Spaces:

jbilcke-hf

/

ai-comic-factory

like 8.46k

Running on CPU Upgrade

hi, currently there aren't enough servers to generate 8 images at once, so it had to be disabled

I'm currently exploring various solutions to enable it back in a way that doesn't double server latency or server costs (such as faster models, and deliberate/manual page generation like on RunwayML when we extend a video), but it will takes a few weeks or months as I am all alone on the comic factory, and I'm working on other AI projects at the same time

hi, currently there aren't enough servers to generate 8 images at once, so it had to be disabled

I'm currently exploring various solutions to enable it back in a way that doesn't double server latency or server costs (such as faster models, and deliberate/manual page generation like on RunwayML when we extend a video), but it will takes a few weeks or months as I am all alone on the comic factory, and I'm working on other AI projects at the same time

Is it possible to fork the project to have two official forks to help with some of the issues?
There has to be some aspects of latency from people trying to work out their ideas to see which prompts work towards their ideas but without needing even a full panel.
I say this because even with other stable diffusion projects, I get wildly different quality in image generation for the same prompt for SD projects that lack a story component and just give outputs to cram all the word ideas into a single image.
When I first began using your space, I went through many many generations just sculpting what I wanted to get and to target my prompt accordingly, generating many panels that became essentially useless and unnecessary generations. Obviously that would contribute to latency issues especially if you iterate it over many users or many scenes and settings that you would encounter in comic books.

Anyway the official forks could be like,
Fork #1 Space:
Say 1 Space that has a story of only two panels but where each panel is it's own separate story component (instead of two panels at a time).
That way users can test out prompts before going for a full story without overloading that system with excess panels that are superfluous for their initial need to get a good prompt pipeline.

Thought I'd offer that idea as it seems it might be easier to fork it as a "comic prompt tester fork",
---> alter page output variable to 2 panels
---> alter the variable for panels per story component to 1 panel instead of 2

Fork #2 Space:
Then in the 2nd space (essentially the same as the space is now), you could have the traditional 1 page of 4 panels, 2 panels at a time for the story, with the later addition of the manual page generation buttons.

Maybe that's a Band-Aid solution for some of the issues that people are complaining about while waiting for the more labor and time intensive coding and testing to add the buttons and associated variable.

Do you have any reading materials that some of us that are coding newbs could look into to educate ourselves and offer proper technical assistance to help cut down on your work load?
(I know hardware and basics of software/OS - high school physics teacher, but background with coding sucks; until AI arts like this never had a need to get into code)

Keep up the good work. Love what you're doing especially keeping it with the open source ethos.