Stable Diffusion 3 support #16030
add SD3 reference implementation from https://github.com/mcmonkey4eva…
a7116aa9
initial SD3 support
5b2a60b8
linter
79de09c3
typo
7ee2114c
prevent accidental creation of CLIP models in float32 type when user …
b443fdcf
add protobuf==3.20.0 to requirements
80f618ea
fix NaN issue when running without --precision half
06d0a5ab
change CLIP links to allow anonymous downloading
58dc35a6
change t5xxl checkpoint to fp8
d4b814ae
add an option (on by default) to disable T5
34b4443c
fix T5
a65dd315
medvram support for SD3
a8fba9af
support for SD3: infinite prompt length, token counting
d686e73d
fix mdevram for SD1/SDXL
42ca30d6
get deepbooru to run with --precision-half
06fe174c
get T5 to work both with and without --precision half
fc8b1266
add the missing get_first_stage_encoding function
0c7bdcc1
fix img2img
0b646335
fix the problem with infinite prompts where empty cond would be calcu…
179ae47d
allow generation to be started with any dimensions specified
d67348a0
support loading clip/t5 from the main model checkpoint
7e4b06fc
do sampler calculations on CPU
1394ecaf
remove AutocastLinear from SD3's MLP
ebe8be90
fix --medvram
9e404c31
Assignees
No one assigned
Login to write a write a comment.
Login via GitHub