Skip to content

Releases: LightDiffusion/LightDiffusion-Next

V1.5.4

02 Mar 21:10
Compare
Choose a tag to compare

Change Log :

  • Changed from standard CFG to CFP++ implementations of euler and euler-ancestral, improving quality of generated images

Full Changelog: V1.5.3...V1.5.4

V1.5.3

02 Mar 20:04
Compare
Choose a tag to compare

Change Log :

  • Opimitzed parts of the sampling pipeline to reducee redundancy
  • Optimized calculations made in samplers and CFG leading to a ~6% speed improvement
  • Changed from standard CFG to CFP++ implementations improving quality of generated images

Full Changelog: V1.5.2...V1.5.3

V1.5.2

15 Feb 15:05
d0b8e7d
Compare
Choose a tag to compare

Change Log :

  • Added missing file for seed history

Full Changelog: V1.5.1...V1.5.2

V1.5.1

15 Feb 15:01
ec48a7a
Compare
Choose a tag to compare

Change Log :

  • Fixed reusing seed only taking seed=0
  • Fixed image saving overwriting after first image after subdirectory addition

Full Changelog: V1.5.0...V1.5.1

V1.5.0

14 Feb 19:45
Compare
Choose a tag to compare

Change Log :

  • Added Automatic HDR for better contrasts and lights in output image

Full Changelog: V1.4.1...V1.5.0

V1.4.1

10 Feb 22:47
Compare
Choose a tag to compare

Change Log :

  • Updated pytorch and xformers versions to work with cuda 12.6 and went from pytorch 2.5.1 to pytorch 2.6.0
  • Minor bug fixes leading to pytorch compiled without cuda on windows
  • Minor bug fixes leading to faulty stable fast install on linux

Full Changelog: V1.4.0...V1.4.1

V1.4.0

04 Feb 16:51
Compare
Choose a tag to compare

Change Log :

  • Added optional WebUI mode using gradio
  • Deployed to huggingface space at https://huggingface.co/spaces/Aatricks/LightDiffusion-Next
  • Added GPU detection for better automated installation scripts
  • Fixed a bug leading to crashes when running only on CPU
  • Fixed a bug leading to crashes when calculating attention with Flux

Full Changelog: V1.3.0...V1.4.0

V1.3.0

03 Feb 17:10
Compare
Choose a tag to compare

Change Log :

  • Fixed Stable-fast model compilation
  • Added stable-fast installation on UNIX shell scripts
  • Added dpmpp sde sampler by default for improved quality and set dpmpp 2m as prefer speed checkbox
  • Changed Flux guidance score
  • Switched to ClipTokenizerFast from ClipTokenizer
  • Changed from llama3.2 to deepseek-r1
  • Added dreamfusion SD1.5 download for choice between anime and realistic
  • Various bug fixes

Full Changelog: V1.2.0...V1.3.0

V1.2.0

24 Jan 20:16
Compare
Choose a tag to compare

Change Log :

  • Add Wavespeed Dynamic Caching optimization for improved inference speed up to 2x when using Flux
  • Updated linux shell scripts

Full Changelog: V1.1.0...V1.2.0

V1.1.0

22 Jan 23:14
Compare
Choose a tag to compare

Change Log :

  • changed model loading to have SD1.5 before Flux
  • added compiler
  • added documentation for code related to Flux