I used AI, but only for the graphics.
Also in the end I didn’t need to train the AI, I just used prompts for everything instead - a lot of it is learning what the AI understands in terms of styles and locations.
So some art styles will get good results on some locations but not on others. Eg. in a general landscape painting style, it will do things like forests and cottages amazingly, but will struggle to make a room inside an office look good.
Yeah, I have Stable Diffusion on my computer, which helps to learn what the AI understands, so then if you use something like Leonardo.ai you’ve done most of the trial-and-error already and can use their enhanced settings to get a more polished version.
The examples you made look good!
Also when you make them 320x200 and they become pixelated, it usually does really nice things to the colours.
I used that model for the character sprites and the items…
No, this is afterwards, in Affinity Photo (similar to Photoshop), just resizing the images to that lower resolution (which is what makes them pixelated). Even with the pixel art model in Leonardo, still have to resize them afterwards to be the correct size.
There is another site now that specializes in pixel art, but I haven’t tried it yet -
Oh, I get it. I tried that in the past, and also tried to reduce the amount of colors after/before resizing. Compared to actual pixel art, the results were mixed, at least for the attempts I made.
I’m sure I saw some time ago an AI-trained pixel art image converter that was able to capture all the nuances of human-made pixel art, but I’ve never tried it.
@Paul : I played your game! Congratulations on creating a fun little pirate journey, I’ll never get tired of them. Your sense of humor has always clicked with me and it was a pleasure to experience it again in the lines of the dialog.
Regarding the graphics, and in the spirit of the topic of this thread, I think your result is a great “proof of concept” of how AI-generated elements can be helpful to artists and creators.
I didn’t have a problem with pixel hunting. I found one hotspot by accident (the crack in the wall), and in general the backgrounds have a tendency to blend everything together, but I think if the interactive objects stand up well enough, old-school users shouldn’t have a problem with that. Still, a hotspot highlighter option might be nice for everyone.
I agree with @seguso that it would make sense for you to pursue other similar projects!
As for the aspects that I think could be improved:
I would say that the story and characters were good for a small production, but I would expect something more defined for a commercial product, regardless of its length.
Knowing where the graphics come from, I am a bit biased and feel that the Leonardo technology is not “there yet” to adequately assist graphic artists who need to create rich adventure game backgrounds. Adventure game backgrounds are often full of elements, sometimes important for the story or puzzles, and at the moment I think this can only be achieved by defining in detail what elements an environment needs to have to serve the story.
I think we’re still in the early days of AI-generated images, and I have no doubt that in the future the technology will be able to empower artists who need to create more detailed images, like DALL-E 3 seems to do.
I can’t wait to see what other projects you have planned. Congratulations again for this first game!
The art is impressive - but something is … hm … “wrong” … in these pictures. If you look at the first one (“BrSw3”) the foreground with the chairs seems do be a 3D cell-shading scene. On the other hand the background seems to be painted and the sea looks like a scaled pixel graphic. The more I look at it, the more I’m confident that the styles won’t fit together - at last in my brain.
Some of them are gorgeous, I love the Santorini street corner.
I think the close-ups are spectacular, even better suited to the backgrounds than the 3D-looking characters Revolution used in “The Serpent’s Curse” and is about to use in the upcoming Parzival’s Stone.
In the first there’s also the lit lantern (in bright daylight) hanging from a branch. At least it has a power line connected!
Similar in the second: there’s a cable ending in the sky. Or maybe it extends to the moon (if that’s what the light blue disc behind the clouds is supposed to be). And of course, the obligatory lantern fruit on the tree on the left hand side.
That actually gives me an idea for a new type of puzzle for newspapers/magazines: instead of “find the 10 differences between those two pictures”, they could do a “find the 10 AI mistakes in that single picture” and use the saved space for an extra ad.
Like @Paul did for his adventure. Modern AI is able to produce such graphics with the right prompt. But note that these are still images, not animations.
I feel like that would be the really tough part in making a game in Broken Sword’s style, it’s pretty heavy on animation from what I remember? I haven’t played it in ages.
The guy said he used Midjourney specifically, and said he didn’t use the name of any specific artist as a reference and I don’t think Midjourney can be trained on images you upload.
So Midjourney might actually know “in the style of Broken Sword” or he might have researched what art styles look like that and created prompts from that info.
Available at all good retailers everywhere!! (ok, maybe just Itch)
Yes, part 5 had a lot of animations. The first two(?) games had great animated cut scenes, but not much in-game-animations AFAIR.
The internet offers a huge heap of Broken Sword pictures, so I’m not surprised that Midjourney knows the style and the series. (A lot of the “big” AIs are trained with the art found by Google.)
I don’t know, they don’t seem to know the style of e.g. Suske & Wiske/Willy Vandersteen/Tintin/Hergé/Asterix so I don’t really see how it’d know a comparatively niche videogame.
Some pictures are sorted out before they are used in the training, plus the AI filters the text entered at the prompt. So maybe the AI blocks the word “Asterix” or the developers have removed all Asterix pictures from the training data.
I assume they blocked Asterix pictures in one of these ways because he is much more popular than Broken Sword and the publisher/copyright holder has more lawyers than Revolution…
Modern AI is able to produce such graphics with the right prompt. But note that these are still images, not animations.