crowsonkb/k-diffusion

Issues

• How to perform EDM (k-diffusion) conditoned on an image ?

  #115 opened 17 days ago by manmeet3591
  0

• Confirm JSON config for FFHQ-1024?

  #103 opened a year ago by tin-sely
  2

• Stabilizing DPM++2M SDE for SDXL

  #85 opened a year ago by LuChengTHU
  2

• Why zero initialize patch_out in hourglass transformer?

  #113 opened 4 months ago by Luciennnnnnn
  1

• About missing "ctx" argument in autograd function "ApplyRotaryEmbeddingInplace"

  #112 opened 6 months ago by sairights
  0

• Scaled cosine similarity can cause NaNs

  #111 opened 7 months ago by bobcao3
  0

• Use of DCT in loss function?

  #110 opened 8 months ago by sundeco
  0

• Blocking ops in several samplers, usually from sigmas being a GPU tensor

  #108 opened 10 months ago by drhead
  0

• Patch size 2 for HDiT

  #107 opened 10 months ago by yuanzhi-zhu
  1

• bad sampler name: DPM++ 2M Karras

  #106 opened 10 months ago by itsklimov
  0

• How to inference the trained model?

  #105 opened 10 months ago by sudheendra-krishnamurthy
  0

• Conditional *image* generation (img2img)

  #70 opened 2 years ago by CallShaul
  2

• Is there any plan to release pretrained HDiT models?

  #102 opened a year ago by iamwangyabin
  0

• LMS sampler question

  #76 opened a year ago by catboxanon
  2

• Sample many files more efficiently

  #49 opened 2 years ago by hotohoto
  1

• sample.py: Error while deserializing header: HeaderTooLarge

  #99 opened a year ago by drscotthawley
  4

• Questions about soft-min-snr loss

  #98 opened a year ago by hmicrobe
  1

• Scaling from 128x128, to 256x256, 512x512 and 1024x1024?

  #95 opened a year ago by tin-sely
  4

• R2 setup in the paper

  #96 opened a year ago by nicolas-dufour
  4

• Consider Making Normalization Optional in MappingFeedForwardBlock

  #94 opened a year ago by mnslarcher
  2

• AttributeError in Oxford Flowers Demo: 'dict' object has no attribute 'convert'

  #93 opened a year ago by mnslarcher
  1

• Pretrained models of Hourglass Diffusion Transformers

  #92 opened a year ago by fenglinglwb
  0

• `--mixed-precision` doesnt work with img transformer 2

  #91 opened a year ago by yoinked-h
  2

• How to cite this?

  #89 opened a year ago by exx8
  1

• FID scores from paper

  #51 opened 2 years ago by darius-lam
  4

• UnboundLocalError: local variable 'h' referenced before assignment

  #88 opened a year ago by LeonZhao28
  0

• [Feature request]Support DDIM-Solver and Seniorious

  #87 opened a year ago by sdbds
  0

• Stabilize the sampling of DPM-Solver++2M by a stabilizing trick

  #43 opened 2 years ago by LuChengTHU
  22

• UnboundLocalError: local variable 'h' referenced before assignment

  #72 opened 2 years ago by yaiol
  2

• k-diffusion triggering of torch.compile/torch.multiprocessing leaves multiple child processes

  #81 opened a year ago by vladmandic
  2

• Loglikelihood Evaluation Question

  #79 opened 2 years ago by Me-lab
  14

• Seek for help about reconstructing the denoiser and sample function

  #75 opened 2 years ago by Aziily
  0

• Common Diffusion Noise Schedules and Sample Steps are Flawed

  #64 opened 2 years ago by sdbds
  1

• sample_dpmpp_2m has a bug?

  #56 opened 2 years ago by hallatore
  19

• Feature Request: Support the "correct" SDE variant of DPM-Solver++

  #59 opened 2 years ago by LuChengTHU
  5

• time consuming

  #63 opened 2 years ago by fuchao01
  0

• Add the newest DPM-Solver

  #40 opened 2 years ago by LuChengTHU
  6

• Addition of UniPC?

  #58 opened 2 years ago by ImpossibleAd436
  0

• Accelerate is not working fine

  #57 opened 2 years ago by frangam
  0

• Both the FID scores and visualization are pretty bad in FFHQ dataset

  #52 opened 2 years ago by IrohCao-NYU
  1

• ModuleNotFoundError: No module named 'torch_directml'

  #54 opened 2 years ago by JoeVX
  0

• Implement k-diffusion with guided-diffusion for image gen with unequal sizes

  #53 opened 2 years ago by luisarandas
  0

• Learning rate scheduler broken

  #50 opened 2 years ago by kyleliang919
  2

• (Missing) Loss weights for the Diffusion Loss?

  #45 opened 2 years ago by mbreuss
  0

• sample sigma scheduler bug

  #37 opened 2 years ago by lockmatrix
  2

• Question: Is it possible to predict step count of DPM adaptive?

  #36 opened 2 years ago by C43H66N12O12S2
  1

• DPM2 ancestral produce odd nosiy/sharpened output during final iteration when using with Stable Difussion

  #34 opened 2 years ago by Hakkin
  0

• Is it possible to add any arbitrary loss terms into k-diffusion, e.g. to use lpips, edge stabilization, etc.

  #31 opened 2 years ago by oxysoft
  2

• [Feature request] Combination of Inpainting and ImgToImg Generation for finetunig

  #29 opened 2 years ago by Serasul1984
  0

• Req: TQDM trange() update frequency / miniters option

  #27 opened 3 years ago by torridgristle
  0