repo stringclasses 1
value | github_id int64 1.27B 4.61B | github_node_id stringlengths 18 24 | number int64 8 13.9k | html_url stringlengths 49 53 | api_url stringlengths 59 63 | title stringlengths 1 402 | body stringlengths 1 62.9k ⌀ | state stringclasses 2
values | state_reason stringclasses 4
values | locked bool 2
classes | comments_count int64 0 99 | labels listlengths 0 5 | assignees listlengths 0 5 | created_at stringdate 2022-06-09 16:28:35 2026-06-07 04:02:58 | updated_at stringdate 2022-06-12 22:18:01 2026-06-07 15:42:15 | closed_at stringdate 2022-06-12 22:18:01 2026-06-05 09:16:43 ⌀ | author_association stringclasses 3
values | milestone_title stringclasses 0
values | snapshot_id stringclasses 63
values | extracted_at stringdate 2026-04-07 13:34:13 2026-06-07 16:35:14 | author_login stringlengths 3 28 | author_id int64 1.54k 290M | author_node_id stringlengths 12 20 | author_type stringclasses 3
values | author_site_admin bool 1
class |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
huggingface/diffusers | 2,555,644,249 | I_kwDOHa8MBc6YVAlZ | 9,558 | https://github.com/huggingface/diffusers/issues/9558 | https://api.github.com/repos/huggingface/diffusers/issues/9558 | The validation image is black when displayed in the tensorboard | ### Describe the bug
when I use the script `train_text_to_image_lora_sdxl.py` to train my lora, the validation image in tensorboard is black .... just like this
### Reproduction
the code in train_text_to_image_lora_sdxl
``... | closed | completed | false | 7 | [
"bug"
] | [] | 2024-09-30T06:03:24Z | 2024-09-30T13:55:37Z | 2024-09-30T13:55:37Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | todochenxi | 108,149,467 | U_kgDOBnI62w | User | false |
huggingface/diffusers | 2,555,890,107 | I_kwDOHa8MBc6YV8m7 | 9,560 | https://github.com/huggingface/diffusers/issues/9560 | https://api.github.com/repos/huggingface/diffusers/issues/9560 | FP32 training for sd3 controlnet | Hi,
I have been use `examples\controlnet\train_controlnet_sd3.py` for controlnet training for a while, and I have some confusion and would like your advice
1. In the line 1097:
`vae.to(accelerator.device, dtype=torch.float32)`
It seems we should use fp32 for VAE, but as far as I know, SD3 currently has no fp32 ch... | closed | completed | false | 11 | [
"stale"
] | [] | 2024-09-30T08:07:04Z | 2024-10-31T15:13:19Z | 2024-10-31T15:13:19Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | xduzhangjiayu | 73,270,275 | MDQ6VXNlcjczMjcwMjc1 | User | false |
huggingface/diffusers | 2,560,049,766 | I_kwDOHa8MBc6Yl0Jm | 9,566 | https://github.com/huggingface/diffusers/issues/9566 | https://api.github.com/repos/huggingface/diffusers/issues/9566 | Documentation about superresolution | ### Describe the bug
Reading this [documentation](https://huggingface.co/docs/diffusers/en/using-diffusers/img2img#image-to-upscaler-to-super-resolution) which is not possible.
Indeed, [super resolution pipeline](https://huggingface.co/docs/diffusers/v0.16.0/en/api/pipelines/stable_diffusion/upscale#diffusers.StableD... | closed | completed | false | 7 | [
"bug"
] | [
"asomoza"
] | 2024-10-01T19:38:30Z | 2025-02-03T17:54:03Z | 2025-02-03T17:54:03Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | christopher5106 | 6,875,375 | MDQ6VXNlcjY4NzUzNzU= | User | false |
huggingface/diffusers | 2,560,596,249 | I_kwDOHa8MBc6Yn5kZ | 9,567 | https://github.com/huggingface/diffusers/issues/9567 | https://api.github.com/repos/huggingface/diffusers/issues/9567 | [community] Improving docstrings and type hints | There are many instances in the codebase where our docstring/typing convention is not followed. We'd like to work on improving this with your help!
Our convention looks like:
```python3
def function_name(parameter_1: Union[str, List[str]], parameter_2: Optional[int] = None, parameter_3: float = 42.0) -> Civiliza... | closed | completed | false | 16 | [
"documentation",
"good first issue",
"contributions-welcome"
] | [] | 2024-10-02T03:20:44Z | 2025-11-13T22:45:59Z | 2025-11-13T22:45:59Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | a-r-r-o-w | 72,266,394 | MDQ6VXNlcjcyMjY2Mzk0 | User | false |
huggingface/diffusers | 2,560,770,892 | I_kwDOHa8MBc6YokNM | 9,568 | https://github.com/huggingface/diffusers/issues/9568 | https://api.github.com/repos/huggingface/diffusers/issues/9568 | cogvideox autoencoder could not reconstruct an image | ### Describe the bug
AutoencoderKLCogVideoX cannot reconstruct single frame image
### Reproduction
```python
from diffusers import AutoencoderKLCogVideoX
device='cuda'
autoencoder = AutoencoderKLCogVideoX.from_pretrained(
"THUDM/CogVideoX-5b",
subfolder="vae",
torch_dtype=torch.bfloat16,
... | closed | completed | false | 2 | [
"bug"
] | [] | 2024-10-02T06:36:00Z | 2024-10-02T19:08:50Z | 2024-10-02T19:08:49Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Xiang-cd | 66,510,463 | MDQ6VXNlcjY2NTEwNDYz | User | false |
huggingface/diffusers | 2,564,217,891 | I_kwDOHa8MBc6Y1twj | 9,575 | https://github.com/huggingface/diffusers/issues/9575 | https://api.github.com/repos/huggingface/diffusers/issues/9575 | diffusers version update to 0.27.0 from 0.20.0, training code seems not work | I have trained an inpainting model using diffusers 0.20.0. The trained model works as expected. However, something seems wrong when I update the diffusers version to 0.27.0, while keeping the training code and other requirements the same. The training code runs successfully, but the inference outputs look like noise. I... | closed | completed | false | 4 | [] | [] | 2024-10-03T14:30:21Z | 2024-10-15T08:58:36Z | 2024-10-15T08:58:35Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | huangjun12 | 22,365,664 | MDQ6VXNlcjIyMzY1NjY0 | User | false |
huggingface/diffusers | 2,564,752,835 | I_kwDOHa8MBc6Y3wXD | 9,576 | https://github.com/huggingface/diffusers/issues/9576 | https://api.github.com/repos/huggingface/diffusers/issues/9576 | use_safetensors parameter not passed to submodules when loading pipeline | ### Describe the bug
When a model is loading from_pretrained, if the model is .bin and not .safetensors i got warnings like
```
"An error occurred while trying to fetch models/stablediffusionapi/yamermix-v8-vae: Error no file named diffusion_pytorch_model.safetensors found in directory models/stablediffusionapi/y... | closed | completed | false | 4 | [
"bug"
] | [] | 2024-10-03T18:41:22Z | 2024-10-07T21:04:43Z | 2024-10-07T21:04:43Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | elismasilva | 40,075,615 | MDQ6VXNlcjQwMDc1NjE1 | User | false |
huggingface/diffusers | 2,567,124,466 | I_kwDOHa8MBc6ZAzXy | 9,585 | https://github.com/huggingface/diffusers/issues/9585 | https://api.github.com/repos/huggingface/diffusers/issues/9585 | APG: Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Models | New paper just released, a new way to do classifier-free guidance (CFG) without affecting image saturation, introducing artifacts, etc. Could be huge.
