Commits · 9fccf4aa031a3e0698c03bd5424c78e56fba9f08 · chenpangpang / ComfyUI

21 Jun, 2023 1 commit
- Add original_shape parameter to transformer patch extra_options. · 9fccf4aa
  comfyanonymous authored Jun 21, 2023
  
  9fccf4aa
20 Jun, 2023 3 commits
- Fix last commits causing an issue with the text encoder lora. · 51581dbf
  comfyanonymous authored Jun 20, 2023
  
  51581dbf
- Keep a set of model_keys for faster add_patches. · 8125b51a
  comfyanonymous authored Jun 20, 2023
  
  8125b51a
- Add a type of model patch useful for model merging. · 45beebd3
  comfyanonymous authored Jun 20, 2023
  
  45beebd3
19 Jun, 2023 3 commits
- Fix k_diffusion math being off by a tiny bit during txt2img. · 036a2207
  comfyanonymous authored Jun 19, 2023
  
  036a2207
- Add a way to set patches that modify the attn2 output. · 8883cb0f
  comfyanonymous authored Jun 18, 2023
```
Change the transformer patches function format to be more future proof.
```
  8883cb0f
- pop clip vision keys after loading them. · cd930d4e
  comfyanonymous authored Jun 18, 2023
  
  cd930d4e
18 Jun, 2023 3 commits
- Not needed anymore. · c9e4a8c9
  comfyanonymous authored Jun 18, 2023
  
  c9e4a8c9
- This is not needed anymore and causes issues with alphas_cumprod. · fb4bf7f5
  comfyanonymous authored Jun 18, 2023
  
  fb4bf7f5
- Fix DDIM v-prediction. · 45be2e92
  comfyanonymous authored Jun 17, 2023
  
  45be2e92
16 Jun, 2023 1 commit
- Fix an issue when alphas_comprod are half floats. · e6e50ab2
  comfyanonymous authored Jun 16, 2023
  
  e6e50ab2
15 Jun, 2023 5 commits
- All the unet weights should now be initialized with the right dtype. · ae43f09e
  comfyanonymous authored Jun 15, 2023
  
  ae43f09e
- Add a --gpu-only argument to keep and run everything on the GPU. · f7edcfd9
  comfyanonymous authored Jun 15, 2023
```
Make the CLIP model work on the GPU.
```
  f7edcfd9
- Initialize more unet weights as the right dtype. · 7bf89ba9
  comfyanonymous authored Jun 15, 2023
  
  7bf89ba9
- Initialize transformer unet block weights in right dtype at the start. · e21d9ad4
  comfyanonymous authored Jun 15, 2023
  
  e21d9ad4
- Properly disable weight initialization in clip models. · bb1f45d6
  comfyanonymous authored Jun 14, 2023
  
  bb1f45d6
14 Jun, 2023 6 commits
- Disable default weight values in unet conv2d for faster loading. · 21f04fe6
  comfyanonymous authored Jun 14, 2023
  
  21f04fe6
- This isn't needed for inference. · 9d54066e
  comfyanonymous authored Jun 14, 2023
  
  9d54066e
- Don't initialize CLIPVision weights to default values. · fa2cca05
  comfyanonymous authored Jun 14, 2023
  
  fa2cca05
- Set model to fp16 before loading the state dict to lower ram bump. · 6b774589
  comfyanonymous authored Jun 14, 2023
  
  6b774589
- Don't initialize clip weights to default values. · 0c7cad40
  comfyanonymous authored Jun 14, 2023
  
  0c7cad40
- Speed up model loading a bit. · 6971646b
  comfyanonymous authored Jun 14, 2023
```
Default pytorch Linear initializes the weights which is useless and slow.
```
  6971646b
13 Jun, 2023 6 commits
- sampler_cfg_function now uses a dict for the argument. · 388567f2
  comfyanonymous authored Jun 13, 2023
```
This means arguments can be added without issues.
```
  388567f2
- Turn on safe load for a few models. · ff9b22d7
  comfyanonymous authored Jun 13, 2023
  
  ff9b22d7
- Remove pytorch_lightning dependency. · 735ac4cf
  comfyanonymous authored Jun 13, 2023
  
  735ac4cf
- Remove useless code. · 2b14041d
  comfyanonymous authored Jun 13, 2023
  
  2b14041d
- Remove more useless files. · 274dff32
  comfyanonymous authored Jun 13, 2023
  
  274dff32
- Cleanup: Remove a bunch of useless files. · f0a2b81c
  comfyanonymous authored Jun 13, 2023
  
  f0a2b81c
12 Jun, 2023 2 commits
- Split the batch in VAEEncode if there's not enough memory. · f8c59310
  comfyanonymous authored Jun 12, 2023
  
  f8c59310
- Auto switch to tiled VAE encode if regular one runs out of memory. · c069fc07
  comfyanonymous authored Jun 11, 2023
  
  c069fc07
11 Jun, 2023 1 commit
- Refactor unCLIP noise augment out of samplers.py · c64ca8c0
  comfyanonymous authored Jun 11, 2023
  
  c64ca8c0
09 Jun, 2023 2 commits
- Simpler base model code. · de142eaa
  comfyanonymous authored Jun 09, 2023
  
  de142eaa
- Fix bug when embedding gets ignored because of mismatched size. · 23cf8ca7
  comfyanonymous authored Jun 08, 2023
  
  23cf8ca7
06 Jun, 2023 2 commits
- Small refactor. · 0e425603
  comfyanonymous authored Jun 06, 2023
  
  0e425603
- Refactor previews into one command line argument. · a3a713b6
  comfyanonymous authored Jun 06, 2023
```
Clean up a few things.
```
  a3a713b6
05 Jun, 2023 4 commits
- preview method autodetection · 3e17971a
  space-nuko authored Jun 05, 2023
  
  3e17971a
- Add latent2rgb preview · d5a28fad
  space-nuko authored Jun 05, 2023
  
  d5a28fad
- Make previews into cli option · 48f7ec75
  space-nuko authored Jun 05, 2023
  
  48f7ec75
- Preview sampled images with TAESD · b4f434ee
  space-nuko authored May 30, 2023
  
  b4f434ee
04 Jun, 2023 1 commit
- Some comments to say what the vram state options mean. · fed0a4dd
  comfyanonymous authored Jun 04, 2023
  
  fed0a4dd