Jedrzej Kosinski
|
4aee1498cf
|
Merge PR #330 from Kosinkadink/develop - SD LoRA masking and scheduling
Add SD LoRA masking, scheduling, and conditioning helpers.
|
2024-04-30 09:18:35 -05:00 |
|
Jedrzej Kosinski
|
0a7ebbf7bd
|
Made all Combine LoRA Hook nodes act as a passthrough if only one hook provided instead of throwing an error
|
2024-04-30 08:55:26 -05:00 |
|
Jedrzej Kosinski
|
fccdb43ea7
|
Add sampler_post_cfg_function support for lora hooks
|
2024-04-30 02:18:03 -05:00 |
|
Jedrzej Kosinski
|
ba66cc1e15
|
Switched order of a couple nodes
|
2024-04-29 11:13:05 -05:00 |
|
Jedrzej Kosinski
|
910475c7b2
|
Added LoRA Hook Keyframe support for scheduling any LoRA Hook, whether it be individual or group
|
2024-04-29 11:01:25 -05:00 |
|
Jedrzej Kosinski
|
5da9bfda92
|
Register Model as LoRA Hook (with clip, not the Model Only version) now works properly thanks to the previous refactor of Model as LoRA Hook code
|
2024-04-28 11:56:42 -05:00 |
|
Jedrzej Kosinski
|
78880ad398
|
Fixed Model as Hooked LoRA from allowing model_sampling keys to be patches
|
2024-04-28 11:08:29 -05:00 |
|
Jedrzej Kosinski
|
33d8a98d77
|
Refactored Registered Model As LoRA Hook to work as a normal lora via weight diff instead of hooked_replace, completely removed hooked_replace code
|
2024-04-28 10:43:01 -05:00 |
|
Jedrzej Kosinski
|
b1893ca36c
|
Fixed default_conds behavior at CFG=1.0, start of LoRA scheduling work
|
2024-04-28 06:09:28 -05:00 |
|
Jedrzej Kosinski
|
005d49a0b2
|
Moved LoraHook stuff into conditioning.py from utilts_motion.py
|
2024-04-28 05:06:15 -05:00 |
|
Jedrzej Kosinski
|
df81a94c84
|
More cleanup, prepared scaffolding for tracking which modules use lowvram weight_function and bias_function
|
2024-04-27 18:23:02 -05:00 |
|
Jedrzej Kosinski
|
aba69c0e55
|
Cleaned up hooked patching code
|
2024-04-27 13:14:47 -05:00 |
|
Jedrzej Kosinski
|
2dd285052d
|
Fixed lowvram I accidentally broke during cleanup
|
2024-04-26 18:03:03 -05:00 |
|
Jedrzej Kosinski
|
92b7b8c0d6
|
Small cleanup of code
|
2024-04-26 17:47:04 -05:00 |
|
Jedrzej Kosinski
|
667042b43e
|
Initial lowvram support for lora hooks
|
2024-04-26 17:24:19 -05:00 |
|
Jedrzej Kosinski
|
2b6c1bf475
|
Added support for Timesteps Conditioning, masks are now optional (so the other parts of the nodes can be used without masks), renaming and reorganizing nodes
|
2024-04-26 10:27:10 -05:00 |
|
Jedrzej Kosinski
|
eaf58bc460
|
Fixed sliding context - accidentally indented the combining code when refactored sliding_calc_conds_batch, making the function return immediately after the first context is sampled
|
2024-04-26 07:41:43 -05:00 |
|
Jedrzej Kosinski
|
44a097d803
|
Added masked conditioning nodes to make masking/combining/applying hooks much easier, including Set Unmasked Conds with supporting code that makes specific conds apply to the areas where no masks were applied automatically
|
2024-04-25 19:08:11 -05:00 |
|
Jedrzej Kosinski
|
89b33022b6
|
Refactored sampling code to fully mirror calc_conds_batch, fixed only the top layer (last) Control object being checked if it is compatible with sliding context
|
2024-04-24 07:12:30 -05:00 |
|
Jedrzej Kosinski
|
5c86862431
|
Merge branch 'main' into develop
|
2024-04-24 03:00:37 -05:00 |
|
Jedrzej Kosinski
|
c3f61c6790
|
Merge PR #357 from Kosinkadink/pag_support - PerturbedAttentionGuidance support w/ context
Support sliding contexts for PerturbedAttentionGuidance (and sampler_post_cfg_function)
|
2024-04-23 06:34:27 -05:00 |
|
Jedrzej Kosinski
|
adb296bb81
|
Added sliding context support for cfg_function so that PerturbedAttentionGuidance works as expected
|
2024-04-23 06:22:14 -05:00 |
|
Jedrzej Kosinski
|
fa08393388
|
Fixed git merging mistake that happened at some point
|
2024-04-23 03:03:33 -05:00 |
|
Jedrzej Kosinski
|
d94273b4c9
|
Merge branch 'main' into develop
|
2024-04-23 02:39:01 -05:00 |
|
Jedrzej Kosinski
|
56b6f9d299
|
Merge PR #356 from Kosinkadink/attn_fallback
Added fallback mechanism for attention to work around pytorch's CUDA limit for scaled_dot_product_attention
|
2024-04-23 02:36:09 -05:00 |
|
Jedrzej Kosinski
|
6ca21f2bb6
|
Added fallback mechanism for attention to work around pytorch's CUDA compiled limit for scaled_dot_product_attention
|
2024-04-23 02:22:10 -05:00 |
|
Jedrzej Kosinski
|
1fde1be80e
|
Merge PR #355 - instructpix2pix models whitelisted for AD
Treat SD15/SDXL instructpix2pix models as compatible SD15/SDXL models
|
2024-04-23 01:25:54 -05:00 |
|
Jedrzej Kosinski
|
7c798ab3e1
|
Treat SD15/SDXL instructpix2pix models as compatible SD15/SDXL models, respectively
|
2024-04-23 01:22:56 -05:00 |
|
Jedrzej Kosinski
|
ded3b7673d
|
Merge branch 'main' into develop
|
2024-04-22 22:01:04 -05:00 |
|
Jedrzej Kosinski
|
4c1ff8467e
|
Merge PR #354 from Kosinkadink/cameractrl - CameraCtrl support
CameraCtrl Support
|
2024-04-22 14:06:31 -05:00 |
|
Jedrzej Kosinski
|
48ced743d1
|
Update README.md - CameraCtrl support
|
2024-04-22 13:59:57 -05:00 |
|
Jedrzej Kosinski
|
078db70d7e
|
Added node to control original_pose_width/height. In original implementation, this was supposed to match up to the source video used to extract features, not sure what should be used for artificially-generated coordinates. Future me's problem.