https://huggingface.co/papers/2410.02416 | open | null | false | 6 | [
"stale",
"consider-for-modular-diffusers"
] | [
"a-r-r-o-w"
] | 2024-10-04T19:09:00Z | 2025-04-25T23:23:38Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | oxysoft | 3,945,277 | MDQ6VXNlcjM5NDUyNzc= | User | false |
huggingface/diffusers | 2,569,444,239 | I_kwDOHa8MBc6ZJpuP | 9,599 | https://github.com/huggingface/diffusers/issues/9599 | https://api.github.com/repos/huggingface/diffusers/issues/9599 | Why there is no LoRA only finetune example of FLUX.1? | **Is your feature request related to a problem? Please describe.**
The only example of LoRA finetune for FLUX.1 I discovered is here:
https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/train_dreambooth_lora_flux.py
which is a dreambooth example. The dreambooth is VRAM intensive and not useful for... | closed | completed | false | 3 | [] | [] | 2024-10-07T06:22:54Z | 2024-10-09T12:48:32Z | 2024-10-09T12:48:32Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | eeyrw | 2,088,817 | MDQ6VXNlcjIwODg4MTc= | User | false |
huggingface/diffusers | 1,420,436,231 | I_kwDOHa8MBc5UqiMH | 960 | https://github.com/huggingface/diffusers/issues/960 | https://api.github.com/repos/huggingface/diffusers/issues/960 | What's the difference between PNDM and plms | What's the difference between PNDM and plms，plms for CompVis/stable-diffusion-v1-4-origin | closed | completed | false | 2 | [
"stale"
] | [] | 2022-10-24T08:27:40Z | 2023-01-26T15:53:03Z | 2022-12-01T15:03:19Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | dingjingzhen | 21,244,263 | MDQ6VXNlcjIxMjQ0MjYz | User | false |
huggingface/diffusers | 2,570,118,884 | I_kwDOHa8MBc6ZMObk | 9,600 | https://github.com/huggingface/diffusers/issues/9600 | https://api.github.com/repos/huggingface/diffusers/issues/9600 | StableDiffusionControlNetPipeline: VRAM exploding up to 1 GB every single step | ### Describe the bug
It seems that when I run ControlNet, the VRAM explodes out of control. I'm not sure if it happens with just one ControlNet. No problem if I run the basic StableDiffusionPipeline without any ControlNet.
### Reproduction
```
# !pip install opencv-python transformers accelerate
from diffuse... | closed | completed | false | 4 | [
"bug"
] | [] | 2024-10-07T11:14:06Z | 2024-10-08T02:27:03Z | 2024-10-07T19:47:56Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | oxysoft | 3,945,277 | MDQ6VXNlcjM5NDUyNzc= | User | false |
huggingface/diffusers | 2,570,644,629 | I_kwDOHa8MBc6ZOOyV | 9,602 | https://github.com/huggingface/diffusers/issues/9602 | https://api.github.com/repos/huggingface/diffusers/issues/9602 | Adapter name(s) {'<my_adapter_name>'} not in the list of present adapters: set() | ### Describe the bug
Trying to load a xlabs lora (with single blocks, ie one that upon conversion executes [this](https://github.com/huggingface/diffusers/blob/3159e60d59819ae874ea3cdbd28e02d9e6c57321/src/diffusers/loaders/lora_conversion_utils.py#L635) line) now breaks since PR #9535 , raising the error `ValueError... | closed | completed | false | 4 | [
"bug"
] | [] | 2024-10-07T14:39:14Z | 2024-10-07T16:39:11Z | 2024-10-07T16:39:11Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Clement-Lelievre | 70,368,164 | MDQ6VXNlcjcwMzY4MTY0 | User | false |
huggingface/diffusers | 2,573,860,989 | I_kwDOHa8MBc6ZagB9 | 9,607 | https://github.com/huggingface/diffusers/issues/9607 | https://api.github.com/repos/huggingface/diffusers/issues/9607 | FlowMatch schedulers - closing the gap | As stated [here](https://github.com/huggingface/diffusers/issues/9490#issuecomment-2369756363), lets close the scheduler gap!
Problem statement:
- Most (if not all quality ones) new models are DiT based
- Implementation of DiT based models in Diffusers typically requires Flow-match type schedulers
- Currently *on... | open | null | false | 11 | [
"wip",
"scheduler"
] | [
"yiyixuxu"
] | 2024-10-08T18:00:58Z | 2025-12-05T18:47:37Z | null | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | vladmandic | 57,876,960 | MDQ6VXNlcjU3ODc2OTYw | User | false |
huggingface/diffusers | 2,574,319,219 | I_kwDOHa8MBc6ZcP5z | 9,610 | https://github.com/huggingface/diffusers/issues/9610 | https://api.github.com/repos/huggingface/diffusers/issues/9610 | add load_lora_weights to pipeline_stable_diffusion_3_img2img.py | **Is your feature request related to a problem? Please describe.*"
Yes, there is no effective way to load lora weights with SD3
A clear and concise description of what the problem is. Ex. I'm always frustrated when [...].
I can't get LoRa weights to load despite the fact there is a thorough and easy way to train t... | closed | completed | false | 3 | [] | [] | 2024-10-08T22:15:37Z | 2024-10-14T21:39:21Z | 2024-10-14T21:39:21Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | TheOneTrueGuy | 3,240,071 | MDQ6VXNlcjMyNDAwNzE= | User | false |
huggingface/diffusers | 2,574,404,018 | I_kwDOHa8MBc6Zckmy | 9,611 | https://github.com/huggingface/diffusers/issues/9611 | https://api.github.com/repos/huggingface/diffusers/issues/9611 | FluxImg2ImgPipeline and FluxInpaintPipeline image batching error (0.31.0.dev) | ### Describe the bug
Providing multiple input images, i.e. a batch number other than 1 to ``FluxImg2ImgPipeline`` or ``FluxInpaintPipeline`` results in a tensor shape error.
### Reproduction
Img2Img:
```python
import PIL.Image
import diffusers
import torch
# img2img
pipe = diffusers.FluxImg2ImgP... | closed | completed | false | 1 | [
"bug"
] | [] | 2024-10-08T23:26:55Z | 2024-10-08T23:47:59Z | 2024-10-08T23:47:59Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Teriks | 14,919,098 | MDQ6VXNlcjE0OTE5MDk4 | User | false |
huggingface/diffusers | 2,574,914,208 | I_kwDOHa8MBc6ZehKg | 9,613 | https://github.com/huggingface/diffusers/issues/9613 | https://api.github.com/repos/huggingface/diffusers/issues/9613 | Error "'Timesteps' object has no attribute 'scale'" on the 0.30.x version while using my own diffusion model | ### Describe the bug
I use my own pretrained model to generate images. When I use diffusers 0.29.2 everything is fine, but when I updated it to 0.30.0, for whatever reason, it just stopped working.
<img width="809" alt="Screenshot 2024-10-09 at 1 28 30 PM" src="https://github.com/user-attachments/assets/b1103560-ea3c... | closed | completed | false | 3 | [
"bug"
] | [] | 2024-10-09T06:39:07Z | 2024-10-09T13:55:55Z | 2024-10-09T13:55:37Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Vasasasasa | 126,963,684 | U_kgDOB5FP5A | User | false |
huggingface/diffusers | 2,575,009,294 | I_kwDOHa8MBc6Ze4YO | 9,615 | https://github.com/huggingface/diffusers/issues/9615 | https://api.github.com/repos/huggingface/diffusers/issues/9615 | UNet1DConditionModel Implementation | ### Model/Pipeline/Scheduler description
[diffusers-unet-1d-condition](https://github.com/lucainiaoge/diffusers-unet-1d-condition)
- This is a counterpart to the [UNet2DConditionModel](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/unets/unet_2d_condition.py)
- Added new blocks for 1d mode... | closed | completed | false | 3 | [] | [] | 2024-10-09T07:14:27Z | 2024-11-12T05:11:47Z | 2024-10-09T13:04:55Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | lucainiaoge | 32,015,775 | MDQ6VXNlcjMyMDE1Nzc1 | User | false |
huggingface/diffusers | 2,575,637,994 | I_kwDOHa8MBc6ZhR3q | 9,619 | https://github.com/huggingface/diffusers/issues/9619 | https://api.github.com/repos/huggingface/diffusers/issues/9619 | Make ControlNet pipeline `StableDiffusionControlNetPipeline` support `interrupt` | **Is your feature request related to a problem? Please describe.**
`StableDiffusionPipeline` has an interrupt that stops the denoising loop when no longer needed. but the ControlNet pipeline doesn't have it.
https://github.com/huggingface/diffusers/blob/31058cdaef63ca660a1a045281d156239fba8192/src/diffusers/pipeli... | closed | completed | false | 2 | [] | [] | 2024-10-09T11:27:27Z | 2024-10-09T22:03:15Z | 2024-10-09T22:03:15Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | pureexe | 1,303,847 | MDQ6VXNlcjEzMDM4NDc= | User | false |
huggingface/diffusers | 1,420,651,491 | I_kwDOHa8MBc5UrWvj | 962 | https://github.com/huggingface/diffusers/issues/962 | https://api.github.com/repos/huggingface/diffusers/issues/962 | The new inpaint pipelines sames not support other size than 512*512 | ### Describe the bug
When I try the 512*512 image and mask, it works well.
