Hey mates, recently I've developed a tool to use the GenerativeAI on the AI Horde to created random avatars and banners on lemmy. To keep things spicy, I wanted to deploy to rotate the /c/piracy banner daily, as I've done in a bunch of other communities like [email protected] and the lemmy.dbzer0.com.
So now this is now active!
In case you're curious,. the current prompt used is this:
"A of a (solo:1.2) {cyberpunk|steampunk|solarpunk|dieselpunk|anarchist|fantasy|science fiction} {disney pixar 3D|realistic|dreamworks 3D animation} {male|female|androgynous} {Abkhazian|Afghan|Åland Island|Albanian|Algerian|American Samoan|Andorran|Angolan|Anguillan|Antarctic|Antiguan|Argentine|Armenian|Aruban|Australian|Austrian|Azerbaijani|Bahamian|Bahraini|Bangladeshi|Barbadian|Belarusian Belgian|Belizean|Beninois|Bermudian|Bhutanese|Bolivian|Bonaire|Herzegovinian|Botswana|Bouvet Island|Brazilian|BIOT|Bruneian|Bulgarian|Burkinabé|Burundian|Cabo Verdean|Cambodian|Cameroonian|Canadian|Caymanian|Central African|Chadian|Chilean|Chinese|Christmas Island|Cocos Island|Colombian|Comorian|Congolese|Congolese|Cook Island|Costa Rican|Croatian|Cuban|Curaçaoan|Cypriot|Czech|Danish|Djiboutian|Dominican|Dominican|Timorese|Ecuadorian|Egyptian|Salvadoran|English|Equatorial Guinean|Eritrean|Estonian|Swati|Ethiopian|European|Falkland Island|Faroese|Fijian|Finnish|French|French Guianese|French Polynesian|French Southern Territories|Gabonese|Gambian|Georgian|German|Ghanaian|Gibraltar|Greek|Greenland|Grenadian|Guadeloupe|Guamanian|Guatemalan|Guernsey|Guinean|Bissau-Guinean|Guyanese|Haitian|Heard Island|Honduran|Cantonese|Magyar|Icelandic|Indian|Indonesian|Iranian|Iraqi|Irish|Manx|Israelite|Italian|Ivorian|Jamaican|Jan Mayen|Japanese|Jersey|Jordanian|Kazakhstani|Kenyan|Kiribati|North Korean|South Korean|Kosovan|Kuwaiti|Kyrgyzstani|Laotian|Latvian|Lebanese|Basotho|Liberian|Libyan|Liechtensteiner|Lithuanian|Luxembourgish|Macanese|Madagascan|Malawian|Malaysian|Maldivian|Malinese|Maltese|Marshallese|Martiniquais|Mauritanian|Mauritian|Mahoran|Mexican|Micronesian|Moldovan|Monégasque|Mongolian|Montenegrin|Montserratian|Moroccan|Mozambican|Myanma Burmese|Namibian|Nauruan|Nepalese|Dutch|New Caledonian|New Zealand|Nicaraguan|Nigerien|Nigerian|Niuean|Norfolk Island|Macedonian|Northern Irish|Northern Marianan|Norwegian|Omani|Pakistani|Palauan|Palestinian|Panamanian|Papuan|Paraguayan|Peruvian|Filipino|Pitcairn Island|Polish|Portuguese|Puerto Rican|Qatari|Réunionnais|Romanian|Russian|Rwandan|Saban|Barthélemois|Saint Helenian|Kittitian|Saint Lucian|Saint-Martinoise|Miquelonnais|Vincentian|Samoan|Sammarinese|São Toméan|Saudi|Scottish|Senegalese|Serbian|Seychellois|Sierra Leonean|Singaporean|Sint EustatiusStatian|Sint Maarten|Slovak|Slovenian|Solomon Island|Somali|South African|South Georgia Island|South Ossetian|South Sudanese|Spanish|Sri Lankan|Sudanese|Surinamese|Svalbard|Swedish|Swiss|Syrian|Taiwanese|Tajikistani|Tanzanian|Thai|Timorese|Togolese|Tokelauan|Tongan|Trinidadian|Tunisian|Turkish|Turkmen|Turks and Caicos Island|Tuvaluan|Ugandan|Ukrainian|Emirati|British|American|Uruguayan|Uzbekistani|Ni-Vanuatu|Vaticanian|Venezuelan|Vietnamese|British Virgin Island|U.S. Virgin Island|Welsh|Wallis and Futuna|Sahrawi|Yemeni|Zambian|Zanzibari|Zimbabwean} pirate {standing in front of a sail|sitting in front of a monitor}, {Rococo|Digital Art|Baroque} style, wearing a {red|blue|yellow|green|black|white} {bandana|hat|scarf} and holding a {crimson|gold|silver|emerald|magic|azure|obsidian|ebony} {sword|keyboard|compass|mug|treasure|flag|club|amulet|fruit}, rich detailed {open|cloudy|stormy|sunny} sky background###multiple people, duo, cleavage, bitcoin"
