brkirch
e3b53fd295
Add UI setting for upcasting attention to float32
...
Adds "Upcast cross attention layer to float32" option in Stable Diffusion settings. This allows for generating images using SD 2.1 models without --no-half or xFormers.
In order to make upcasting cross attention layer optimizations possible it is necessary to indent several sections of code in sd_hijack_optimizations.py so that a context manager can be used to disable autocast. Also, even though Stable Diffusion (and Diffusers) only upcast q and k, unfortunately my findings were that most of the cross attention layer optimizations could not function unless v is upcast also.
2023-01-25 01:13:04 -05:00
brkirch
84d9ce30cb
Add option for float32 sampling with float16 UNet
...
This also handles type casting so that ROCm and MPS torch devices work correctly without --no-half. One cast is required for deepbooru in deepbooru_model.py, some explicit casting is required for img2img and inpainting. depth_model can't be converted to float16 or it won't work correctly on some systems (it's known to have issues on MPS) so in sd_models.py model.depth_model is removed for model.half().
2023-01-25 01:13:02 -05:00
InvincibleDude
44c0e6b993
Merge branch 'AUTOMATIC1111:master' into master
2023-01-24 15:44:09 +03:00
brkirch
f64af77adc
Fix different first gen with Approx NN previews
...
The loading of the model for approx nn live previews can change the internal state of PyTorch, resulting in a different image. This can be avoided by preloading the approx nn model in advance.
2023-01-23 22:49:20 -05:00
invincibledude
3bc8ee998d
Gen params paste improvement
2023-01-22 16:35:42 +03:00
invincibledude
7f62300f7d
Gen params paste improvement
2023-01-22 16:29:08 +03:00
invincibledude
a5c2b5ed89
UI and PNG info improvements
2023-01-22 15:50:20 +03:00
invincibledude
bbb1e35ea2
UI and PNG info improvements
2023-01-22 15:44:59 +03:00
invincibledude
34f6d66742
hr conditioning
2023-01-22 15:32:47 +03:00
invincibledude
125d5c8d96
hr conditioning
2023-01-22 15:31:11 +03:00
invincibledude
2ab2bce74d
hr conditioning
2023-01-22 15:28:38 +03:00
invincibledude
c5d4c87c02
hr conditioning
2023-01-22 15:17:43 +03:00
invincibledude
4e0cf7d4ed
hr conditioning
2023-01-22 15:15:08 +03:00
invincibledude
a9f0e7d536
hr conditioning
2023-01-22 15:12:00 +03:00
invincibledude
f774a8d24e
Hr-fix separate prompt experimentation
2023-01-22 14:52:01 +03:00
invincibledude
81e0723d65
Logging for debugging
2023-01-22 14:41:41 +03:00
invincibledude
b331ca784a
Fix
2023-01-22 14:35:34 +03:00
invincibledude
8114959e7e
Hr separate prompt test
2023-01-22 14:28:53 +03:00
invincibledude
0f6862ef30
PLMS edge-case handling fix 5
2023-01-22 00:11:05 +03:00
invincibledude
6cd7bf9f86
PLMS edge-case handling fix 3
2023-01-22 00:08:58 +03:00
invincibledude
3ffe2e768b
PLMS edge-case handling fix 2
2023-01-22 00:07:46 +03:00
invincibledude
9e1f49c4e5
PLMS edge-case handling fix
2023-01-22 00:03:16 +03:00
AUTOMATIC
78f59a4e01
enable compact view for train tab
...
prevent previews from ruining hypernetwork training
2023-01-22 00:02:51 +03:00
invincibledude
6c0566f937
Type mismatch fix
2023-01-21 23:25:36 +03:00
invincibledude
3bd898b6ce
First test of different sampler for hi-res fix
2023-01-21 23:14:59 +03:00
AUTOMATIC
3deea34135
extract extra network data from prompt earlier
2023-01-21 19:36:08 +03:00
AUTOMATIC
92fb1096db
make it so that extra networks are not removed from infotext
2023-01-21 16:41:25 +03:00
AUTOMATIC
40ff6db532
extra networks UI
...
rework of hypernets: rather than via settings, hypernets are added directly to prompt as <hypernet:name:weight>
2023-01-21 08:36:07 +03:00
AUTOMATIC1111
a8322ad75b
Merge pull request #6854 from EllangoK/master
...
Saves Extra Generation Parameters to params.txt
2023-01-18 23:25:56 +03:00
AUTOMATIC
b186d44dcd
use DDIM in hires fix is the sampler is PLMS
2023-01-18 23:20:23 +03:00
EllangoK
5e15a0b422
Changed params.txt save to after manual init call
2023-01-17 11:42:44 -05:00
AUTOMATIC
e0e8005009
make StableDiffusionProcessing class not hold a reference to shared.sd_model object
2023-01-16 23:09:08 +03:00
AUTOMATIC
9991967f40
Add a check and explanation for tensor with all NaNs.
2023-01-16 22:59:46 +03:00
AUTOMATIC
f9ac3352cb
change hypernets to use sha256 hashes
2023-01-14 10:25:37 +03:00
space-nuko
88416ab5ff
Fix extension parameters not being saved to last used parameters
2023-01-12 13:46:59 -08:00
AUTOMATIC
d4fd2418ef
add an option to use old hiresfix width/height behavior
...
add a visual effect to inactive hires fix elements
2023-01-09 14:57:47 +03:00
noodleanon
50e2536279
Merge branch 'AUTOMATIC1111:master' into img2img-api-scripts
2023-01-07 14:18:09 +00:00
AUTOMATIC
1a5b86ad65
rework hires fix preview for #6437 : movie it to where it takes less place, make it actually account for all relevant sliders and calculate dimensions correctly
2023-01-07 09:56:37 +03:00
noodleanon
b5253f0dab
allow img2img api to run scripts
2023-01-05 21:21:48 +00:00
AUTOMATIC
847f869c67
experimental optimization
2023-01-05 21:00:52 +03:00
AUTOMATIC
2e30997450
move sd_model assignment to the place where we change the sd_model
2023-01-05 10:21:17 +03:00
Philpax
83ca8dd0c9
Merge branch 'AUTOMATIC1111:master' into fix-sd-arch-switch-in-override-settings
2023-01-05 05:00:58 +01:00
AUTOMATIC
99b67cff0b
make hires fix not do anything if the user chooses the second pass resolution to be the same as first pass resolution
2023-01-05 01:25:52 +03:00
AUTOMATIC
bc43293c64
fix incorrect display/calculation for number of steps for hires fix in progress bars
2023-01-04 23:56:43 +03:00
AUTOMATIC
8149078094
added the option to specify target resolution with possibility of truncating for hires fix; also sampling steps
2023-01-04 22:04:40 +03:00
AUTOMATIC
097a90b88b
add XY plot parameters to grid image and do not add them to individual images
2023-01-04 19:19:11 +03:00
AUTOMATIC
525cea9245
use shared function from processing for creating dummy mask when training inpainting model
2023-01-04 17:58:07 +03:00
AUTOMATIC
4d66bf2c0d
add infotext to "-before-highres-fix" images
2023-01-04 17:24:46 +03:00
AUTOMATIC1111
6281c1bdb4
Merge pull request #6299 from stysmmaker/feat/latent-upscale-modes
...
Add more latent upscale modes
2023-01-04 13:47:36 +03:00
MMaker
15fd0b8bc4
Update processing.py
2023-01-04 05:12:54 -05:00