But when I test other size, the new inpaint pipe tip a error:
```
Traceback (most recent call last):
File "/root/work/tasker/test.py", line 33, in <module>
image = pipe(prompt=prompt, image=image, mask_image=mask_image).images[0]
... | closed | completed | false | 5 | [
"bug"
] | [
"patrickvonplaten"
] | 2022-10-24T11:06:05Z | 2022-10-26T15:55:17Z | 2022-10-26T15:55:17Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | CrazyBoyM | 35,400,185 | MDQ6VXNlcjM1NDAwMTg1 | User | false |
huggingface/diffusers | 2,576,357,159 | I_kwDOHa8MBc6ZkBcn | 9,624 | https://github.com/huggingface/diffusers/issues/9624 | https://api.github.com/repos/huggingface/diffusers/issues/9624 | Its hard to clear VRam to change pipelines | ### Describe the bug
Its not trivial to clear VRam to change pipelines.
This is a snippet of my code:
```
pipe.to("cpu")
try:
del pipe.transformer
except:
print("no transformer")
try:
del pipe.sched... | closed | completed | false | 3 | [
"bug"
] | [] | 2024-10-09T16:04:21Z | 2024-10-09T17:02:24Z | 2024-10-09T17:00:25Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | brurpo | 24,902,022 | MDQ6VXNlcjI0OTAyMDIy | User | false |
huggingface/diffusers | 2,576,505,418 | I_kwDOHa8MBc6ZklpK | 9,625 | https://github.com/huggingface/diffusers/issues/9625 | https://api.github.com/repos/huggingface/diffusers/issues/9625 | process_encoder_hidden_states function is applied on each denoising step | https://github.com/huggingface/diffusers/blob/31058cdaef63ca660a1a045281d156239fba8192/src/diffusers/models/unets/unet_2d_condition.py#L1004
Hey! I noticed that `process_encoder_hidden_states` is applied on each denoising step that influences on an inference performance. This function can be executed only once befor... | open | null | false | 6 | [
"stale"
] | [] | 2024-10-09T17:19:40Z | 2024-12-11T15:03:54Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Benkovichnikita | 10,713,288 | MDQ6VXNlcjEwNzEzMjg4 | User | false |
huggingface/diffusers | 2,576,802,992 | I_kwDOHa8MBc6ZluSw | 9,628 | https://github.com/huggingface/diffusers/issues/9628 | https://api.github.com/repos/huggingface/diffusers/issues/9628 | ValueError: Attempting to unscale FP16 gradients. when using examples/dreambooth/train_dreambooth_lora_sd3.py | ### Describe the bug
I was using the example training script `examples/dreambooth/train_dreambooth_lora_sd3.py`.
I changed `autocast_ctx = nullcontext()` to `autocast_ctx = torch.autocast(accelerator.device.type, dtype=torch.float32)` to address RuntimeError: Input type (float) and bias type (c10::Half) should be ... | closed | completed | false | 5 | [
"bug"
] | [] | 2024-10-09T19:40:33Z | 2024-10-16T16:46:31Z | 2024-10-16T16:46:31Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | yafeim | 29,441,946 | MDQ6VXNlcjI5NDQxOTQ2 | User | false |
huggingface/diffusers | 2,577,046,938 | I_kwDOHa8MBc6Zmp2a | 9,629 | https://github.com/huggingface/diffusers/issues/9629 | https://api.github.com/repos/huggingface/diffusers/issues/9629 | [Schedulers] Add methods for conversion between noise and denoised predictions | Guidance methods like APG (#9585 #9626) require conversion between noise and denoised predictions. The computation of these predictions varies across schedulers and depends on `prediction_type`. Currently, this is embedded in `step` functions of the schedulers which complicates implementation of APG.
Adding separate... | open | null | false | 5 | [
"stale",
"scheduler"
] | [] | 2024-10-09T21:59:35Z | 2024-11-09T15:02:48Z | null | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | hlky | 106,811,348 | U_kgDOBl3P1A | User | false |
huggingface/diffusers | 2,577,128,557 | I_kwDOHa8MBc6Zm9xt | 9,630 | https://github.com/huggingface/diffusers/issues/9630 | https://api.github.com/repos/huggingface/diffusers/issues/9630 | Error when setting num_single_layers=0 while training flux-controlnet on a multi-GPU server using a single GPU | ### Describe the bug
While training flux-controlnet on a multi-GPU server and restricting the training to a single GPU, setting **_num_single_layers=0_** leads to an error:
[rank0]: Parameter indices which did not receive grad for rank 0: 64 65 72 73 74 75
### Reproduction
`
accelerate launch --gpu_ids='0,' ... | open | null | false | 8 | [
"bug",
"stale"
] | [] | 2024-10-09T23:01:47Z | 2024-11-26T15:03:26Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | wangherr | 68,674,487 | MDQ6VXNlcjY4Njc0NDg3 | User | false |
huggingface/diffusers | 2,578,094,005 | I_kwDOHa8MBc6Zqpe1 | 9,633 | https://github.com/huggingface/diffusers/issues/9633 | https://api.github.com/repos/huggingface/diffusers/issues/9633 | Confusion about accelerator.num_processes in get_scheduler | In the example code from [train_text_to_image_sdxl.py](https://github.com/huggingface/diffusers/blob/e16fd93d0a40156c1f49fde07f6f2eb438983927/examples/text_to_image/train_text_to_image_sdxl.py#L974):
```python
num_warmup_steps = args.lr_warmup_steps * args.gradient_accumulation_steps
```
But in [train_text_to_image... | closed | completed | false | 5 | [
"stale"
] | [] | 2024-10-10T08:39:12Z | 2024-11-09T15:37:33Z | 2024-11-09T15:37:33Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | hj13-mtlab | 184,484,550 | U_kgDOCv8Cxg | User | false |
huggingface/diffusers | 2,578,713,715 | I_kwDOHa8MBc6ZtAxz | 9,635 | https://github.com/huggingface/diffusers/issues/9635 | https://api.github.com/repos/huggingface/diffusers/issues/9635 | [Flux ControlNet] Add support for de-distilled models with CFG | Flux with Inpaiting and ControlNet currently yields bad result with the base model.
To echo [this comment](https://github.com/huggingface/diffusers/pull/9571#issuecomment-2404857656), using de-distilled models could potentially help getting better outputs.