Pretty long ye? The way this works is that each a random option is picked from each collection wrapped in { } which allows me to generate versatile options every day to keep things fresh
Here's some sample images that will be created
If you have improvements you want to see on the prompt, do let me know in the comments. In the future I'm having thoughts of allowing the community here to automatically send adjustments ideas to the prompt for funsies.
And the process of setting that up is creative?
Hate to break it to you, but programming is creative plus I don't like making visual art manually.
"A conservative estmate of 800 watts"? Where the fuck are you getting these number mate? My GPU at full power is not even half that.
I do automation for a living, I'm well aware that computer engineering can be a creative process, but it is not an artistic one.
I don't like seeing visual art automated.
I don't know what's in your computer, but some high-end graphics cards can pull more than 300W at full load all on their own. 400W total power output seems a pretty conservative estimate to me.
Great? I find the automation creative.
I don't care?
Most high-end graphic cards actually use close to 100W when not in demanding workloads. Mine certainly does.
You said double that. Most people's PCs idle at 200-W300W as a total. And even if 400W was the total PC usage, as I said, my PC is already on.
If you didn't, then you wouldn't be in this thread defending your creative choices. You could have ignored me like a person who doesn't care would do.
As per Tom's Hardware, the average high-end card comes in somewhere north of 250W. But even if we half the total power estimate again we're still talking about an eigth of a watt-hour, or 50,000 times more than a 4mb download, per image.
I revised my estimate while you were replying.
My PC idles at about 75W, and tops out just shy of 600W with all cores, cards, and drives running.
I am autistic. I actually hate that my brain is forcing me to reply to this drivel.
Great, so a generated image would be done at 2-10 seconds at like 200W usage. Anyway, this is a stupid argument. Even if it's not as an mp3, it's like a movie, what-the-fuck-ever.
I am also autistic, but I enjoy a good argument and will never pass up the chance to complain when people put generated images on my screen. There's more real art out there in the world to see than could be seen in a lifetime, so why should anyone waste time with computer-generated approximations of art?
We are comparing apples to oranges here, but the fact remains that serving existing content will always be cheaper than serving content one has to generate first.
And generated content can be randomized and automated, which you can't do with existing content, which is what I wanted to do
Sure you can, you just need a sufficiently diverse database of images at which to point your random number generator. No generation necessary, just aggregate human artistic output.
That's orders of magnitude harder in a bad non automated than what I'm going now for worse results and more risk. It's a frankly terrible idea
Randomly generated content is usually a bad idea on public-facing websites. Generating and displaying it automatically means you only get to inspect the results after the fact, so you're relying on the image generator to never produce something unacceptable.
Having a DB of prepared images means you can pick through and remove all the unacceptable ones before they have a chance to be displayed.
The ai horde already takes care of any problematic stuff. This is all solved