Upgrade your account to a Community Supporter account and remove most of the site ads.

Reply to thread

Message: <blockquote data-quote="Jfdlsjfd" data-source="post: 9715996" data-attributes="member: 42856">(side discussion before I resume doing the spell list)For those who like AI image but are reluctant to use models trained on copyrighted artworks, the works to train model on licensed art is progressing.<ul> <li data-xf-list-type="ul">Firefly (inside Adobe's suite) is now, in arena matches, where people are asked to tell which image is better when confronted to the same prompt, is on par with Flux Schnell, the lightweight edition of open-source models you can run on your own computer.</li> <li data-xf-list-type="ul">If you prefer models you can run at home, Bria 3.2 is having SDXL-like performance, which is where the best performing model was 8 months ago.</li> </ul>It is far from perfect and below current models of course, but if one wants to create tokens for their online games without regret for uncompensated authors, it might be interesting.I have played with it a little:A warforged with a leather belt and a leather satchel on his torso[ATTACH=full]412562[/ATTACH]A wizard with a fireball over his palm:[ATTACH=full]412563[/ATTACH](the 4 fingered-hand was a classic blunder back then and the improvement comes from a different training, not from using more images)Two space marines in front of a glass tube holding a green fluorescent liquid.[ATTACH=full]412579[/ATTACH]A floating castle in a gloomy forest, in front of the moon[ATTACH=full]412564[/ATTACH] An orange large industrial truck with large wheels and a futuristic design, the driver is standing next to the vehicle, parked in a desert environment next to some black metal structures.[ATTACH=full]412566[/ATTACH]A 1920 british gentleman studying over a tome in his library. Besides him, a brain floats in a jar.[ATTACH=full]412567[/ATTACH]a spaceship flying in front of Jupiter[ATTACH=full]412569[/ATTACH]a middle-aged woman sitting on a bench in a public park, wearing red dress, pink hat, white shoes. A dog sits at her feet.[ATTACH=full]412568[/ATTACH]While these images will be a throwback from what is recently presented on the thread, it's because it's a 4 billion-parameters models, while current models -- which are more costly to train -- are 12 to 18 billions parameters. The nice thing with small models is that on current harware it takes 5 seconds to generate an image, so you can easily get a bunch of them, choose the closer to your need and correct the details. Maybe there will be interest for this line of models, whose main attraction, for those concerned, is that it can be used guilt-free (and for everyone, it dismisses the myth that AI images need to be trained on copyrighted work to get anywhere).</blockquote>

Verification