Currently the Flux ControlNet pipelines do not support mo... | open | null | false | 11 | [
"good first issue",
"contributions-welcome"
] | [] | 2024-10-10T12:33:01Z | 2026-03-22T06:15:45Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | simbrams | 25,414,628 | MDQ6VXNlcjI1NDE0NjI4 | User | false |
huggingface/diffusers | 2,579,156,814 | I_kwDOHa8MBc6Zus9O | 9,637 | https://github.com/huggingface/diffusers/issues/9637 | https://api.github.com/repos/huggingface/diffusers/issues/9637 | XFormer fails when passing attention mask while using bfloat and key's sequence length not being a multiple of 8 | ### Describe the bug
XFormer will fail when passing attention mask with its last dimension not being a multiple of 8 (i.e. key's sequence length) under bfloat16. This seems to be because xformer needs to attn_bias's stride to be a multiple of 8. Padding the attention mask on its last dimension will solve this probl... | open | null | false | 4 | [
"bug",
"stale"
] | [] | 2024-10-10T15:13:54Z | 2024-12-05T15:03:37Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | dhmbb2 | 123,251,854 | U_kgDOB1isjg | User | false |
huggingface/diffusers | 2,580,034,560 | I_kwDOHa8MBc6ZyDQA | 9,639 | https://github.com/huggingface/diffusers/issues/9639 | https://api.github.com/repos/huggingface/diffusers/issues/9639 | How to use my own trained lora in local computer? | local_model_path = r"D:\downloads\FLUX.1-schnell"
pipe = FluxPipeline.from_pretrained(local_model_path, torch_dtype=torch.bfloat16)
#lora not working by this way
pipe.load_lora_weights("XLabs-AI/flux-lora-collection", weight_name="disney_lora.safetensors")
pipe.load_lora_weights(r"D:\AI\stable-diffusion-webui-forg... | closed | completed | false | 6 | [] | [] | 2024-10-10T23:19:47Z | 2024-11-10T08:49:08Z | 2024-10-13T14:06:49Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | derekcbr | 31,892,214 | MDQ6VXNlcjMxODkyMjE0 | User | false |
huggingface/diffusers | 1,420,864,789 | I_kwDOHa8MBc5UsK0V | 964 | https://github.com/huggingface/diffusers/issues/964 | https://api.github.com/repos/huggingface/diffusers/issues/964 | Inpainting example not working | ### Describe the bug
```
Fetching 15 files: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████| 15/15 [00:00<00:00, 45756.04it/s]
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
Traceb... | closed | completed | false | 1 | [
"bug"
] | [] | 2022-10-24T13:49:06Z | 2022-10-24T14:22:54Z | 2022-10-24T14:22:54Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | seann999 | 6,847,529 | MDQ6VXNlcjY4NDc1Mjk= | User | false |
huggingface/diffusers | 2,580,228,414 | I_kwDOHa8MBc6Zyyk- | 9,641 | https://github.com/huggingface/diffusers/issues/9641 | https://api.github.com/repos/huggingface/diffusers/issues/9641 | Code With Strange Logic in CogVideoX's Dynamic CFG | ### Describe the bug
As shown at [pipeline_cogvideox_image2video.py L778](https://github.com/huggingface/diffusers/blob/38a3e4df926c59bc122191c0fc8066755e98b6d2/src/diffusers/pipelines/cogvideo/pipeline_cogvideox_image2video.py#L778C25-L778C118), [pipeline_cogvideox_video2video.py L778](https://github.com/huggingfac... | open | null | false | 3 | [
"bug",
"wip"
] | [] | 2024-10-11T01:56:14Z | 2024-11-20T01:56:24Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | immortalCO | 22,522,323 | MDQ6VXNlcjIyNTIyMzIz | User | false |
huggingface/diffusers | 2,580,359,918 | I_kwDOHa8MBc6ZzSru | 9,643 | https://github.com/huggingface/diffusers/issues/9643 | https://api.github.com/repos/huggingface/diffusers/issues/9643 | Flux does not support multiple Controlnets? | ### Describe the bug
I'm encountering an issue with the FluxControlNetPipeline. The `controlnet` parameter is supposed to accept a `List[FluxControlNetModel]`. However, when I attempt to execute my code, I run into the following error:
```
Traceback (most recent call last):
File "/opt/tiger/test_1/h.py", line... | closed | completed | false | 1 | [
"bug"
] | [] | 2024-10-11T03:47:06Z | 2024-10-11T17:39:20Z | 2024-10-11T17:39:20Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | RimoChan | 20,064,807 | MDQ6VXNlcjIwMDY0ODA3 | User | false |
huggingface/diffusers | 2,580,706,834 | I_kwDOHa8MBc6Z0nYS | 9,645 | https://github.com/huggingface/diffusers/issues/9645 | https://api.github.com/repos/huggingface/diffusers/issues/9645 | ImportError: cannot import name 'StableDiffusionLoraLoaderMixin' from 'diffusers.loaders' | ### Describe the bug
Hello, I've encountered the error as shown in the image since some hours ago (before that everything worked fine for me)
Could you check it for me? Thank you
### Reproduction
!pip install git+https:/... | closed | completed | false | 2 | [
"bug"
] | [] | 2024-10-11T07:52:50Z | 2024-10-11T13:51:49Z | 2024-10-11T13:51:49Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | alansmithee-cpu | 184,599,712 | U_kgDOCwDEoA | User | false |
huggingface/diffusers | 2,580,788,116 | I_kwDOHa8MBc6Z07OU | 9,648 | https://github.com/huggingface/diffusers/issues/9648 | https://api.github.com/repos/huggingface/diffusers/issues/9648 | enable_vae_tiling instead leads to memory overflow | ### Describe the bug
Using enable_vae_tiling instead causes an overflow of memory, and it is normal not to use it. This is very strange.
### Reproduction
```
from diffusers import AutoencoderKL
from diffusers.utils import load_image, numpy_to_pil
import PIL.Image
import os
import torch
from torchvision import... | closed | completed | false | 3 | [
"bug"
] | [] | 2024-10-11T08:34:45Z | 2024-10-11T13:57:58Z | 2024-10-11T13:57:58Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | TianmengChen | 110,011,222 | U_kgDOBo6jVg | User | false |
huggingface/diffusers | 1,421,242,624 | I_kwDOHa8MBc5UtnEA | 966 | https://github.com/huggingface/diffusers/issues/966 | https://api.github.com/repos/huggingface/diffusers/issues/966 | `UNet2DConditionModel` inheritance issue | ### Describe the bug
Running the Stable Diffusion in-painting example, I got an inheritance error
> RuntimeError: `@register_for_config` was applied to UNet2DConditionModel init method, but this class does not inherit from `ConfigMixin`.
### Reproduction
Code
```python
import PIL
import requests
impor... | closed | completed | false | 2 | [
"bug"
] | [] | 2022-10-24T18:27:38Z | 2022-11-09T04:38:22Z | 2022-11-09T04:38:22Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | vict0rsch | 9,283,470 | MDQ6VXNlcjkyODM0NzA= | User | false |
huggingface/diffusers | 2,583,127,410 | I_kwDOHa8MBc6Z92Vy | 9,660 | https://github.com/huggingface/diffusers/issues/9660 | https://api.github.com/repos/huggingface/diffusers/issues/9660 | The from_pretrained method of the DiffusionPipeline doesn't set the passed modules' mode to evaluation. | ### Describe the bug
When passing some modules to the from_pretrained method of DiffusionPipeline, this method will instantizate the pipeline using passed modules instead of loading from the cached folder (see [here](https://github.com/huggingface/diffusers/blob/6a5f06488c0d88c1827c016835cd5f64abe4b52c/src/diffusers/... | closed | completed | false | 5 | [
"bug",
"stale"
] | [] | 2024-10-12T14:49:18Z | 2024-11-11T16:40:04Z | 2024-11-11T16:40:04Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | wtyuan96 | 38,099,141 | MDQ6VXNlcjM4MDk5MTQx | User | false |
huggingface/diffusers | 2,583,385,279 | I_kwDOHa8MBc6Z-1S_ | 9,661 | https://github.com/huggingface/diffusers/issues/9661 | https://api.github.com/repos/huggingface/diffusers/issues/9661 | from_pretrained: filename argument removed? | **What API design would you like to have changed or added to the library? Why?**
I do believe there was a `filename` argument in the past to load a specific checkpoint in a huggingface repository. It appears that this has been removed with no replacement.
**What use case would this enable or better enable? Can yo... | closed | completed | false | 4 | [
"stale"
] | [] | 2024-10-12T20:02:31Z | 2024-11-13T00:37:52Z | 2024-11-13T00:37:52Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | oxysoft | 3,945,277 | MDQ6VXNlcjM5NDUyNzc= | User | false |
huggingface/diffusers | 2,584,344,815 | I_kwDOHa8MBc6aCfjv | 9,664 | https://github.com/huggingface/diffusers/issues/9664 | https://api.github.com/repos/huggingface/diffusers/issues/9664 | subprocess.CalledProcessError running train_dreambooth_lora_sdxl.py | ### Describe the bug
I'm trying to run train_dreambooth_lora_sdxl.py described here:
https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/README_sdxl.md
Also using this example:
https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/SDXL_DreamBooth_LoRA_.ipynb#scrollT... | closed | completed | false | 2 | [
"bug"
] | [] | 2024-10-13T22:53:41Z | 2024-10-15T23:58:56Z | 2024-10-15T23:58:55Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sebastianherreramonterrosa | 103,291,953 | U_kgDOBigcMQ | User | false |
huggingface/diffusers | 2,584,722,044 | I_kwDOHa8MBc6aD7p8 | 9,665 | https://github.com/huggingface/diffusers/issues/9665 | https://api.github.com/repos/huggingface/diffusers/issues/9665 | [Training] DeepSpeed config seems to be affecting `accelerator.prepare()` | `deepspeed.yaml`:
```yaml
compute_environment: LOCAL_MACHINE
debug: false
deepspeed_config:
gradient_accumulation_steps: 1
gradient_clipping: 1.0
offload_optimizer_device: cpu
offload_param_device: cpu
zero3_init_flag: false
zero_stage: 2
distributed_type: DEEPSPEED
downcast_bf16: 'no'
enable... | closed | completed | false | 3 | [] | [] | 2024-10-14T04:47:42Z | 2024-10-14T08:20:55Z | 2024-10-14T08:20:55Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sayakpaul | 22,957,388 | MDQ6VXNlcjIyOTU3Mzg4 | User | false |
huggingface/diffusers | 2,584,975,650 | I_kwDOHa8MBc6aE5ki | 9,666 | https://github.com/huggingface/diffusers/issues/9666 | https://api.github.com/repos/huggingface/diffusers/issues/9666 | smaller image encoder for IP Adapter | Are there smaller image encoder for IP Adapter?
