:comparing two methods of animated stable diffusion. same prompts, same seeds. 100 frames, 25 steps each. the top one took about 4x longer than the bottom one. Whatever you did with that top one works really well. Having an AI animation as a time lapse is an interesting idea, as it overcomes a lot of the viewer's issues with lack of consistency between frames.
|
|
#
?
Jun 28, 2023 10:41
|
|
|
KakerMix posted:MidJounrney, but in this case it's the original version 5, rather than the new hotness that is 5.2. I was testing to see how the different models handle car wheels since that is a 'tell' with AI images, and I sure do love my cars I really like these!
|
|
#
?
Jun 28, 2023 12:08
|
|
|
This is kind of off topic as it's a video, but: https://www.youtube.com/watch?v=8Z3tiVPzMtU I know it's based on a very slightly older meme, but remaking it with the combination of pirates and fashion seems somehow even more appropriate than Harry Potter and fashion. Using AI to I assume additionally blend the animation of the mouths and to help with the voice lines really impressed me. Obviously you would need to be able to edit video, mix audio streams, produce the AI images (With the help of prompts that might be already available to be fair), generate additional audio lines in the voice of the actor, animate the mouths with some software that matches it up to an audio track (unless you bother to learn classical mouth animation instead?). It feels both high effort and poo poo-post level. I know about some of these tools and how easy they are to use, but am I so out of date that producing a work of this quality can take less than say, 10 hours? I'm surprised it's not the more successful meme, but I have no idea how things go viral anyway.
|
|
#
?
Jun 28, 2023 16:22
|
|
|
|
| # ? May 30, 2024 19:01 |
|
|
         
|
|
#
?
Jun 28, 2023 17:47
|
|
