Code

Stable Diffusion 3 support#16030

Code

Merged

AUTOMATIC1111 merged 24 commits intodevAUTOMATIC1111/stable-diffusion-webui:devfrom sd3AUTOMATIC1111/stable-diffusion-webui:sd3Copy head branch name to clipboardJul 6, 2024

+2,058-72

Commits

Commits on Jun 16, 2024

• add SD3 reference implementation from https://github.com/mcmonkey4eva/sd3-ref/

  mcmonkey4eva

  committed
  a7116aa

  Copy full SHA for a7116aa

• initial SD3 support

  AUTOMATIC1111

  committed
  5b2a60b

  Copy full SHA for 5b2a60b

• linter

  AUTOMATIC1111

  committed
  79de09c

  Copy full SHA for 79de09c

• typo

  AUTOMATIC1111

  committed
  7ee2114

  Copy full SHA for 7ee2114

• prevent accidental creation of CLIP models in float32 type when user wants float16

  AUTOMATIC1111

  committed
  b443fdc

  Copy full SHA for b443fdc

• add protobuf==3.20.0 to requirements

  AUTOMATIC1111

  committed
  80f618e

  Copy full SHA for 80f618e

• fix NaN issue when running without --precision half

  AUTOMATIC1111

  committed
  06d0a5a

  Copy full SHA for 06d0a5a

• change CLIP links to allow anonymous downloading

  AUTOMATIC1111

  committed
  58dc35a

  Copy full SHA for 58dc35a

• change t5xxl checkpoint to fp8

  AUTOMATIC1111

  committed
  d4b814a

  Copy full SHA for d4b814a

• add an option (on by default) to disable T5

  Show description for 34b4443
  AUTOMATIC1111

  committed
  34b4443

  Copy full SHA for 34b4443

Commits on Jun 24, 2024

• fix T5

  AUTOMATIC1111

  committed
  a65dd31

  Copy full SHA for a65dd31

• medvram support for SD3

  AUTOMATIC1111

  committed
  a8fba9a

  Copy full SHA for a8fba9a

Commits on Jun 26, 2024

• support for SD3: infinite prompt length, token counting

  AUTOMATIC1111

  committed
  d686e73

  Copy full SHA for d686e73

Commits on Jun 27, 2024

• fix mdevram for SD1/SDXL

  AUTOMATIC1111

  committed
  42ca30d

  Copy full SHA for 42ca30d

Commits on Jun 28, 2024

• get deepbooru to run with --precision-half

  AUTOMATIC1111

  committed
  06fe174

  Copy full SHA for 06fe174

• get T5 to work both with and without --precision half

  AUTOMATIC1111

  committed
  fc8b126

  Copy full SHA for fc8b126

• add the missing get_first_stage_encoding function

  AUTOMATIC1111

  committed
  0c7bdcc

  Copy full SHA for 0c7bdcc

• fix img2img

  AUTOMATIC1111

  committed
  0b64633

  Copy full SHA for 0b64633

• fix the problem with infinite prompts where empty cond would be calculated incorrectly

  AUTOMATIC1111

  committed
  179ae47

  Copy full SHA for 179ae47

• allow generation to be started with any dimensions specified

  AUTOMATIC1111

  committed
  d67348a

  Copy full SHA for d67348a

• support loading clip/t5 from the main model checkpoint

  AUTOMATIC1111

  committed
  7e4b06f

  Copy full SHA for 7e4b06f

Commits on Jun 29, 2024

• do sampler calculations on CPU

  AUTOMATIC1111

  committed
  1394eca

  Copy full SHA for 1394eca

• remove AutocastLinear from SD3's MLP

  AUTOMATIC1111

  committed
  ebe8be9

  Copy full SHA for ebe8be9

Commits on Jun 30, 2024

• fix --medvram

  AUTOMATIC1111

  committed
  9e404c3

  Copy full SHA for 9e404c3