Official IP Adapter repository using laion/CLIP-ViT-H-14-laion2B-s32B-b79K, that is very large in size. Are there smaller and faster image encoder that work with IP Adapter?
| closed | completed | true | 5 | [] | [] | 2024-10-14T06:54:51Z | 2024-10-15T14:33:32Z | 2024-10-15T14:33:32Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | kmpartner | 17,810,734 | MDQ6VXNlcjE3ODEwNzM0 | User | false |
huggingface/diffusers | 2,585,532,590 | I_kwDOHa8MBc6aHBiu | 9,667 | https://github.com/huggingface/diffusers/issues/9667 | https://api.github.com/repos/huggingface/diffusers/issues/9667 | Error in loading flux fp8 model with local transformer_flux.py file | ### Describe the bug
Unable to use flux fp8 model from `Kijai/flux-fp8` while having transformer_flux.py file in local. I have modified the scripts to remove any import error. I put some print statements in single_model_file.py to check why it is not loading the model.
### Reproduction
The below code works fin... | open | null | false | 2 | [
"bug",
"stale"
] | [] | 2024-10-14T10:17:08Z | 2024-11-13T15:02:50Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | DhavalTaunk08 | 31,320,833 | MDQ6VXNlcjMxMzIwODMz | User | false |
huggingface/diffusers | 1,421,243,589 | I_kwDOHa8MBc5UtnTF | 967 | https://github.com/huggingface/diffusers/issues/967 | https://api.github.com/repos/huggingface/diffusers/issues/967 | Schedulers not compatible with OnnxStableDiffusionPipeline: TypeError: unsupported operand | ### Describe the bug
Hi,
I tried to use different schedulers with OnnxStableDiffusionPipeline, but it throw errors.
Schedulers are not compatible with numpy used in onnx pipeline.
Onnx checkpoints converted with: [convert_stable_diffusion_checkpoint_to_onnx.py](https://raw.githubusercontent.com/huggingface/diffus... | closed | completed | false | 6 | [
"bug"
] | [
"anton-l"
] | 2022-10-24T18:28:27Z | 2022-11-08T13:39:13Z | 2022-11-08T13:39:13Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Tnifey | 8,442,019 | MDQ6VXNlcjg0NDIwMTk= | User | false |
huggingface/diffusers | 2,585,684,726 | I_kwDOHa8MBc6aHmr2 | 9,670 | https://github.com/huggingface/diffusers/issues/9670 | https://api.github.com/repos/huggingface/diffusers/issues/9670 | Could not load lora: PEFT backend is required for this method | ### Describe the bug
I am trying to use a lora with the StableDiffusion ControlNet Inpaint Pipeline but i am getting the error 'PEFT backend is required for this method.' even though PEFT is installed (globally and in the venv)
i am using 'path/to/lora' for the lora because the lora model does not matter, same er... | closed | completed | false | 2 | [
"bug"
] | [] | 2024-10-14T11:07:40Z | 2024-10-14T16:24:46Z | 2024-10-14T16:24:45Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | FOSSEnjoyer69 | 129,515,422 | U_kgDOB7g_ng | User | false |
huggingface/diffusers | 2,587,416,032 | I_kwDOHa8MBc6aONXg | 9,673 | https://github.com/huggingface/diffusers/issues/9673 | https://api.github.com/repos/huggingface/diffusers/issues/9673 | high cpu usage when loading multiple loras at once. | ### Describe the bug
Hi, I was making a synthesis system using celery and diffusers,
and I found the cpu usage of program goes high when loading loras,
it is okay when I use just one worker, but it becomes hard when using 8 workers at once.
It happens when lora loaded first time, and I think it is because of p... | closed | completed | false | 5 | [
"bug"
] | [] | 2024-10-15T01:49:37Z | 2024-10-15T05:07:40Z | 2024-10-15T05:05:48Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | gudwns1215 | 36,942,890 | MDQ6VXNlcjM2OTQyODkw | User | false |
huggingface/diffusers | 2,587,461,508 | I_kwDOHa8MBc6aOYeE | 9,674 | https://github.com/huggingface/diffusers/issues/9674 | https://api.github.com/repos/huggingface/diffusers/issues/9674 | Train sd3 controlnet but result is bad | ### Describe the bug
Hi,
I tried to train sd3 controlnet with `examples\controlnet\train_controlnet_sd3.py`, but it seems the validation results are really bad, and the training loss was oscillating all the time, it really confused me for a long time, could you please give some advice? Thanks a lot !
Here are s... | closed | completed | true | 1 | [
"bug"
] | [] | 2024-10-15T02:42:25Z | 2024-10-15T03:26:14Z | 2024-10-15T03:26:14Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | xduzhangjiayu | 73,270,275 | MDQ6VXNlcjczMjcwMjc1 | User | false |
huggingface/diffusers | 2,587,768,164 | I_kwDOHa8MBc6aPjVk | 9,677 | https://github.com/huggingface/diffusers/issues/9677 | https://api.github.com/repos/huggingface/diffusers/issues/9677 | keep text encoders in fp32 in flux | Hi,
I want to test outputs when text encoders are fp32 and pipeline is fp16 in FLUX according this [issue](https://github.com/huggingface/diffusers/pull/9097#issuecomment-2272292516). I write below code but get error.
code :
```
import torch
from diffusers import FluxPipeline
from transformers import T5Encod... | closed | not_planned | false | 5 | [] | [] | 2024-10-15T06:26:06Z | 2024-11-27T18:04:20Z | 2024-10-15T08:38:50Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | saeedkhanehgir | 65,589,645 | MDQ6VXNlcjY1NTg5NjQ1 | User | false |
huggingface/diffusers | 1,421,408,102 | I_kwDOHa8MBc5UuPdm | 968 | https://github.com/huggingface/diffusers/issues/968 | https://api.github.com/repos/huggingface/diffusers/issues/968 | Model loading is slow / Time before generating | ### Describe the bug
I have used a simple realization on T4 (Google Cloud)
```
import torch
from torch import autocast
from diffusers import StableDiffusionPipeline
access_token = ""
pipe = StableDiffusionPipeline.from_pretrained(
"CompVis/stable-diffusion-v1-4",
revision="fp16",
torch_dtype=t... | closed | completed | false | 6 | [
"bug"
] | [
"patrickvonplaten"
] | 2022-10-24T20:34:02Z | 2022-11-16T21:53:01Z | 2022-11-16T21:53:01Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Feniksss | 7,080,230 | MDQ6VXNlcjcwODAyMzA= | User | false |
huggingface/diffusers | 2,588,169,596 | I_kwDOHa8MBc6aRFV8 | 9,680 | https://github.com/huggingface/diffusers/issues/9680 | https://api.github.com/repos/huggingface/diffusers/issues/9680 | SDXL unconditional generation is broken | ### Describe the bug
Running SDXL generation with CFG scale 1.0 and 0.0 give the exact same results, but CFG scale 0.0 should perform unconditional generation.
