2022 Coffee Shop and Gardens

[rrkknight3]

Good morning, PDG, Fafnir and McG.

PDG, check out https://github.com/krea-ai/open-prompts

[parkdalegardener]

There is a lot going on in this space at the moment Sir Knight. Thanks for the link.

Now I do need to say this in general about the text to image thing. I am in a bit of a quandary ATM regarding it. I have no issue with the generation in and of itself. I completely understand the scrape and have maintained for years that data scraping is being used a lot more than people may understand. I used to mine Google when training ML/AI with OpenCV myself. They put a limit in the API now limit that type of thing to folks that don't pay them.

I have two issues. My first problem is with the way people blend styles of existing or well recognized artists to flavour the result; using them by name. I am not sure who's art is cheapened by that type of thing, but it makes me a little uneasy. An example would be "A naked Vicky in a temple with a sword in the style of Frank Frazzeta and Boris Vallejo, trending on Artstation" Don't know what it would generate on a first pass without tweeking; but it wouldn't feel right to me. I don't do that, and I find that has become the favourite way of operation for many.

The other thing is speed. Stuff that would take me days to make and an hour or more to render happens in seconds. If I was illustrating some of the stuff I generate it would take ages to finish to the level of detail these generators produce. Still, I haven't used actual physical media, like charcoal or oils, in years; and can do it quicker on the computer myself so I that argument may not carry a lot of weight.

Now; one the plus side, I use it like photoshop on meth. Screaming fast and highly unstable running head long into madness. I find it quite addictive to artfully craft a prompt with a pretty good idea what I am going to get out of it depending upon the sampling method used and the generation parameters tweeked.

I've worked a lot with computer vision so I know how these diffusion GANS function. I understand why a prompt will generate or compose an image in a particular way based upon the sampling criteria.

For a guy that always hated doing post work and usually didn't; I find myself strangely addicted to this. The worst thing is it is post work. Post work in the highest degree.

In-painting to replace a part of the image that needs work. Putting a hat on a character that never had one to start with.

Out-painting to replace an environment a character is currently in.

The notorious img2img used for fine tuning a diffuser and creating "Deep Fakes" amongs't just general image manipulation and trippy morphing video.

It's all post work if you actually look at it. Otherwise it's just parroting a prompt and hoping for the best random seed result.

Again; thanks. I'll digest this over lunch.

Good morning, PDG, Fafnir and McG.

PDG, check out https://github.com/krea-ai/open-prompts

[parkdalegardener]

I'm liking this database Sir Knight. Good find.

[Agent0013]

"Text to Image". Hmmm. Is this about NFTs?
From what I gather, one uses text prompts to define what a project will have, and then allows to computer to do all the actual work putting the project together. Am I right here? Or is this something else entirely?
When I get to work on a project, say a piece of art I want to do, I don't want the actual work composing it and either painting, drawing, sculpting, or working in CGI taken away from me. I do it because I enjoy doing the work myself. The more difficult a project is, the more I enjoy doing it, as it challenges me, and I love a challenge. Challenges keep the mind sharp.

[Aelin]

Good afternoon!
I have no envy to put W11 on any comp, that's enough it tried since the first days it was launched to install itself on my W8.1. Between that and the fact preinstalled McAfee was acting like a virus, I don't wonder too much why my hard drive failed after 2 years...

McG, your trick was one thing I found online in the first week I obtained the W10 comp. Otherwise, it was useless for me in all time to work properly, even only writting. It was fully using the fans over and over. Once the comp was with a lot removed of useless stuff, suddenly it was almost the silence and fans are working only when it's necessary for them to do their true job.
I also go each day in services.msc to stop Windows Update.
And no possible update when I'm not online with comp, because I simply turn it off.
Good luck to you with the W11.

[parkdalegardener]

"Text to Image". Hmmm. Is this about NFTs?
From what I gather, one uses text prompts to define what a project will have, and then allows to computer to do all the actual work putting the project together. Am I right here? Or is this something else entirely?
When I get to work on a project, say a piece of art I want to do, I don't want the actual work composing it and either painting, drawing, sculpting, or working in CGI taken away from me. I do it because I enjoy doing the work myself. The more difficult a project is, the more I enjoy doing it, as it challenges me, and I love a challenge. Challenges keep the mind sharp.

No nothing to do with nfts.

Text to image is just as it sounds and you surmise.

[parkdalegardener]

Agent look at Wambo, DALL-e, Midjourney, Stability ai, Huggingface; and any other number of others.

[Aelin]

By curiosity looked also for the little Craiyon, or NightCafe.
I still fell i's too bad Midjourney is on discord Not that I want to do art with, but sometimes to create a quick background for our images would be good that way. Rarely I find what I need when I can't do it.

[parkdalegardener]

And you seem to forget that my time upright atm is minimal and motion limited. You forget why I looked at it in the first place.

[parkdalegardener]

By curiosity looked also for the little Craiyon, or NightCafe.
I still fell i's too bad Midjourney is on discord Not that I want to do art with, but sometimes to create a quick background for our images would be good that way. Rarely I find what I need when I can't do it.

Exactly Aelin. It's fun to play with and it is fine as a texture toy when tiling the images or as a background maker. A quick and dirty way to make a lot of artwork for a gallery scene or library shelves full of books. On it's own without modification the resulting images are not ready for prime time, but they are perfect to take into photoshop.