|
AI animated video and voice short. Gonna be interesting to see what can be done a year from now. https://twitter.com/mrjonfinger/status/1645953033636048896 Youtube https://www.youtube.com/watch?v=s62xY_Z1raM
|
#
?
Apr 12, 2023 02:45
|
|
|
|
| # ? May 28, 2024 17:44 |
|
|
Looks much more impressive than that Will Smith eating pasta thing.
|
|
#
?
Apr 12, 2023 05:46
|
|
|
https://voca.ro/131ztgXzB0g7
|
|
#
?
Apr 13, 2023 07:30
|
|
|
Anyone try the Dolly LLM? Just got released by databricks, open source. https://www.databricks.com/blog/2023/04/12/dolly-first-open-commercially-viable-instruction-tuned-llm Edit: apparently a fine tune of a previous model rather than a new model. Still curious to give it a try at some point. wait for it fucked around with this message at 10:13 on Apr 13, 2023 |
|
#
?
Apr 13, 2023 09:58
|
|
|
Been working on upscaling with more detail on really, really huge images
|
|
#
?
Apr 13, 2023 11:55
|
|
|
BARONS CYBER SKULL posted:Been working on upscaling with more detail on really, really huge images do you have the original image or some progression images to compare?
|
|
#
?
Apr 13, 2023 17:34
|
|
|
BARONS CYBER SKULL posted:Been working on upscaling with more detail on really, really huge images Which model are you using for this? Midjourney doesn't really upscale for poo poo anymore.
|
|
#
?
Apr 13, 2023 18:48
|
|
|
LifeSunDeath posted:do you have the original image or some progression images to compare?  Bob Ross Nuke Test posted:Which model are you using for this? Midjourney doesn't really upscale for poo poo anymore. RevAnimated 1.2.1
|
|
#
?
Apr 13, 2023 20:45
|
|
|
BARONS CYBER SKULL posted:original: wow yeah, it added a details like crazy      
|
|
#
?
Apr 13, 2023 21:02
|
|
|
wait for it posted:Anyone try the Dolly LLM? Just got released by databricks, open source. Jeff, Leonard "J." Crabs bot when?
|
|
#
?
Apr 13, 2023 21:04
|
|
|
LifeSunDeath posted:wow yeah, it added a details like crazy basic workflow! generate your base image, with hiresfix as per usual, nothing special then, make sure you have https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111 installed as an extension send your base image to img2img, and mirror your settings exactly (except wildcards if you use those, just paste that image's prompts in), seed and all, EXCEPT denoising, which for adding detail should be around 0.3-0.37 then go to Tiled Diffusion and do this:  swap the tile sizes around if you're doing landscape! then Tiled Vae and do this:  so you should end up something like this:  slam generate, wait a while, then when it's done drag the image from the output over to the input, and do it again, and wait CONSIDERABLY longer then you'd end up with a image that's doubled in size, then doubled again, with enough coherency on the original that it doesn't just become a mass of the same prompt tiled into itself, but also added a ton of detail!
|
|
#
?
Apr 13, 2023 22:11
|
|
|
oh so you did a crazy ivan on the AI and it gave you the old razzle dazzle, v nice.
|
|
#
?
Apr 13, 2023 22:21
|
|
|
|
|
#
?
Apr 14, 2023 01:11
|
|
|
BARONS CYBER SKULL posted:basic workflow! Thank you for this write-up. (Have to link, it's 133MB) Korn.png
|
|
#
?
Apr 14, 2023 04:17
|
|
|
KwegiboHB posted:Thank you for this write-up. very upsetting thank you
|
|
#
?
Apr 14, 2023 04:37
|
|
|
BARONS CYBER SKULL posted:basic workflow! Could this be used on images generated outside of A1111, say with content originally from MidJourney V5?
|
|
#
?
Apr 14, 2023 05:46
|
|
|
Bob Ross Nuke Test posted:Could this be used on images generated outside of A1111, say with content originally from MidJourney V5? This can be used on any image. The model you choose will have a slight to great effect based on the amount of de-noising done. You can use that on purpose for specific effects. Should pass on the tip of making an inpainting model out of any model by way of simple A+(B-C) merge. Base model in A, SD1.5-Inpainting in B, SD1.5 in C. That will give you an impainting model for img2img or upscaling to go with the model for base generations.
|
|
#
?