### Reproduction
from diffusers import StableDiffusionXLPipeline, StableDiffusionXLImg2ImgPipeline
import torch
pipe = StableDiffusionXLPipeline.from_pre... | closed | completed | false | 3 | [
"bug"
] | [] | 2024-10-15T09:24:43Z | 2024-10-29T07:59:02Z | 2024-10-29T07:59:02Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | morrisalp | 8,263,996 | MDQ6VXNlcjgyNjM5OTY= | User | false |
huggingface/diffusers | 2,589,345,425 | I_kwDOHa8MBc6aVkaR | 9,686 | https://github.com/huggingface/diffusers/issues/9686 | https://api.github.com/repos/huggingface/diffusers/issues/9686 | [SD3 ControlNet] bug in pipeline 'controlnet_pooled_projections' | ### Describe the bug
Hi,
I think I found an issue that causes a misalignment between training and inference in SD3 ControlNet.
https://github.com/huggingface/diffusers/blob/a3e8d3f7deed140f57a28d82dd0b5d965bd0fb09/src/diffusers/pipelines/controlnet_sd3/pipeline_stable_diffusion_3_controlnet.py#L977
I think th... | open | null | false | 11 | [
"bug",
"stale"
] | [] | 2024-10-15T16:52:46Z | 2025-01-06T15:04:27Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | tobiasfshr | 36,965,290 | MDQ6VXNlcjM2OTY1Mjkw | User | false |
huggingface/diffusers | 1,421,466,771 | I_kwDOHa8MBc5UudyT | 969 | https://github.com/huggingface/diffusers/issues/969 | https://api.github.com/repos/huggingface/diffusers/issues/969 | Accelerate tests fail on 0.6.0 with more than 1 cuda devices | ### Describe the bug
Running the tests from the 0.6.0 tag, the accelerate tests fail with the following stack trace:
```
self = <tests.test_models_unet.UNetLDMModelTests testMethod=test_from_pretrained_accelerate_wont_change_results>
@require_accelerate
@unittest.skipIf(torch_device != "cuda", "This te... | closed | completed | false | 7 | [
"bug",
"wip"
] | [
"williamberman"
] | 2022-10-24T21:21:14Z | 2024-11-21T11:31:25Z | 2024-11-21T11:31:25Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | antoche | 1,627,384 | MDQ6VXNlcjE2MjczODQ= | User | false |
huggingface/diffusers | 2,590,859,478 | I_kwDOHa8MBc6abWDW | 9,690 | https://github.com/huggingface/diffusers/issues/9690 | https://api.github.com/repos/huggingface/diffusers/issues/9690 | FLUX.1-dev FP8 Example Code Cleanup | I was looking at the [FLUX.1-dev FP8 example code](https://huggingface.co/docs/diffusers/main/en/api/pipelines/flux#single-file-loading-for-the-fluxtransformer2dmodel) in the documentation and noticed some unnecessary imports and variable declarations, which make it appear longer and cluttered. Here is the shorter vers... | open | null | false | 8 | [] | [] | 2024-10-16T07:01:45Z | 2025-03-31T21:56:37Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | HashingTag | 24,617,729 | MDQ6VXNlcjI0NjE3NzI5 | User | false |
huggingface/diffusers | 2,591,017,486 | I_kwDOHa8MBc6ab8oO | 9,692 | https://github.com/huggingface/diffusers/issues/9692 | https://api.github.com/repos/huggingface/diffusers/issues/9692 | Starting cogview3-plus-3b reported an error | `###` Describe the bug
Starting the cogview3-plus model diffusers/models/modeling_utils.py line 774 yields an error TypeError: load_checkpoint_and_dispatch() got an unexpected keyword argument 'strict'. Line 783 exegesis 'strict=True,' can start the model
### Reproduction
```
from diffusers import CogView3Pl... | closed | completed | false | 2 | [
"bug"
] | [
"a-r-r-o-w"
] | 2024-10-16T08:04:18Z | 2024-10-17T07:44:48Z | 2024-10-17T07:44:48Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Malowking | 169,113,066 | U_kgDOChR16g | User | false |
huggingface/diffusers | 2,593,520,654 | I_kwDOHa8MBc6alfwO | 9,697 | https://github.com/huggingface/diffusers/issues/9697 | https://api.github.com/repos/huggingface/diffusers/issues/9697 | train_text_to_image_sdxl training effect is very poor | I use DeepSpeed for training: train_text_to_image_sdxl.py
1.The data volume is 231 pieces
2. deepspeed json
3.Training Script
![企业微信截图_17291362274700](https://github.com/user-attachments/assets/ae5a6207-dbc8-... | closed | completed | true | 2 | [] | [] | 2024-10-17T03:40:17Z | 2024-10-17T08:32:44Z | 2024-10-17T08:32:44Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | wzhiyuan2016 | 33,740,622 | MDQ6VXNlcjMzNzQwNjIy | User | false |
huggingface/diffusers | 2,593,577,154 | I_kwDOHa8MBc6altjC | 9,698 | https://github.com/huggingface/diffusers/issues/9698 | https://api.github.com/repos/huggingface/diffusers/issues/9698 | Unable to Retrieve Intermediate Gradients with CogVideoXPipeline | ### Describe the bug
When generating videos using the CogVideoXPipeline model, we need to access the gradients of intermediate tensors. However, we do not require additional training or parameter updates for the model.
We tried using register_forward_hook to capture the gradients, but this approach failed because t... | closed | completed | false | 4 | [
"bug"
] | [] | 2024-10-17T04:30:56Z | 2024-10-27T10:24:41Z | 2024-10-27T10:24:41Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | lovelyczli | 39,312,624 | MDQ6VXNlcjM5MzEyNjI0 | User | false |
huggingface/diffusers | 1,421,706,283 | I_kwDOHa8MBc5UvYQr | 970 | https://github.com/huggingface/diffusers/issues/970 | https://api.github.com/repos/huggingface/diffusers/issues/970 | Integration tests are not passing | ### Describe the bug
when I run integration tests, e.g. with `pytest -k test_stable_diffusion_inpaint_pipeline_fp16 tests\pipelines\stable_diffusion\test_stable_diffusion_inpaint.py` the tests fail as the output is not close enough to the expected output
### Reproduction
1. you have a windows 10 machine with p... | closed | completed | false | 2 | [
"bug"
] | [] | 2022-10-25T01:38:28Z | 2022-10-27T04:33:37Z | 2022-10-27T04:33:37Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Lewington-pitsos | 24,558,325 | MDQ6VXNlcjI0NTU4MzI1 | User | false |
huggingface/diffusers | 2,593,824,523 | I_kwDOHa8MBc6amp8L | 9,700 | https://github.com/huggingface/diffusers/issues/9700 | https://api.github.com/repos/huggingface/diffusers/issues/9700 | Flux inversion | current img2img is not so well, [RF Inversion](https://rf-inversion.github.io/)) provides an inverse method for Flux real image editing, can we implement it using diffusers?
