Commits · c62e8361676e53f86422e5cb5511673ef3b61e78 · chenpangpang / ComfyUI

01 Mar, 2024 1 commit
- Move noise scaling to object with sampling math. · c62e8361
  comfyanonymous authored Mar 01, 2024
  
  c62e8361
29 Feb, 2024 1 commit
- Allow image_only_indicator to be None. · cb7c3a29
  comfyanonymous authored Feb 29, 2024
  
  cb7c3a29
28 Feb, 2024 2 commits
- Koala 700M and 1B support. · b3e97fc7
  comfyanonymous authored Feb 28, 2024
```
Use the UNET Loader node to load the unet file to use them.
```
  b3e97fc7
- Remove duplicate text_projection key from some saved models. · 37a86e46
  comfyanonymous authored Feb 28, 2024
  
  37a86e46
27 Feb, 2024 5 commits
- Auto detect playground v2.5 model. · 8daedc5b
  comfyanonymous authored Feb 27, 2024
  
  8daedc5b
- Playground V2.5 support with ModelSamplingContinuousEDM node. · d46583ec
  comfyanonymous authored Feb 27, 2024
```
Use ModelSamplingContinuousEDM with edm_playground_v2.5 selected.
```
  d46583ec
- Make XL checkpoints save in a more standard format. · 1e0fcc9a
  comfyanonymous authored Feb 27, 2024
  
  1e0fcc9a
- Make the text projection saved in the checkpoint the right format. · b416be7d
  comfyanonymous authored Feb 27, 2024
  
  b416be7d
- Add a min_length property to tokenizer class. · 03c47fc0
  comfyanonymous authored Feb 26, 2024
  
  03c47fc0
25 Feb, 2024 5 commits
- Make return_projected_pooled setable from the __init__ · 8ac69f62
  comfyanonymous authored Feb 25, 2024
  
  8ac69f62
- Support loading old CLIP models saved with CLIPSave. · ca7c310a
  comfyanonymous authored Feb 25, 2024
  
  ca7c310a
- Always return unprojected pooled output for gligen. · c2cb8e88
  comfyanonymous authored Feb 25, 2024
  
  c2cb8e88
- Move text projection into the CLIP model code. · 1cb3f6a8
  comfyanonymous authored Feb 25, 2024
```
Fix issue with not loading the SSD1B clip correctly.
```
  1cb3f6a8
- Support text encoder text_projection in lora. · 6533b172
  comfyanonymous authored Feb 24, 2024
  
  6533b172
23 Feb, 2024 3 commits
- Support lora keys with lora_prior_unet_ and lora_prior_te_ · 1e5f0f66
  comfyanonymous authored Feb 23, 2024
  
  1e5f0f66
- Fix model and cond transformer options merge · e1cb93c3
  logtd authored Feb 23, 2024
  
  e1cb93c3
- Cleanup uni_pc inpainting. · 10847dfa
  comfyanonymous authored Feb 23, 2024
```
This causes some small changes to the uni pc inpainting behavior but it
seems to improve results slightly.
```
  10847dfa
20 Feb, 2024 3 commits
- Add some latent2rgb matrices for previews. · 18c151b3
  comfyanonymous authored Feb 20, 2024
  
  18c151b3
- Pass pooled CLIP to stage b. · 0d0fbabd
  comfyanonymous authored Feb 20, 2024
  
  0d0fbabd
- Align simple scheduling closer to official stable cascade scheduler. · c6b7a157
  comfyanonymous authored Feb 20, 2024
  
  c6b7a157
19 Feb, 2024 6 commits
- Enable fp16 by default on mps. · 88f30040
  comfyanonymous authored Feb 19, 2024
  
  88f30040
- Remove print. · e93cdd0a
  comfyanonymous authored Feb 19, 2024
  
  e93cdd0a
- Support Stable Cascade in checkpoint format. · 3711b31d
  comfyanonymous authored Feb 19, 2024
  
  3711b31d
- Some cleanups to how the text encoders are loaded. · d91f45ef
  comfyanonymous authored Feb 19, 2024
  
  d91f45ef
- Forgot to commit this. · a7b5eaa7
  comfyanonymous authored Feb 19, 2024
  
  a7b5eaa7
- Support loading the Stable Cascade effnet and previewer as a VAE. · 3b2e5799
  comfyanonymous authored Feb 19, 2024
```
The effnet can be used to encode images for img2img with Stage C.
```
  3b2e5799
18 Feb, 2024 2 commits
- Fix gligen lowvram mode. · dccca1da
  comfyanonymous authored Feb 18, 2024
  
  dccca1da
- Add ModelSamplingStableCascade to control the shift sampling parameter. · 8b60d33b
  comfyanonymous authored Feb 18, 2024
```
shift is 2.0 by default on Stage C and 1.0 by default on Stage B.
```
  8b60d33b
17 Feb, 2024 10 commits
- Fix attention masks properly for multiple batches. · 6bcf57ff
  comfyanonymous authored Feb 17, 2024
  
  6bcf57ff
- fp8 weight support for Stable Cascade. · 11e3221f
  comfyanonymous authored Feb 17, 2024
  
  11e3221f
- Fix attention mask batch size in some attention functions. · f8706546
  comfyanonymous authored Feb 17, 2024
  
  f8706546
- Properly fix attention masks in CLIP with batches. · 3b9969c1
  comfyanonymous authored Feb 17, 2024
  
  3b9969c1
- Implement shift schedule for cascade stage C. · 5b40e7a5
  comfyanonymous authored Feb 17, 2024
  
  5b40e7a5
- Manual cast for bf16 on older GPUs. · 929e266f
  comfyanonymous authored Feb 17, 2024
  
  929e266f
- Fix clip attention mask issues on some hardware. · 6c875d84
  comfyanonymous authored Feb 17, 2024
  
  6c875d84
- Make Stable Cascade work on old pytorch 2.0 · 805c36ac
  comfyanonymous authored Feb 17, 2024
  
  805c36ac
- Support Stable Cascade Stage B lite. · f2d1d16f
  comfyanonymous authored Feb 16, 2024
  
  f2d1d16f
- Make --force-fp32 disable loading models in bf16. · 0b3c5048
  comfyanonymous authored Feb 16, 2024
  
  0b3c5048
16 Feb, 2024 2 commits
- StableCascade CLIP model support. · 97d03ae0
  comfyanonymous authored Feb 16, 2024
  
  97d03ae0
- Stable Cascade Stage B. · 667c9281
  comfyanonymous authored Feb 16, 2024
  
  667c9281