|
2024-04-22 13:47:31 -05:00 |
|
Jedrzej Kosinski
|
eb8e7c0b7a
|
Reworked nodes_cameractrl.py to use CAMERACTRL_POSES inputs and outputs for everything
|
2024-04-22 10:51:07 -05:00 |
|
Jedrzej Kosinski
|
a97a4e2e4e
|
Use internal view_options for CameraPoseEncoder to overcome 16 max context_length, made rearrange_hidden_shapes irrelevant by figuring out the proper rearrange inside CameraPoseEncoder's forward
|
2024-04-21 06:54:59 -05:00 |
|
Jedrzej Kosinski
|
078465a98e
|
Added AnimateDiff+CameraCtrl Keyframe node to allow scheduling of cameractrl_multival in addition to the normal AD multivals, moved pose nodes into submenu in CameraCtrl
|
2024-04-21 05:09:56 -05:00 |
|
Jedrzej Kosinski
|
2bdf695853
|
Merge branch 'cameractrl' of https://github.com/Kosinkadink/ComfyUI-AnimateDiff into cameractrl
|
2024-04-21 04:44:52 -05:00 |
|
Jedrzej Kosinski
|
171b1f46fa
|
Added cameractrl_multival to Apply AnimateDiff+CameraCtrl Model nodel
|
2024-04-21 04:38:52 -05:00 |
|
Jedrzej Kosinski
|
dffbf24452
|
Merge pull request #352 from chaojie/cameractrl
camera poses nodes
|
2024-04-21 04:33:17 -05:00 |
|
fles
|
1cbf898083
|
camera poses nodes
|
2024-04-21 17:13:04 +08:00 |
|
Jedrzej Kosinski
|
bda8b6aad8
|
Moved CameraCtrl and AnimateLCM-I2V nodes into their own files
|
2024-04-21 02:48:10 -05:00 |
|
Jedrzej Kosinski
|
eef203568c
|
Fix sub_idxs not being used if orig_camera_entries has the same length as full_length
|
2024-04-21 01:55:10 -05:00 |
|
Jedrzej Kosinski
|
8c8a9ca67c
|
Commented out some test log statements
|
2024-04-19 13:20:31 -05:00 |
|
Jedrzej Kosinski
|
9d2d4e747b
|
Got CameraCtrl working, fixed weird edge case causing AD device mismatches after switching models + prompts
|
2024-04-19 13:13:33 -05:00 |
|
Jedrzej Kosinski
|
45493cf8a9
|
Almost got CameraCtrl working, just need to compare with running diffusers impl side by side now to debug
|
2024-04-08 19:01:19 -05:00 |
|
Jedrzej Kosinski
|
11dac28eac
|
More progress, plucker_embeddings created now
|
2024-04-08 06:08:39 -05:00 |
|
Jedrzej Kosinski
|
dad9d18589
|
Merge branch 'main' into cameractrl
|
2024-04-07 18:16:55 -05:00 |
|
Jedrzej Kosinski
|
f9e0343f4c
|
Merge PR #337 from Kosinkadink/freeinit_samplersigma_fix
Make FreeInit [sampler sigma] work with new ComfyUI versions
|
2024-04-06 17:32:15 -05:00 |
|
Jedrzej Kosinski
|
93af734521
|
Make FreeInit [sampler sigma] work with new ComfyUI versions
|
2024-04-06 17:30:44 -05:00 |
|
Jedrzej Kosinski
|
6219ea438d
|
More progress on CameraCtrl
|
2024-04-06 16:37:46 -05:00 |
|
Jedrzej Kosinski
|
bc1bb6d630
|
Merge branch 'main' into cameractrl
|
2024-04-05 00:48:53 -05:00 |
|