The following concept art comes directly from the current pipeline, but we haven't yet gotten it fully stable for animation. It's way closer than it was, but in the mean time, here are some frame grabs from the AI layers in progress.
cgi Save Point: Concept Art
- Thread starter Nate North
- Start date
There are so many of these from this weeks tests that I can't even show even 1 in 100. I've installed a new stage that allows the AI layer to read directly from our existing Mpg frames, and it saves a huge amount of time. The next step for animation is training a custom model calibrated for our input layer UE5, and a more focused output layer.
Thanks. I am the artist on all these frames, and lol, about a billion more.
It's a paradigm shift situation, which was what the whole Save Point Idea was about in the first place. I'm not the artist of the paintings exactly, I'm the artist who drew the artist robot that drew the paintings. It's still a work in progress, but I'm way closer than I was even 3 months ago.
For the last few years, I've been building what's called "a pipeline" which in this case is mostly a very long chain of programs and processes, which once fully calibrated and automated will allow an exponential leap in creative output at a viable quality, first for lower markets like animation, then after further development, for photorealistic film at fractions of a cent on the dollar.
Most of the videos you've seen over the last year were what I call the "ghost layer" a way for me to have total control over a film set, and consistency from any angle. The CGI layer you've been seeing is a bit ugly for broadcast, but allows this control, the AI layer on the other end of the pipeline looks great, but has no control. So this whole thing is about connecting and hybriding these two technologies. Then you have total creative control with a broadcast ready look.
That answer probably isn't super clear. In all of the above pictures, I first fed the system an image I created, a lengthy description of the imagery I was envisioning, tweaked many variables, arranged modular processors, swapped out code components that kind of thing. Once the pipeline is outputting the type of image I want to get out of input X, I can print out an unlimited number of variations on that artistic mentality. It would be accurate to say that each theme and style is an artwork, taking a lot of time and effort to design, but images within those groups are simply derivations of the prime creation.
It's a paradigm shift situation, which was what the whole Save Point Idea was about in the first place. I'm not the artist of the paintings exactly, I'm the artist who drew the artist robot that drew the paintings. It's still a work in progress, but I'm way closer than I was even 3 months ago.
For the last few years, I've been building what's called "a pipeline" which in this case is mostly a very long chain of programs and processes, which once fully calibrated and automated will allow an exponential leap in creative output at a viable quality, first for lower markets like animation, then after further development, for photorealistic film at fractions of a cent on the dollar.
Most of the videos you've seen over the last year were what I call the "ghost layer" a way for me to have total control over a film set, and consistency from any angle. The CGI layer you've been seeing is a bit ugly for broadcast, but allows this control, the AI layer on the other end of the pipeline looks great, but has no control. So this whole thing is about connecting and hybriding these two technologies. Then you have total creative control with a broadcast ready look.
That answer probably isn't super clear. In all of the above pictures, I first fed the system an image I created, a lengthy description of the imagery I was envisioning, tweaked many variables, arranged modular processors, swapped out code components that kind of thing. Once the pipeline is outputting the type of image I want to get out of input X, I can print out an unlimited number of variations on that artistic mentality. It would be accurate to say that each theme and style is an artwork, taking a lot of time and effort to design, but images within those groups are simply derivations of the prime creation.
Last edited:
The ninth frame, showing two figures watching lightning among trees, seems to show one of the figures as Darth Vader. In any case, they look good, but there should be a story around them.
There's a story in place that links all these images together. It's just a very large scale story with branching plotlines. In extremely short, tagline form, a man living a normal life on earth gains the ability to revert to previous time with his memory intact. There are many opportunities for the viewer to use this to their advantage, which quickly propels them (via the protagonist) into unexpected lives. While the narrative branches over time, a central plotline revolves around the discovery that earth is only one of a seemingly endless number of civilizations, and a lot of people have known this and been travelling between the planets for decades, unbeknownst to the general population. Once off planet the viewer can pursue numerous adventures connected by a single galactic scale mystery, the resolution of which forms the final act. A lot of it will probably play out similar to Star Trek, with episodic adventures interconnected by overarching plots and world building.
It's important to note that my plotlines are only intended to represent a fraction of the possible stories as the narrative branches. The design is such that a viewer who shows disinterest in sci fi plotlines can make a series of decisions that put them into a storyline they are interested in. They just have to be built one at a time, and I'm still working on the technology and workflow needed to render the first one. Getting a lot closer though.
It's all intentional, the winged astronauts are from a recurring dream the protagonist has that provides subtle clues to solving the endgame. I like weird surrealism, but it will be confined mostly to short dream sequences that happen rarely.
Not much on the screen is happening by accident compared to the old days (dec 2022), with many of these having both an image and a description of 75-225 words in addition to running through multiple stacked and modded models.
The Horsetronaut was in our earliest concept art, and it was just some dumb throwaway idea like many others, but I got a lot of response from that particular image, so I kept it as a dream element though it wont play a role in the actual story.
Not much on the screen is happening by accident compared to the old days (dec 2022), with many of these having both an image and a description of 75-225 words in addition to running through multiple stacked and modded models.
The Horsetronaut was in our earliest concept art, and it was just some dumb throwaway idea like many others, but I got a lot of response from that particular image, so I kept it as a dream element though it wont play a role in the actual story.
As we develop new techniques in the various subprocesses that contribute to the art layer matrix, it becomes possible to create more complex styles that are more detailed and consistent. You may sort of recognize some of the images below, as many are previously seen SP frames from last years demo work, run through this particular style implementation I developed.
Attachments
I was thinking, would clouds form and move the same way in a different atmosphere? Probably not. Seems like each planet would have it's own variations of cloud behavior. I have this set up to store all the information about a frame's concept inside each frame. What that means is that once a particular visual concept is achieved, every output frame using that concept contains the "DNA" that created it. I don't have to save and organize these thousands of ideas and formulas, because they are literally burned into every frame of film across Save Point. I can drop any frame into a reader function and create an infinite amount of that concept, style, item, etc. So if I make a good weeping willow tree genome, and it makes a tree I think is ideal, I can just make an infinite line of trees based around that "Archetype file"