or how can we use DDIM inversion in Flux? | closed | completed | false | 8 | [] | [
"linoytsaban"
] | 2024-10-17T07:03:59Z | 2024-12-17T16:00:30Z | 2024-12-17T16:00:29Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | yuxu915 | 94,051,337 | U_kgDOBZscCQ | User | false |
huggingface/diffusers | 2,595,479,428 | I_kwDOHa8MBc6as9-E | 9,704 | https://github.com/huggingface/diffusers/issues/9704 | https://api.github.com/repos/huggingface/diffusers/issues/9704 | RuntimeError: cuDNN Frontend error: [cudnn_frontend] Error: No execution plans support the graph. | ### Describe the bug
Hello. I tried the Img2Img Pipeline and encountered the error in the images. Could you please check it for me? Thank you
<img width="1363" alt="Screenshot 2024-10-17 at 11 39 30" src="https://github.com/user-attachments/assets/8128883f-ae9c-4e70-9ac0-437651fb9163">
<img width="1362" alt="Scree... | closed | completed | false | 59 | [
"bug",
"stale"
] | [] | 2024-10-17T18:44:42Z | 2025-09-26T08:26:00Z | 2024-12-13T06:16:57Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | alansmithee-cpu | 184,599,712 | U_kgDOCwDEoA | User | false |
huggingface/diffusers | 2,598,439,461 | I_kwDOHa8MBc6a4Qol | 9,713 | https://github.com/huggingface/diffusers/issues/9713 | https://api.github.com/repos/huggingface/diffusers/issues/9713 | Ctrlora implementation : Easy solution to train a controlnet with < 1000 examples | ### Model/Pipeline/Scheduler description
Authors of the paper trained a base controlnet (with a new architecture if I'm not mistaken) on 9 different conditions to allow finetuning on new conditions easily using a LoRA rank 128. This method allows finetuning a novel condition using less than 1000 examples with less t... | open | null | false | 5 | [
"stale",
"contributions-welcome"
] | [] | 2024-10-18T21:46:20Z | 2025-03-02T15:05:46Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | mamad-sd | 179,823,030 | U_kgDOCrfhtg | User | false |
huggingface/diffusers | 2,599,186,728 | I_kwDOHa8MBc6a7HEo | 9,717 | https://github.com/huggingface/diffusers/issues/9717 | https://api.github.com/repos/huggingface/diffusers/issues/9717 | Flux-dev docker image error | ### Describe the bug
```
PS F:\cog-flux-dev> docker images
REPOSITORY TAG IMAGE ID CREATED SIZE
flux-dev-model latest c19d0ffc3660 11 minutes ago 55.1GB
```
After build docker image..i have run that flux-dev image
Run command
```
docker run --gpus all -it flux-dev-model... | closed | completed | false | 6 | [
"bug"
] | [] | 2024-10-19T13:43:02Z | 2024-11-17T07:36:29Z | 2024-11-17T07:36:28Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | tzktok | 169,971,463 | U_kgDOCiGPBw | User | false |
huggingface/diffusers | 2,599,438,200 | I_kwDOHa8MBc6a8Ed4 | 9,719 | https://github.com/huggingface/diffusers/issues/9719 | https://api.github.com/repos/huggingface/diffusers/issues/9719 | `disable_progress_bar` is ignored for some models (Loading checkpoint shards) | ### Describe the bug
When loading some pipelines, `diffusers.utils.logging.disable_progress_bar()` doesn't disable all progress bars. In particular the "Loading checkpoint shards" progress bar still appears. The "Loading pipeline components..." progress bar, however, is disabled as expected. Models I found, where this... | closed | completed | false | 2 | [
"bug"
] | [] | 2024-10-19T17:42:37Z | 2024-10-19T19:29:12Z | 2024-10-19T19:29:12Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | JonasLoos | 33,965,649 | MDQ6VXNlcjMzOTY1NjQ5 | User | false |
huggingface/diffusers | 2,601,179,615 | I_kwDOHa8MBc6bCtnf | 9,726 | https://github.com/huggingface/diffusers/issues/9726 | https://api.github.com/repos/huggingface/diffusers/issues/9726 | FLUX.1-dev dreambooth problem trained on multigpu | ### Describe the bug
I tried to use accelerate and deepspeed to train flux, and it worked fine when using two L40s, but an error occurred when using two a100s. What is the reason?
### Reproduction
train.sh:
accelerate launch --config_file config.yaml train_flux.py \
--pretrained_model_name_or_path="./FLUX.1-dev"... | closed | completed | false | 0 | [
"bug"
] | [] | 2024-10-21T03:20:44Z | 2024-10-21T03:32:42Z | 2024-10-21T03:32:41Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | jyy-1998 | 55,531,840 | MDQ6VXNlcjU1NTMxODQw | User | false |
huggingface/diffusers | 2,601,208,682 | I_kwDOHa8MBc6bC0tq | 9,727 | https://github.com/huggingface/diffusers/issues/9727 | https://api.github.com/repos/huggingface/diffusers/issues/9727 | FLUX.1-dev dreambooth save problem trained on multigpu | ### Describe the bug
I tried to train flux using accelerate and deepspeed, but when using two L40s, the model could not be saved properly. What is the problem?
### Reproduction
train.sh:
accelerate launch --config_file config.yaml train_flux.py \
--pretrained_model_name_or_path="./FLUX.1-dev" \
--resolution=1... | closed | completed | false | 1 | [
"bug"
] | [] | 2024-10-21T03:37:23Z | 2024-10-29T06:38:00Z | 2024-10-29T06:38:00Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | jyy-1998 | 55,531,840 | MDQ6VXNlcjU1NTMxODQw | User | false |
huggingface/diffusers | 2,601,216,566 | I_kwDOHa8MBc6bC2o2 | 9,728 | https://github.com/huggingface/diffusers/issues/9728 | https://api.github.com/repos/huggingface/diffusers/issues/9728 | shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = emb.chunk(6, dim=1) ValueError: not enough values to unpack (expected 6, got 4) | ### Discussed in https://github.com/huggingface/diffusers/discussions/9710
<div type='discussions-op-text'>
<sup>Originally posted by **LianShuaiLong** October 18, 2024</sup>
To save GPU memory during training, I pre-extracted the text embeddings. During the training process, I did not load the text encoders and... | closed | completed | false | 1 | [] | [] | 2024-10-21T03:43:52Z | 2024-10-29T07:57:47Z | 2024-10-29T07:57:46Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | LianShuaiLong | 23,699,530 | MDQ6VXNlcjIzNjk5NTMw | User | false |
huggingface/diffusers | 2,601,356,295 | I_kwDOHa8MBc6bDYwH | 9,729 | https://github.com/huggingface/diffusers/issues/9729 | https://api.github.com/repos/huggingface/diffusers/issues/9729 | train_dreambooth_lora_flux_advanced ValueError: unexpected save model: <class 'transformers.models.t5.modeling_t5.T5EncoderModel'> | ### Describe the bug
Was exploring train_dreambooth_lora_flux_advanced.py from the advanced diffusion training examples. When I used --enable_t5_ti was getting the error ValueError: unexpected save model: <class 'transformers.models.t5.modeling_t5.T5EncoderModel'>
### Reproduction
export MODEL_NAME="black-forest-la... | closed | completed | false | 5 | [
"bug"
] | [] | 2024-10-21T05:17:04Z | 2024-12-06T07:59:33Z | 2024-10-28T10:52:28Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | hackkhai | 51,231,270 | MDQ6VXNlcjUxMjMxMjcw | User | false |
huggingface/diffusers | 1,422,148,930 | I_kwDOHa8MBc5UxEVC | 973 | https://github.com/huggingface/diffusers/issues/973 | https://api.github.com/repos/huggingface/diffusers/issues/973 | [ Feature request] Stable Diffusion 1.5 inpainting without prompt | Use stable-diffusion 1.5 inpainting model without prompt to do regular inpainting. Example image come from [latent-diffusion](https://github.com/CompVis/latent-diffusion).
| closed | completed | false | 5 | [
"stale"
] | [
"patil-suraj"
] | 2022-10-25T09:24:56Z | 2022-11-24T19:12:16Z | 2022-11-24T19:12:16Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Sanster | 3,998,421 | MDQ6VXNlcjM5OTg0MjE= | User | false |
huggingface/diffusers | 2,602,356,770 | I_kwDOHa8MBc6bHNAi | 9,731 | https://github.com/huggingface/diffusers/issues/9731 | https://api.github.com/repos/huggingface/diffusers/issues/9731 | How to use Playground2.5 to train lora with own dataset to generate pictures of a specific style？ | ### Describe the bug
Hi,
I have been working on training models using the same dataset as "stabilityai/stable-diffusion-xl-base-1.0" with the script examples/text_to_image/train_text_to_image_lora_sdxl.py, and I achieved quite promising results.
Now, I am trying to further improve the performance by switching to... | open | null | false | 1 | [
"bug",
"stale"
] | [] | 2024-10-21T12:10:12Z | 2024-11-20T15:03:04Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | hjw-0909 | 56,182,057 | MDQ6VXNlcjU2MTgyMDU3 | User | false |
huggingface/diffusers | 2,602,828,908 | I_kwDOHa8MBc6bJARs | 9,732 | https://github.com/huggingface/diffusers/issues/9732 | https://api.github.com/repos/huggingface/diffusers/issues/9732 | With examples/dreambooth /README_flux.md guide setting up and training, got cuda OOM with 3090Ti 24GB | ### Describe the bug
Followed the guide examples/dreambooth/README_flux.md guide setting up and training, got cuda OOM with 3090Ti 24GB.
