36a7953142
Greatly improve lowvram sampling speed by getting rid of accelerate.
...
Let me know if this breaks anything.
1 year ago
261bcbb0d9
A few missing comfy ops in the VAE.
1 year ago
9a7619b72d
Fix regression with inpaint model.
1 year ago
571ea8cdcc
Fix SAG not working with cfg 1.0
1 year ago
8cf1daa108
Fix SDXL area composition sometimes not using the right pooled output.
1 year ago
2258f85159
Support stable zero 123 model.
...
To use it use the ImageOnlyCheckpointLoader to load the checkpoint and
the new Stable_Zero123 node.
1 year ago
2f9d6a97ec
Add --deterministic option to make pytorch use deterministic algorithms.
1 year ago
e45d920ae3
Don't resize clip vision image when the size is already good.
1 year ago
13e6d5366e
Switch clip vision to manual cast.
...
Make it use the same dtype as the text encoder.
1 year ago
719fa0866f
Set clip vision model in eval mode so it works without inference mode.
1 year ago
574363a8a6
Implement Perp-Neg
1 year ago
a5056cfb1f
Remove useless code.
1 year ago
329c571993
Improve code legibility.
1 year ago
6c5990f7db
Fix cfg being calculated more than once if sampler_cfg_function.
1 year ago
ba04a87d10
Refactor and improve the sag node.
...
Moved all the sag related code to comfy_extras/nodes_sag.py
1 year ago
6761233e9d
Implement Self-Attention Guidance ( #2201 )
...
* First SAG test
* need to put extra options on the model instead of patcher
* no errors and results seem not-broken
* Use @ashen-uncensored formula, which works better!!!
* Fix a crash when using weird resolutions. Remove an unnecessary UNet call
* Improve comments, optimize memory in blur routine
* SAG works with sampler_cfg_function
1 year ago
b454a67bb9
Support segmind vega model.
1 year ago
824e4935f5
Add dtype parameter to VAE object.
1 year ago
32b7e7e769
Add manual cast to controlnet.
1 year ago
3152023fbc
Use inference dtype for unet memory usage estimation.
1 year ago
77755ab8db
Refactor comfy.ops
...
comfy.ops -> comfy.ops.disable_weight_init
This should make it more clear what they actually do.
Some unused code has also been removed.
1 year ago
b0aab1e4ea
Add an option --fp16-unet to force using fp16 for the unet.
1 year ago
ba07cb748e
Use faster manual cast for fp8 in unet.
1 year ago
57926635e8
Switch text encoder to manual cast.
...
Use fp16 text encoder weights for CPU inference to lower memory usage.
1 year ago
340177e6e8
Disable non blocking on mps.
1 year ago
614b7e731f
Implement GLora.
1 year ago
cb63e230b4
Make lora code a bit cleaner.
1 year ago
174eba8e95
Use own clip vision model implementation.
1 year ago
97015b6b38
Cleanup.
1 year ago
a4ec54a40d
Add linear_start and linear_end to model_config.sampling_settings
1 year ago
9ac0b487ac
Make --gpu-only put intermediate values in GPU memory instead of cpu.
1 year ago
efb704c758
Support attention masking in CLIP implementation.
1 year ago
fbdb14d4c4
Cleaner CLIP text encoder implementation.
...
Use a simple CLIP model implementation instead of the one from
transformers.
This will allow some interesting things that would too hackish to implement
using the transformers implementation.
1 year ago
2db86b4676
Slightly faster lora applying.
1 year ago
1bbd65ab30
Missed this one.
1 year ago
9b655d4fd7
Fix memory issue with control loras.
1 year ago
26b1c0a771
Fix control lora on fp8.
1 year ago
be3468ddd5
Less useless downcasting.
1 year ago
ca82ade765
Use .itemsize to get dtype size for fp8.
1 year ago
31b0f6f3d8
UNET weights can now be stored in fp8.
...
--fp8_e4m3fn-unet and --fp8_e5m2-unet are the two different formats
supported by pytorch.
1 year ago
af365e4dd1
All the unet ops with weights are now handled by comfy.ops
1 year ago
61a123a1e0
A different way of handling multiple images passed to SVD.
...
Previously when a list of 3 images [0, 1, 2] was used for a 6 frame video
they were concated like this:
[0, 1, 2, 0, 1, 2]
now they are concated like this:
[0, 0, 1, 1, 2, 2]
1 year ago
c97be4db91
Support SD2.1 turbo checkpoint.
1 year ago
983ebc5792
Use smart model management for VAE to decrease latency.
1 year ago
c45d1b9b67
Add a function to load a unet from a state dict.
1 year ago
f30b992b18
.sigma and .timestep now return tensors on the same device as the input.
1 year ago
13fdee6abf
Try to free memory for both cond+uncond before inference.
1 year ago
be71bb5e13
Tweak memory inference calculations a bit.
1 year ago
39e75862b2
Fix regression from last commit.
1 year ago
50dc39d6ec
Clean up the extra_options dict for the transformer patches.
