Sana

Sana copied to clipboard

Even faster Sana Sprint?

2

Hi, wondering if you have a 0.3B or even smaller version of SANA sprint that could be run continuously with decent FPS for real-time systems. Do you have experiments with...

johannes-stelzer

1. update readme; 2. update CI test code;

lawrence-cj

Image Editting via Inversion

9

Hi, please correct me if I'm wrong. I tried using the inverse function in [DPM-Solver](https://github.com/NVlabs/Sana/blob/47777fbdb245e8584ba829caea3d2326c13b2b50/diffusion/model/dpm_solver.py#L549) to invert the source latent to the noisy latent. After obtaining the noisy latent, I...

KhoiDOO

Hello, thank you for the great work! Is it possible to provide us with the Dockerfile for setting up the training environment? I have seen the dockerfile for running inference...

Raman1121

video generation

2

Hi authors, Thanks for the brilliant wotk! Do you have any plans on publishing video generation sana models?

weimengting

Add train_sana_sprint_diffusers file

5

Initial implementation of SANA-Sprint training script adapted for Diffusers. This needs further refinement and optimization. @lawrence-cj @sayakpaul

scxue

licencing of the generated images

1

hi, can we use the generated images safely for commercial purpose , for example in video games ? thanks .

issam1975

OOM when finetuning Sana-1.6M on 4K images — memory requirement & tiled VAE?

1

Hi, I'm finetuning Sana-1.6M on 4K images (4096×4096) and encountered OOM on an H20-96G GPU during vae_encode. Mixed precision is enabled, batch size = 1. My questions: 1.How much memory...

Twinkle-ce

Minor Typo Error

1

Great work! For your full table benchmark, the emoji for FID seems to be inverted.

WeizhenWang-1210

Large intermediate value in linear attention

2

I tried your linear attention module and found that, in attn_matmul, `vk` has extremely large values especially when the sequence is long. I guess it is because your relu kernel...

thuliu-yt16