Sign In

SD-Turbo

Updated: Dec 4, 2023

base model

Download

1 variant available

fp32 SafeTensor

sd_turbo.safetensors

Full precision, largest file • 4.86 GB

Verified:

Type
Checkpoint Trained
Stats

1,484

Reviews
Published

Dec 4, 2023

Base Model

SD 2.1

Hash
AutoV2
3F067A1B94
default creator card background decoration
Followers - 91

91

Likes - 193

193

output_tile.jpg

Yes, this is the SD(2.1, not quite 1.5) version of SDXL-turbo. Dont know why it hasnt been uploaded before now. But the license seems to allow non-comercial resitribution, so.. here it is!

LIcense at https://huggingface.co/stabilityai/sd-turbo/blob/main/LICENSE
Raw original file at https://huggingface.co/stabilityai/sd-turbo/resolve/main/sd_turbo.safetensors?download=true

Set to CFG=1 STEPS=1.
Sometimes you get nicer results with STEPS=3 or 4. Sometimes you dont.

YOU MAY HAVE TO MANUALLY SET RESOLUTION TO 512x512, if your tool defaults to 1024x1024

Note: like SDXL-turbo, it is bad at humans by default. You have to bang on it. See examples.

Its not blindingly fast on my old laptop.. but it DOES generate an image in a single step instead of 20.
So that does make it "fastER"