...
Now everything in transformer_options gets put in extra_options.
1 year ago
5d6dfce548
Fix importing diffusers unets.
1 year ago
3e5ea74ad3
Make buggy xformers fall back on pytorch attention.
1 year ago
871cc20e13
Support SVD img2vid model.
1 year ago
410bf07771
Make VAE memory estimation take dtype into account.
1 year ago
32447f0c39
Add sampling_settings so models can specify specific sampling settings.
1 year ago
c3ae99a749
Allow controlling downscale and upscale methods in PatchModelAddDownscale.
1 year ago
72741105a6
Remove useless code.
1 year ago
6a491ebe27
Allow model config to preprocess the vae state dict on load.
1 year ago
cd4fc77d5f
Add taesd and taesdxl to VAELoader node.
...
They will show up if both the taesd_encoder and taesd_decoder or taesdxl
model files are present in the models/vae_approx directory.
1 year ago
ce67dcbcda
Make it easy for models to process the unet state dict on load.
1 year ago
d9d8702d8d
percent_to_sigma now returns a float instead of a tensor.
1 year ago
0cf4e86939
Add some command line arguments to store text encoder weights in fp8.
...
Pytorch supports two variants of fp8:
--fp8_e4m3fn-text-enc (the one that seems to give better results)
--fp8_e5m2-text-enc
1 year ago
107e78b1cb
Add support for loading SSD1B diffusers unet version.
...
Improve diffusers model detection.
1 year ago
7e3fe3ad28
Make deep shrink behave like it should.
1 year ago
9f00a18095
Fix potential issues.
1 year ago
7ea6bb038c
Print warning when controlnet can't be applied instead of crashing.
1 year ago
dcec1047e6
Invert the start and end percentages in the code.
...
This doesn't affect how percentages behave in the frontend but breaks
things if you relied on them in the backend.
percent_to_sigma goes from 0 to 1.0 instead of 1.0 to 0 for less confusion.
Make percent 0 return an extremely large sigma and percent 1.0 return a
zero one to fix imprecision.
1 year ago
57eea0efbb
heunpp2 sampler.
1 year ago
728613bb3e
Fix last pr.
1 year ago
ec3d0ab432
Merge branch 'master' of https://github.com/Jannchie/ComfyUI
1 year ago
c962884a5c
Make bislerp work on GPU.
1 year ago
420beeeb05
Clean up and refactor sampler code.
...
This should make it much easier to write custom nodes with kdiffusion type
samplers.
1 year ago
f2e49b1d57
fix: adaptation to older versions of pytroch
1 year ago
94cc718e9c
Add a way to add patches to the input block.
1 year ago
7339479b10
Disable xformers when it can't load properly.
1 year ago
4781819a85
Make memory estimation aware of model dtype.
1 year ago
dd4ba68b6e
Allow different models to estimate memory usage differently.
1 year ago
2c9dba8dc0
sampling_function now has the model object as the argument.
1 year ago
8d80584f6a
Remove useless argument from uni_pc sampler.
1 year ago
248aa3e563
Fix bug.
1 year ago
4a8a839b40
Add option to use in place weight updating in ModelPatcher.
1 year ago
412d3ff57d
Refactor.
1 year ago
58d5d71a93
Working RescaleCFG node.
...
This was broken because of recent changes so I fixed it and moved it from
the experiments repo.
1 year ago
3e0033ef30
Fix model merge bug.
...
Unload models before getting weights for model patching.
1 year ago
002aefa382
Support lcm models.
...
Use the "lcm" sampler to sample them, you also have to use the
ModelSamplingDiscrete node to set them as lcm models to use them properly.
1 year ago
ec12000136
Add support for full diff lora keys.
1 year ago
064d7583eb
Add a CONDConstant for passing non tensor conds to unet.
1 year ago
794dd2064d
Fix typo.
1 year ago
0a6fd49a3e
Print leftover keys when using the UNETLoader.
1 year ago
fe40109b57
Fix issue with object patches not being copied with patcher.
1 year ago
a527d0c795
Code refactor.
1 year ago
2a23ba0b8c
Fix unet ops not entirely on GPU.
1 year ago
844dbf97a7
Add: advanced->model->ModelSamplingDiscrete node.
...
This allows changing the sampling parameters of the model (eps or vpred)
or set the model to use zsnr.
1 year ago
656c0b5d90
CLIP code refactor and improvements.
...
More generic clip model class that can be used on more types of text
encoders.
Don't apply weighting algorithm when weight is 1.0
Don't compute an empty token output when it's not needed.
1 year ago
b3fcd64c6c
Make SDTokenizer class work with more types of tokenizers.
1 year ago
7e455adc07
fix unet_wrapper_function name in ModelPatcher
1 year ago
1ffa8858e7
Move model sampling code to comfy/model_sampling.py
1 year ago
ae2acfc21b
Don't convert Nan to zero.
...
Converting Nan to zero is a bad idea because it makes it hard to tell when
something went wrong.