Apr 14, 2023 06:07
|
|
|
Just some random images I generated while experimenting with the describe function in Midjourney the last couple days                     
|
|
#
?
Apr 14, 2023 07:09
|
|
|
The middle third all look like an online web comic from about a decade ago called "copper" particularly the green one with the boat and the fish
|
|
#
?
Apr 14, 2023 07:22
|
|
|
Bob Ross Nuke Test posted:Could this be used on images generated outside of A1111, say with content originally from MidJourney V5? Yes but it’s not going to be nearly as effective, as running it through with the same prompt and seed is basically forcing it to image2image the original with a new, larger version of the same image with added detail You’d have to run significantly less denoise on a image you don’t know the prompt/model/etc or else SD is just going to hallucinate entirely new images in the tiles it renders, and with less denoise it’s not going to significantly change or add details
|
|
#
?
Apr 14, 2023 08:59
|
|
|
BARONS CYBER SKULL posted:basic workflow! Thanks! Quoting so I can find later, too, let's be honest
|
|
#
?
Apr 14, 2023 12:03
|
|
|
Is this and these quote:
|
|
#
?
Apr 14, 2023 12:37
|
|
|
[Edit: starting to think this is probably fake news Edit Edit: Upon even further inspection it does in fact appear to be real news] AI wins again! https://www.eldagsen.com/sony-world-photography-awards-2023/ The guy won a big Sony photography contest with an image that is quite frankly obviously AI generated. He refused the prize though because it wasn't art created in the medium of the contest (rightly so IMO), and just did it to show people they'll have to look out for this stuff. Now we don't need to have a derail about what is or isn't art or unfair or whatever, but I would like to remark on how I think anyone with experience in AI generation would have had flags going off right away. It's similar to how I can spot ChatGPT's writing style a mile away now. 
Nice Van My Man fucked around with this message at 15:32 on Apr 14, 2023 |
|
#
?
Apr 14, 2023 15:05
|
|
|
Fuschia tude posted:Is this For the bottom 6, the keywords seem to be "in the style of dark blue and red", "Indonesian illustration" and "Konstantin Somov". I basically generated a close approximation to these images and then used "describe" to see what it would generate and what keywords it used. I had personally never heard of Konstantin Somov and looking up his art on Google images, I don't really see the resemblance but hey. If you like what you see here, maybe use his name in the prompt. The ships I'll have to scroll back through Discord and find the prompts but I know woodcut illustration wasn't a part of them.        
|
|
#
?
Apr 14, 2023 15:06
|
|
|
Actually the only place I'm seeing that AI art thing reported is hackernews and that website, so I'm starting to suspect they've been taken in by fake news.
|
|
#
?
Apr 14, 2023 15:17
|
|
|
Nice Van My Man posted:Actually the only place I'm seeing that AI art thing reported is hackernews and that website, so I'm starting to suspect they've been taken in by fake news. I don't see how anyone qualified to judge a photography contest (or anyone with eyes) wouldn't immediately spot that it isn't a photograph. There's so much wrong with it, from the inconsistent depth of field to the weird hands (and the extra hand). Maybe they're getting AIs to judge photography contests now. (Or more likely it's probably a fake story...)
|
|
#
?
Apr 14, 2023 15:23
|
|
|
It's tricky because he says Sony took it down. I wouldn't believe him, but it does appear in this Fox News (sorry) gallery of winners: https://www.foxnews.com/lifestyle/wonderful-world-pictures-jaw-dropping-photos-sony-world-photography-awards-finalists I don't even know what's real anymore! drat you AI! Edit: Oh, it also appears here (https://www.worldphoto.org/sony-world-photography-awards/winners-galleries/2023/open/winners/creative-2023-open-competition) on the web archive. In light of this evidence I declaring my final verdict is that it's real. Nice Van My Man fucked around with this message at 15:33 on Apr 14, 2023 |
|
#
?
Apr 14, 2023 15:25
|
|
|
|
|
#
?
