3 ms·
...of all the possible formats, it outputs.. a powerpoint presentation..? What.
by Llamamoe 10mo ago
...of all the possible formats, it outputs.. a powerpoint presentation..? What.
- djfobbz 10mo agoLol, right?!?! I would've expected sequential PNGs followed by SVGs once the model improved.
- CamperBob2 10mo agoThat's what the example code at https://old.reddit.com/r/StableDiffusion/comments/1pqnghp/qwenimagelayered_released_on_huggingface/nuw349x/ https://old.reddit.com/r/StableDiffusion/comments/1pqnghp/qw... generates. You get 0.png, 1.png ... n.png, where n= the requested number of layers-1. It'll drop a 600W RTX 6000 to its knees for about a minute, but it does work.
- dvrp 10mo agoI saw some people at a company called Pruna AI got it down to 8 seconds with Cloudflare/Replicate, but I don't know if it was on consumer hardware or an A100/H100/H200, and I don't know if the inference optimization is open-source yet.
- dragonwriter 10mo agoThe github repo includes (among other things) a script (relying on python-pptx) to output decomposed layer images into a pptx file “where you can edit and move these layers flexibly.” (I've never user Powerpoint for this, but maybe it is good enough for this and ubiquitous enough that this is sensible?)