1 year ago
d2e27b48f1
sampler_cfg_function now gets the noisy output as argument again.
...
This should make things that use sampler_cfg_function behave like before.
Added an input argument for those that want the denoised output.
This means you can calculate the x0 prediction of the model by doing:
(input - cond) for example.
1 year ago
2455aaed8a
Allow model or clip to be None in load_lora_for_models.
1 year ago
ecb80abb58
Allow ModelSamplingDiscrete to be instantiated without a model config.
1 year ago
e73ec8c4da
Not used anymore.
1 year ago
111f1b5255
Fix some issues with sampling precision.
1 year ago
7c0f255de1
Clean up percent start/end and make controlnets work with sigmas.
1 year ago
a268a574fa
Remove a bunch of useless code.
...
DDIM is the same as euler with a small difference in the inpaint code.
DDIM uses randn_like but I set a fixed seed instead.
I'm keeping it in because I'm sure if I remove it people are going to
complain.
1 year ago
1777b54d02
Sampling code changes.
...
apply_model in model_base now returns the denoised output.
This means that sampling_function now computes things on the denoised
output instead of the model output. This should make things more consistent
across current and future models.
1 year ago
c837a173fa
Fix some memory issues in sub quad attention.
1 year ago
125b03eead
Fix some OOM issues with split attention.
1 year ago
a12cc05323
Add --max-upload-size argument, the default is 100MB.
1 year ago
2a134bfab9
Fix checkpoint loader with config.
1 year ago
e60ca6929a
SD1 and SD2 clip and tokenizer code is now more similar to the SDXL one.
1 year ago
6ec3f12c6e
Support SSD1B model and make it easier to support asymmetric unets.
1 year ago
434ce25ec0
Restrict loading embeddings from embedding folders.
1 year ago
723847f6b3
Faster clip image processing.
1 year ago
a373367b0c
Fix some OOM issues with split and sub quad attention.
1 year ago
7fbb217d3a
Fix uni_pc returning noisy image when steps <= 3
1 year ago
3783cb8bfd
change 'c_adm' to 'y' in ControlNet.get_control
1 year ago
d1d2fea806
Pass extra conds directly to unet.
1 year ago
036f88c621
Refactor to make it easier to add custom conds to models.
1 year ago
3fce8881ca
Sampling code refactor to make it easier to add more conds.
1 year ago
8594c8be4d
Empty the cache when torch cache is more than 25% free mem.
1 year ago
8b65f5de54
attention_basic now works with hypertile.
1 year ago
e6bc42df46
Make sub_quad and split work with hypertile.
1 year ago
a0690f9df9
Fix t2i adapter issue.
1 year ago
9906e3efe3
Make xformers work with hypertile.
1 year ago
4185324a1d
Fix uni_pc sampler math. This changes the images this sampler produces.
1 year ago
e6962120c6
Make sure cond_concat is on the right device.
1 year ago
45c972aba8
Refactor cond_concat into conditioning.
1 year ago
430a8334c5
Fix some potential issues.
1 year ago
782a24fce6
Refactor cond_concat into model object.
1 year ago
0d45a565da
Fix memory issue related to control loras.
...
The cleanup function was not getting called.
1 year ago
d44a2de49f
Make VAE code closer to sgm.
1 year ago
23680a9155
Refactor the attention stuff in the VAE.
1 year ago
c8013f73e5
Add some Quadro cards to the list of cards with broken fp16.
1 year ago
bb064c9796
Add a separate optimized_attention_masked function.
1 year ago
fd4c5f07e7
Add a --bf16-unet to test running the unet in bf16.
1 year ago
9a55dadb4c
Refactor code so model can be a dtype other than fp32 or fp16.
1 year ago
88733c997f
pytorch_attention_enabled can now return True when xformers is enabled.
1 year ago
20d3852aa1
Pull some small changes from the other repo.
1 year ago
ac7d8cfa87
Allow attn_mask in attention_pytorch.
1 year ago
1a4bd9e9a6
Refactor the attention functions.
...
There's no reason for the whole CrossAttention object to be repeated when
only the operation in the middle changes.
1 year ago
8cc75c64ff
Let unet wrapper functions have .to attributes.
1 year ago
5e885bd9c8
Cleanup.
1 year ago
851bb87ca9
Merge branch 'taesd_safetensors' of https://github.com/mochiya98/ComfyUI
1 year ago
9eb621c95a
Supports TAESD models in safetensors format
1 year ago
d1a0abd40b
Merge branch 'input-directory' of https://github.com/jn-jairo/ComfyUI
1 year ago
72188dffc3
load_checkpoint_guess_config can now optionally output the model.
1 year ago
63e5fd1790
Option to input directory
1 year ago
9bfec2bdbf
Fix quality loss due to low precision
1 year ago
0f17993d05
fix: typo in extra sampler
1 year ago
comfyanonymous
Hari
Rafie Walker
Jianqi Pan
gameltb
Jedrzej Kosinski
Yukimasa Funaoka
Jairo Correa
City
badayvedat