Apr 14, 2023 15:32
|
|
|
WhiteHowler posted:I don't see how anyone qualified to judge a photography contest (or anyone with eyes) wouldn't immediately spot that it isn't a photograph. There's so much wrong with it, from the inconsistent depth of field to the weird hands (and the extra hand). the "inconsistent dof" is not. it's emulating large-format cameras, with their flexible focal fields that the accordion lenses could do funky stuff with. that one looks like a darkroom composite of two large-format portraits; it may have been trained on exactly such photos.
|
|
#
?
Apr 14, 2023 19:36
|
|
|
Are there any (free ideally) online generators with inpainting and/or img2img? I'm using BlueWillow currently which is "okay" but otherwise lacking.
|
|
#
?
Apr 14, 2023 21:03
|
|
|
You should be able to do all that stuff with Stable Horde. There is even an addon to use it from the auto1111 webui so you can play with all the same knobs that offline generator users can.
|
|
#
?
Apr 14, 2023 21:11
|
|
|
Dog Kisser posted:Are there any (free ideally) online generators with inpainting and/or img2img? I'm using BlueWillow currently which is "okay" but otherwise lacking. A stable diffusion Google Colab notebook? You get 4-8 hours a day generally without any issues. Google is very vague and cagey about the hours, complete waste to pay them you still aren't guaranteed anything I'd use a different provider to run a notebook if you are actually going to pay. Google is the best free though.
|
|
#
?
Apr 14, 2023 21:17
|
|
|
pixaal posted:A stable diffusion Google Colab notebook? You get 4-8 hours a day generally without any issues. Google is very vague and cagey about the hours, complete waste to pay them you still aren't guaranteed anything I'd use a different provider to run a notebook if you are actually going to pay. Google is the best free though. I've never even heard of that - can I get a... Link? I don't even know what to ask for hehe
|
|
#
?
Apr 14, 2023 22:03
|
|
|
Paid google colab was cool as hell a few years ago before the AI stuff really took off because it was $10/month and that would give you essentially 24/7 access to at least an nvidia titan  deep dish peat moss fucked around with this message at 22:12 on Apr 14, 2023 |
|
#
?
Apr 14, 2023 22:10
|
|
|
I did a month of the $10 colab pro and it's kind of an odd value proposition. They give you compute credits, but they're still used when you're only using free tier resources. So the only thing that makes sense is to use the beefiest configuration and run your compute credits down to nothing in just a few hours. After you're out of compute credits you're basically limited to the free tier except you can choose a high system ram configuration and run multiple colabs at the same time. They do cut you off after using it for 12 hours in a day though.
|
|
#
?
Apr 15, 2023 00:57
|
|
|
BARONS CYBER SKULL posted:basic workflow! Thanks for the workflow! I got it working for my first upscale (basically 960x540 to 1920x1080) but if I try sending the upscaled image back to img2img and hit generate I run out of memory. I have a 12GB card, is there a way to get this method to work on that? I kept dropping the tile sizes lower and lower but couldn't find something that worked.
|
|
#
?
Apr 15, 2023 05:12
|
|
|
repugnant posted:Thanks for the workflow! I got it working for my first upscale (basically 960x540 to 1920x1080) but if I try sending the upscaled image back to img2img and hit generate I run out of memory. I have a 12GB card, is there a way to get this method to work on that? I kept dropping the tile sizes lower and lower but couldn't find something that worked. drag it over instead of resending resending will change the resolutions, dragging it over just puts the new image in there
|
|
#
?
Apr 15, 2023 05:13
|
|
|
BARONS CYBER SKULL posted:drag it over instead of resending Awesome, that did the trick! Got about 20 minutes left on my first 4K upscale repugnant fucked around with this message at 04:01 on Apr 16, 2023 |
|
#
?
Apr 15, 2023 18:50
|
|
|
Pvt. Parts posted:This right the hell here. Lol at anyone calling training a model on your image "theft". Guess inspiration is theft now. capitalism rewards lovely people and so encourages more people to be lovely also it is by definition a lovely exploitative system, a system is what it does
|
|
#
?
Apr 15, 2023 23:32
|
|
|
|
| # ? May 28, 2024 17:44 |
|
|
Why are we quoting week old posts
|
|
|
#
?
Apr 16, 2023 03:44
|
|
