Commit Graph

5596 Commits

Author SHA1 Message Date
4eb7a5fc60 (ci) clean up pip tests 2023-06-26 13:08:43 -04:00
587203d589 (tests) make fixture reusable; support boards
fixes the test suite generally, but some tests needed to be
skipped/xfailed due to recent refactor

- ignore three test suites that broke following the model manager
  refactor
- move InvocationServices fixture to conftest.py
- add `boards` InvocationServices to the fixture
2023-06-26 13:08:34 -04:00
e3f136cdda Update 060_INSTALL_PATCHMATCH.md
installing the packaged 'blas' is needed in Archlinux, otherwise patchmatch fails initializing with a "libblas.so.3 missing" error.
2023-06-26 14:23:10 +02:00
af566adf56 For MediapipeFace ControlNet preprocessor, if input image is RGBA format then convert to RGB (otherwise MediapipeFace image processing throws an error) 2023-06-26 04:29:43 -07:00
873c18bc4b Added TileResampler ControlNet preprocessor node.
Also fixes to SegmentAnything ControlNet preprocessor node.
2023-06-26 04:27:26 -07:00
d905d0e42a feat(ui): only show canvas image fallback on loading error (#3589) 2023-06-26 21:40:10 +12:00
6ccf62a863 feat(ui): only show canvas image fallback on loading error 2023-06-26 19:20:05 +10:00
6390af229d feat(ui): add dynamic prompts to t2i tab
- add param accordion for dynamic prompts
- update graphs
2023-06-26 19:15:54 +10:00
47e651225d query for 'main' model type when populating UI lists
to support renaming of 'pipeline' models to 'main'
2023-06-26 01:39:46 -04:00
9cfac4175f feat(ui): improved node parsing (#3584)
- use `swagger-parser` to dereference openapi schema
- tidy vite plugins
- use mantine select for node add menu
2023-06-26 17:38:23 +12:00
3a19be1606 fix: Add missing IAIMantineSelect disabled styles 2023-06-26 17:37:47 +12:00
b51ab056f2 Merge branch 'main' into feat/ui/update-node-parsing 2023-06-26 17:32:44 +12:00
e206fad22a fix(ui): fix controlnet image size (#3585) 2023-06-26 17:32:07 +12:00
7b97639961 Merge branch 'main' into lstein/installer-for-new-model-layout 2023-06-26 01:24:30 -04:00
60780e990d fix(ui): fix controlnet image size 2023-06-26 12:03:11 +10:00
8d43cf92f6 feat(ui): update action santizer for schema actions 2023-06-26 12:00:38 +10:00
862bf7546c feat(ui): improved node parsing
- use `swagger-parser` to dereference openapi schema
- tidy vite plugins
- use mantine select for node add menu
2023-06-26 11:53:54 +10:00
91c3a58fb6 Fix lycoris layers init 2023-06-26 04:33:37 +03:00
5cebf67ee4 Apply lora by patching lora instead of hooks 2023-06-26 03:57:33 +03:00
1ba94a92b3 Fixes 2023-06-26 03:54:42 +03:00
23c22ac933 Refactor logic/small fixes 2023-06-26 03:07:54 +03:00
160b5d7992 add support for an autoimport models directory scanned at startup time 2023-06-25 18:50:15 -04:00
10e8389fa4 Commenting out LeReS ControlNet image preprocessor until release of controlnet_aux v0.0.6 (supported on controlnet_aux current main, but not on latest release v0.0.5) 2023-06-25 14:25:14 -07:00
45aa338a98 Changed pyproject.toml to require controlnet_aux >= 0.0.5 (to enable use of SAM ControlNet preprocessor) 2023-06-25 14:22:34 -07:00
414a04774c Added LeReS ControlNet image preprocessor. 2023-06-25 14:19:55 -07:00
c91d1eacba Merge branch 'lstein/installer-for-new-model-layout' of github.com:invoke-ai/InvokeAI into lstein/installer-for-new-model-layout 2023-06-25 16:04:48 -04:00
60b37b7ff4 fix model manager documentation 2023-06-25 16:04:43 -04:00
b872e7a5e0 Simplifying ControlNet SAM preprocessor segmentation color mapping. 2023-06-25 12:54:48 -07:00
de4064bdac Fixed problem with with non-reproducible results from ControlNet SegmentAnything preprocessor. Cause was controlnet_aux randomization of segmentation coloring, which seems to lead to some randomization of resulting images using ControlNet seg model. Switched to using deterministic ADE20K color palette instead, which solved the problem. 2023-06-25 12:38:17 -07:00
10c3753d7f Added SAM preprocessor 2023-06-25 11:16:39 -07:00
a3c22b5fe6 Remove upcast_attention and prediction_type from stable diffusion model logic, fix ckpt conversion according to this 2023-06-25 21:06:22 +03:00
922468b836 Add control_mode parameter to ControlNet (#3535)
This PR adds the "control_mode" option to ControlNet implementation. 
Possible control_mode options are: 

- balanced -- this is the default, same as previous implementation
without control_mode
- more_prompt -- pays more attention to the prompt
- more _control -- pays more attention to the ControlNet (in earlier
implementations this was called "guess_mode")
- unbalanced -- pays even more attention to the ControlNet 

balanced, more_prompt, and more_control should be nearly identical to
the equivalent options in the [auto1111 sd-webui-controlnet
extension](https://github.com/Mikubill/sd-webui-controlnet#more-control-modes-previously-called-guess-mode)

