Today, we open-source PrunaVAED, a faster drop-in replacement decoder for video generation with LTX-2.3! - 𝗛𝗼𝘄 𝗲𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝘁 𝗶𝘀 𝗶𝘁? ~1.7-2.1x faster decoder & ~50% lower peak VRAM - 𝗛𝗼𝘄 𝗴𝗼𝗼𝗱 𝗶𝘀 𝗶𝘁? Near-original visual quality & no changes to the latent encoding - 𝗛𝗼𝘄 𝗱𝗼𝗲𝘀 𝗶𝘁 𝘄𝗼𝗿𝗸? We pruned,... trained & distilled video decoders to maximize efficiency of the LTX-2.3 series from Lightricks Want to try the fast open-source integration? Check it here on Diffusers: Want to try the fastest video generation endpoint? Check p-video here on API:show more

where is the comfyui workflow?

please guys release diffusion version for comfyui. this is diffusers.

Thanks for creating this! Really cool

🔥🔥

The collab we’ve been waiting for

nicee 🔥

Normally I wouldn't care to save a few seconds on the decode, but doing a 1280x704 res 25 second clip does take some time with the default VAE. Shaving that almost in half could be nice. Plus with less VRAM maybe I could push to 1080p 25 second clips on my RTX 3060 without OOM.

Ltx 2.5 is out hope you make another fast vae for it, thanks

is just download, replace?

Noice 👌

lets get this in @ComfyUI

@wildmindai might be something for you?

Half the VRAM with barely any quality loss is a great trade for video gen.

Huge win for open-source video AI. Faster decoding + lower VRAM without sacrificing quality is exactly what builders need. 🚀

Please opensource some of your research on speeding up image inference

Faster LTX is not the problem. A better -and useful- LTX is the problem that's not solved.

Any weights for Diffusers?

Interesting! So using it coupled with an int8 model should be very fast!