 Jedrzej Kosinski
|
3d00251fc2
|
modified prepare_mask_batch to work with sd3
|
2024-06-27 19:37:00 -05:00 |
|
|
Jedrzej Kosinski
|
4454c47707
|
Added sd3 support from upcoming ComfyUI update
|
2024-06-27 18:23:10 -05:00 |
|
|
Jedrzej Kosinski
|
96d957e9ec
|
Fixed get_calc_pow to work the same as before (only current difference would be with SDXL T2IAdapter models with custom/soft weights), automatically resize T2IAdapter control tensors to match batch_size to make my life easier
|
2024-06-27 17:43:21 -05:00 |
|
|
Jedrzej Kosinski
|
7e7fdf0b83
|
Also reverse T2IAdapter's get_calc_pow
|
2024-06-27 13:59:43 -05:00 |
|
|
Jedrzej Kosinski
|
4646bb3a2b
|
Reverse T2IAdapter weights to account for ComfyUI changes (means SDXL T2IAdapter softweights have breaking changes due to extra middle weight)
|
2024-06-27 13:51:18 -05:00 |
|
|
Jedrzej Kosinski
|
bc528228c7
|
Make 1 mask input work with T2IAdapter (some ComfyUI optimization caused feature regression at some point)
|
2024-06-27 12:46:06 -05:00 |
|
|
Jedrzej Kosinski
|
59e7d1a7af
|
Make require_vae something for me to work on in the future for SparseCtrl and ReferenceCN
|
2024-06-27 12:24:18 -05:00 |
|
|
Jedrzej Kosinski
|
86e6ed301e
|
Fixed control weights not being applied properly
|
2024-06-27 11:15:03 -05:00 |
|
|
Jedrzej Kosinski
|
e800d52acf
|
Made all controlnets works with upcoming ComfyUI changes, started work on making latent preprocs obsolete, still need to make control weights apply the same way as before
|
2024-06-27 05:03:10 -05:00 |
|
|
Jedrzej Kosinski
|
e6085cc111
|
Change what I previously called latent_format to model_latent_format to avoid conflicts with upcoming ComfyUI changes
|
2024-06-27 00:46:55 -05:00 |
|
|
Jedrzej Kosinski
|
2c1e9f6cdd
|
Add workaround for outdated Steerable-Motion workflows to not throw errors (for my own peace of mind)
|
2024-06-13 07:54:03 -05:00 |
|
|
Jedrzej Kosinski
|
b77125bdb9
|
Prepare for update that removes model_keys prop from ModelPatcher
|
2024-06-02 19:04:38 -05:00 |
|
|
Jedrzej Kosinski
|
576426a734
|
Renamed extras to cn_extras to differentiate from potential extras on AnimateDiff in the future (which would be called ad_extras there)
|
2024-05-29 01:37:56 -05:00 |
|
|
Jedrzej Kosinski
|
949e1acf7e
|
Turned context_aware into a dropdown in case I add other context_aware types in future, renamed sparse_mask_strength to sparse_mask_mult
|
2024-05-29 01:23:21 -05:00 |
|
|
Jedrzej Kosinski
|
5ef18991ee
|
Refactored SparseMethod code to have a common get_indexes func, with _get_indexes the one that is abstract
|
2024-05-29 01:00:54 -05:00 |
|
|
Jedrzej Kosinski
|
74b6cf576f
|
Implemented context_aware
|
2024-05-28 22:32:19 -05:00 |
|
|
Jedrzej Kosinski
|
20d54c0660
|
Added support for weight extras, added sparse_hint_mult, sparse_nonhint_mult, and sparse_mask_strength for SparseCtrl, renamed Force Default Weights node to Default Weights
|
2024-05-27 18:49:28 -05:00 |
|
|
Jedrzej Kosinski
|
1f95fac964
|
Fixed odd behavior when context options are active and there are less condhints than latents, condhints/masks are extended (last image/mask) instead of repeated, starting to work on uncond_mask to extend uncond_multiplier feature, modified motion_strength > 1.0 behavior
|
2024-05-24 00:11:26 -05:00 |
|
|
Jedrzej Kosinski
|
5660a110df
|
Made SparseCtrl work with fp8 and cleaner lowvram support
|
2024-05-23 18:12:28 -05:00 |
|
|
Jedrzej Kosinski
|
c808e875c4
|
Implement fallback_attention_mm for SparseCtrl to mirror AnimateDiff-Evolved
|
2024-05-23 15:59:50 -05:00 |
|
|
Jedrzej Kosinski
|
e346228937
|
Make attn RefCN compatible with most recent comfy (backwards compatible)
|
2024-05-19 23:13:21 -05:00 |
|
|
Jedrzej Kosinski
|
bed20c2e28
|
Fixed contains_uncond_multipier tailend recursion instead of returning False
|
2024-05-17 03:24:23 -05:00 |
|
|
Jedrzej Kosinski
|
ab894722f6
|
Made len hack for uncond_multiplier more robust
|
2024-05-16 21:20:09 -05:00 |
|
|
Jedrzej Kosinski
|
86c528ddab
|
Make uncond_multiplier work with ControlLLLite
|
2024-05-16 17:55:06 -05:00 |
|
|
Jedrzej Kosinski
|
