This is a merge I threw together for a side-project for amateur filmmakers to generate storyboards for short films. Turned out pretty well. It's not a generalist model, but rather a merge for a very specific task. Based on the very excellent Scribbler, merged with SDXL Lightning 4 step LoRA and a couple of sketch-based LoRA's. Credit for the base model generators who did such an excellent job.
I find DPMPP+ SDE SGM Uniform at 5 steps, CFG 1 gives the nicest and quickest results, perfect for storyboarding. Olde skool, using sketches for storyboards, I know, with all these new AI tools available, but some people just like using drawings to visualise their film ideas...Hope people find it useful.
Description
I use it for a local install of an app to generate story boards.
FAQ
Comments (4)
This looks wonderful, thank you for sharing it!
If you could upload it without lightning merged it, users could apply different accelerators and control the ratio of speed/quality loss.
I very much enjoy using the PCM models as of recent and find different configurations useful for different tasks, where more stylized or glitched models are actually not even showing any noticeable deterioration to more drastic means of acceleration.
For general use, the 8-step variants of PCM in both low- and normal-cfg versions are usually my go-to, in case you want to take a look. But it goes as far down as 2-steps and works on the same level as TCD I would say, in case you need more speed for some quick story-boarding.
Either way, I would appreciate a non-accelerated model, if it's not too much of a hassle. Thank you!
Hi. I used another model, linked in the description that would suit your needs better, in that case. It’s their work that’s made this possible, I just merged the accelerator and made it available with a LoRA I created for my other model…All credit goes to the original creator…
@FiditeNemini I see. Yeah I got that one already and thought yours would be substantially different. It's always difficult to tell how much was really merged into a model until you try it. Thank you very much either way!
Thank you.