The changes to enable balanced, more_prompt, and more_control are
managed deeper in the code by two booleans, "soft_injection" and
"cfg_injection". The three control mode options in sd-webui-controlnet
map to these booleans like:
 
!soft_injection && !cfg_injection ⇒  BALANCED            
 soft_injection &&  cfg_injection ⇒  MORE_CONTROL 
 soft_injection && !cfg_injection ⇒  MORE_PROMPT   
 
The "unbalanced" option simply exposes the fourth possible combination
of these two booleans:
!soft_injection &&  cfg_injection ⇒ UNBALANCED

With "unbalanced" mode it is very easy to overdrive the controlnet
inputs. It's recommended to use a cfg_scale between 2 and 4 to mitigate
this, along with lowering controlnet weight and possibly lowering "end
step percent". With those caveats, "unbalanced" can yield interesting
results.

Example of all four modes using Canny edge detection ControlNet with
prompt "old man", identical params except for control_mode:

![Screenshot from 2023-06-11
23-53-00](https://github.com/invoke-ai/InvokeAI/assets/303100/c9e31e7f-50de-4d85-94f2-b5a4af3d067b)
Top middle:       BALANCED
Top right:          MORE_CONTROL
Bottom middle: MORE_PROMPT
Bottom right :    UNBALANCED

I kind of chose this seed because it shows pretty rough results with
BALANCED (the default), but in my opinion better results with both
MORE_CONTROL and MORE_PROMPT. And you can definitely see how MORE_PROMPT
pays more attention to the prompt, and MORE_CONTROL pays more attention
to the control image. And shows that UNBALANCED with default cfg_scale
etc is unusable.

But here are four examples from same series (same seed etc), all have
control_mode = UNBALANCED but now cfg_scale is set to 3.
![Screenshot from 2023-06-11
23-48-44](https://github.com/invoke-ai/InvokeAI/assets/303100/5a495306-2164-40aa-9cc8-ce737d7671e7)
And param differences are:
Top middle: prompt="old man", control_weight=0.3, end_step_percent=0.5
Top right: prompt="old man", control_weight=0.4, end_step_percent=1.0
Bottom middle: prompt=None, control_weight=0.3, end_step_percent=0.5
Bottom right: prompt=None, control_weight=0.4, end_step_percent=1.0

So with the right settings UNBALANCED seems useful.
2023-06-25 16:09:26 +12:00
57e719702d fix(ui): add missing ControlNetInvocation type; tidy schema-derived types 2023-06-25 14:04:53 +10:00
11378a9236 chore(ui): regen api schema 2023-06-25 14:04:16 +10:00
132829c88f fix(ui): fix path of generated schema types 2023-06-25 14:04:00 +10:00
4d4b5b56dc Merge branch 'main' into feat/controlnet-control-modes 2023-06-25 15:48:07 +12:00
a9334128c9 chore(ui): bump all packages (#3579)
Everything seems to be working.

- Due to a change to `reactflow`, I regenerated `yarn.lock`
- New chakra CLI fixes issue I had made a patch for; removed the patch
- Change to fontsource changed how we import that font
- Change to fontawesome means we lost the txt2img tab icon, just chose a
similar one
2023-06-25 15:45:39 +12:00
6b276587d8 chore(ui): bump all packages
Everything seems to be working.

- Due to a change to `reactflow`, I regenerated `yarn.lock`
- New chakra CLI fixes issue I had made a patch for; removed the patch
- Change to fontsource changed how we import that font
- Change to fontawesome means we lost the txt2img tab icon, just chose a similar one
2023-06-25 13:44:10 +10:00
c5faffc18b Merge branch 'main' of github.com:invoke-ai/InvokeAI into feat/controlnet-control-modes
Only "real" conflicts were in:
     invokeai/frontend/web/src/features/controlNet/components/ControlNet.tsx
     invokeai/frontend/web/src/features/controlNet/store/controlNetSlice.ts
2023-06-24 17:05:57 -07:00
c3c4a71173 implemented Stalker's suggested improvements 2023-06-24 12:37:26 -04:00
d5f742620f Merge branch 'main' into lstein/installer-for-new-model-layout 2023-06-24 11:58:06 -04:00
ba1371a88f rename ModelType.Pipeline to ModelType.Main 2023-06-24 11:45:49 -04:00
3ae996ebcb fix(ui): fix metadata viewer too stronk 2023-06-24 18:15:49 +10:00
3d16605762 fix(ui): fix controlnet upload button 2023-06-24 18:15:49 +10:00
b6dec2b826 fix(ui): fix controlnet dnd overlay not showing on dragover 2023-06-24 18:15:49 +10:00
013e2aa2a1 fix(ui): fix control image sizes
they were all weird
2023-06-24 18:15:49 +10:00
8f9fa15fc8 fix(ui): fix image fetching query string 2023-06-24 18:15:49 +10:00
dde497404b fix(ui): fix init image display buttons
- Reset and Upload buttons along top of initial image
- Also had to mess around with the control net & DnD image stuff after changing the styles
- Abstract image upload logic into hook - does not handle native HTML drag and drop upload - only the button click upload
2023-06-24 18:15:49 +10:00
0472b33164 fix(ui): fix duplicate is_intermediate query param when fetching images 2023-06-24 17:57:39 +10:00
a6c615a98c fix(ui): fix canvas staging area
Missed some of the `imageUpdated` stuff
2023-06-24 17:57:39 +10:00