4ed14e9ae0
|
Make T2IAdapter work with uncond_multiplier (results are not great when < 1.0, but match up exactly with auto1111 results)
|
2024-05-16 17:40:46 -05:00 |
|
|
Jedrzej Kosinski
|
db0bf14dac
|
Add support for uncond_multiplier in all Control Weights (identical to auto1111's 'controlnet is more important when set to 0.0)
|
2024-05-16 04:15:36 -05:00 |
|
|
Jedrzej Kosinski
|
f61c3e2be3
|
Fix Timestep Keyframe Interpolation/From List nodes
|
2024-05-14 23:30:22 -05:00 |
|
|
Jedrzej Kosinski
|
cdfbdc0bbe
|
Refactored Timestep Keyframes to use guarantee_steps instead of guarantee_usage, Timestep Keyframes with same start_percent will no longer overwrite one another, allowing guarantee_steps to be used to schedule on a per-step basis without worrying about unique start_percents, added Timestep Keyframe Interpolation and From List nodes to ease creation of Timestep Keyframe schedules, small cleanup/upkeep
|
2024-05-14 04:05:18 -05:00 |
|
|
Jedrzej Kosinski
|
1a5a48d37d
|
Account for use_motion being off
|
2024-04-04 14:42:47 -05:00 |
|
|
Jedrzej Kosinski
|
0556fd2265
|
Fixed SparseCtrl not working with lowvram
|
2024-04-04 14:35:28 -05:00 |
|
|
Jedrzej Kosinski
|
3bf357e662
|
Removed extra empty line I accidentally added
|
2024-04-02 23:43:49 -05:00 |
|
|
Jedrzej Kosinski
|
3e975bb841
|
Fix for cases where predefined weights length doesn't match expected layer lengths (will need to generalize for for ControlLora to properly do soft weights for actual layer number)
|
2024-04-02 23:40:48 -05:00 |
|
|
Jedrzej Kosinski
|
0357841c15
|
Removed unnecessary dropdown from Reference ControlNet (Finetune) node, as should always be adain+attn anyway to prevent confusion
|
2024-04-02 01:14:06 -05:00 |
|
|
Jedrzej Kosinski
|
019452c790
|
Made sure ref cns match up to expected bank indexes, made adain properly react to strengths and masks
|
2024-04-02 01:08:19 -05:00 |
|
|
Jedrzej Kosinski
|
5e3c34027f
|
Added Reference ControlNet (Finetune) to control attn and adain values separately
|
2024-04-01 18:20:35 -05:00 |
|
|
Jedrzej Kosinski
|
414ac8d6ec
|
Started separating out attn and adain variables to prevent issues when using multiple RefCN
|
2024-04-01 04:28:00 -05:00 |
|
|
Jedrzej Kosinski
|
cb309c6078
|
Added initial reference_adain and reference_adain+attn support
|
2024-04-01 03:44:11 -05:00 |
|
|
Jedrzej Kosinski
|
829d10ddc0
|
Fixed BIGMIN and BIGMAX to conform with javascript limits, pretty-fied deprecated node
|
2024-03-31 01:30:07 -05:00 |
|
|
Jedrzej Kosinski
|
fb4cd18a7f
|
Only expose reference_attn for now since it's the only one with code support (adain coming soon)
|
2024-03-31 01:13:49 -05:00 |
|
|
Jedrzej Kosinski
|
e5841d1759
|
Added latent_kf and masking support (masks don't work per-area currently, but do work per-latent)
|
2024-03-31 00:34:56 -05:00 |
|
|
Jedrzej Kosinski
|
2fdf8236f6
|
Don't apply normal CNs during ref cn WRITE runs
|
2024-03-30 19:48:45 -05:00 |
|
|
Jedrzej Kosinski
|
7696628fdd
|
Simplified RefCN injection, model_opt no longer needed and should fix some intermittent bugs
|
2024-03-30 19:16:21 -05:00 |
|
|
Jedrzej Kosinski
|
37ae77b486
|
Applied proper strength control to attn1_replace_patch as well
|
2024-03-30 05:29:22 -05:00 |
|
|
Jedrzej Kosinski
|
25892b7b7d
|
Added proper strength control to ReferenceCN, replaced old strength with ref_weight
|
2024-03-30 05:26:09 -05:00 |
|
|
Jedrzej Kosinski
|
111632917c
|
Cleaned up ref cn so I can add more features
|
2024-03-30 04:02:44 -05:00 |
|
|
Jedrzej Kosinski
|
5c3786e1a7
|
Merge branch 'main' into develop
|
2024-03-19 10:55:07 -05:00 |
|
|
Jedrzej Kosinski
|
6e05ca03fe
|
Fixed ModelPatcher import in control.py
|
2024-03-19 10:40:44 -05:00 |
|
|
Jedrzej Kosinski
|
7678232a80
|
More refcn debugging and fix for refcn issue for ancestral samplers while testing
|
2024-03-19 10:38:52 -05:00 |
|
|
Jedrzej Kosinski
|
654988cf98
|
Merge branch 'main' into develop
|
2024-03-07 04:26:30 -06:00 |
|
|
Jedrzej Kosinski
|
0082222162
|
Updated T2IAdapterAdvanced to match recent ComfyUI changes
|
2024-03-07 04:21:37 -06:00 |
|