### Reproduction
PC got 256GB RAM
3090Ti VRAM 24GB
torch 2.4.1 + cuda 12.1
export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
accelerate==1.0.1
transformers==4.45.2
### ... | closed | completed | false | 18 | [
"bug",
"stale"
] | [] | 2024-10-21T14:59:26Z | 2025-02-21T09:19:01Z | 2024-12-03T16:16:02Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | riflemanl | 23,468,948 | MDQ6VXNlcjIzNDY4OTQ4 | User | false |
huggingface/diffusers | 2,603,274,148 | I_kwDOHa8MBc6bKs-k | 9,733 | https://github.com/huggingface/diffusers/issues/9733 | https://api.github.com/repos/huggingface/diffusers/issues/9733 | Xlabs controlnets do not work with FluxControlNetInpaintPipeline | ### Describe the bug
errors when Xlabs controlnets are used with FluxControlNetInpaintPipeline
### Reproduction
controlnet = FluxControlNet.from_pretrained(“xlabs…”)
pipe = FluxControlNetInpaintPipeline.from_pretrained(…, controlnet=controlnet)
image = pipe(…)
### Logs
```shell
Given groups=1, weight size of [... | closed | completed | false | 11 | [
"bug",
"good first issue",
"contributions-welcome"
] | [] | 2024-10-21T17:53:43Z | 2024-10-25T21:50:57Z | 2024-10-25T21:50:57Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | neuron-party | 96,799,331 | U_kgDOBcUKYw | User | false |
huggingface/diffusers | 2,605,484,708 | I_kwDOHa8MBc6bTIqk | 9,740 | https://github.com/huggingface/diffusers/issues/9740 | https://api.github.com/repos/huggingface/diffusers/issues/9740 | Add Ultimate SD Upscale pipeline for high-quality tiled image upscaling | **Is your feature request related to a problem? Please describe.**
Currently, diffusers library lacks advanced tiled upscaling capabilities that are available in other Stable Diffusion implementations. While the library supports basic img2img and upscaling, there's no built-in solution for handling large images throug... | open | null | false | 22 | [] | [
"asomoza"
] | 2024-10-22T13:36:27Z | 2025-04-05T23:17:39Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | BasimBashir | 45,428,008 | MDQ6VXNlcjQ1NDI4MDA4 | User | false |
huggingface/diffusers | 2,606,364,221 | I_kwDOHa8MBc6bWfY9 | 9,744 | https://github.com/huggingface/diffusers/issues/9744 | https://api.github.com/repos/huggingface/diffusers/issues/9744 | Add Mochi Video Model | ### Model/Pipeline/Scheduler description
A new state-of-the-art open-source video model was released today under the Apache-2.0 license: Mochi.
Would be fantastic to get support for this model via a pipeline in Diffusers.
[Blog](https://www.genmo.ai/blog) | [Hugging Face](https://huggingface.co/genmo/mochi-1-pre... | closed | completed | false | 2 | [] | [
"a-r-r-o-w"
] | 2024-10-22T19:58:23Z | 2024-11-05T15:03:43Z | 2024-11-05T15:03:43Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ghunkins | 12,562,057 | MDQ6VXNlcjEyNTYyMDU3 | User | false |
huggingface/diffusers | 2,606,834,841 | I_kwDOHa8MBc6bYSSZ | 9,745 | https://github.com/huggingface/diffusers/issues/9745 | https://api.github.com/repos/huggingface/diffusers/issues/9745 | Problem when install the nodes.. [ERROR] An error occurred while retrieving information for the 'controlaux... | FETCH DATA from: D:\SD\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Manager\extension-node-map.json
[ERROR] An error occurred while retrieving information for the 'controlaux_hed' node.
Traceback (most recent call last):
File "D:\SD\ComfyUI_windows_portable\ComfyUI\server.py", line 564, in get_object_info... | closed | completed | false | 2 | [] | [] | 2024-10-23T00:45:40Z | 2024-10-23T00:50:35Z | 2024-10-23T00:50:35Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | inglob | 124,811,471 | U_kgDOB3B4zw | User | false |
huggingface/diffusers | 2,607,347,884 | I_kwDOHa8MBc6baPis | 9,748 | https://github.com/huggingface/diffusers/issues/9748 | https://api.github.com/repos/huggingface/diffusers/issues/9748 | 关于提交信息国家一栏选项错误 | 
![image](https://github.com/user-attachments/as... | closed | completed | false | 1 | [] | [] | 2024-10-23T05:47:29Z | 2024-10-23T20:53:39Z | 2024-10-23T20:53:39Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | tangyuming99 | 166,014,357 | U_kgDOCeUtlQ | User | false |
huggingface/diffusers | 2,607,599,144 | I_kwDOHa8MBc6bbM4o | 9,750 | https://github.com/huggingface/diffusers/issues/9750 | https://api.github.com/repos/huggingface/diffusers/issues/9750 | Is it possible to provide img2img code for CogView3? | Is it possible to provide img2img code for CogView3? | open | null | false | 3 | [
"stale",
"contributions-welcome"
] | [] | 2024-10-23T07:40:38Z | 2024-12-20T15:04:01Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ChalvYongkang | 82,589,387 | MDQ6VXNlcjgyNTg5Mzg3 | User | false |
huggingface/diffusers | 2,608,532,268 | I_kwDOHa8MBc6bewss | 9,756 | https://github.com/huggingface/diffusers/issues/9756 | https://api.github.com/repos/huggingface/diffusers/issues/9756 | Could not find loading_adapters.ipynb | ### Describe the bug
while reading doc [Load adapters](https://huggingface.co/docs/diffusers/using-diffusers/loading_adapters)
I tried to open in Colab to run an example on this page.
<img width="504" alt="open_colab" src="https://github.com/user-attachments/assets/0b1397f1-d266-4d83-84ab-276ea796a2a4">
I... | closed | completed | false | 6 | [
"bug"
] | [] | 2024-10-23T13:03:11Z | 2024-11-01T15:27:56Z | 2024-11-01T15:27:56Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | thliang01 | 21,286,104 | MDQ6VXNlcjIxMjg2MTA0 | User | false |
huggingface/diffusers | 2,610,499,600 | I_kwDOHa8MBc6bmRAQ | 9,761 | https://github.com/huggingface/diffusers/issues/9761 | https://api.github.com/repos/huggingface/diffusers/issues/9761 | [docstring] Incorrect docstring reference for `return_dict` in `pipeline_stable_diffusion_3.py` | Hi there,
I noticed a small issue in the docstring for the `return_dict` parameter in `StableDiffusion3Pipeline` class [here](https://github.com/huggingface/diffusers/blob/24c7d578baf6a8b79890101dd280278fff031d12/src/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3.py#L765).
Seems like this shou... | closed | completed | false | 2 | [] | [] | 2024-10-24T05:20:43Z | 2024-10-25T11:03:36Z | 2024-10-25T11:03:36Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | shenzhiy21 | 90,144,843 | MDQ6VXNlcjkwMTQ0ODQz | User | false |
huggingface/diffusers | 2,611,460,879 | I_kwDOHa8MBc6bp7sP | 9,763 | https://github.com/huggingface/diffusers/issues/9763 | https://api.github.com/repos/huggingface/diffusers/issues/9763 | Flux Controlnet Train will truncate more than 77 Token descriptions | ### Describe the bug
At the start of the training, an excessively long description will cause an error to be reported, suggesting that ：The following part of your input was truncated because CLIP can only handle sequences up to 77 tokens:
`Map: 16%|█████████████████▍ ... | closed | completed | false | 4 | [
"bug"
] | [] | 2024-10-24T12:40:11Z | 2024-10-24T13:18:21Z | 2024-10-24T13:18:20Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ScilenceForest | 45,549,187 | MDQ6VXNlcjQ1NTQ5MTg3 | User | false |
huggingface/diffusers | 2,611,588,178 | I_kwDOHa8MBc6bqaxS | 9,764 | https://github.com/huggingface/diffusers/issues/9764 | https://api.github.com/repos/huggingface/diffusers/issues/9764 | EMAModel class bug: "from_pretrained" method bug | ### Describe the bug
```python
@classmethod
def from_pretrained(cls, path, model_cls, foreach=False) -> "EMAModel":
_, ema_kwargs = model_cls.load_config(path, return_unused_kwargs=True)
model = model_cls.from_pretrained(path)
ema_model = cls(model.parameters(), model_cls=model_cls, model_config=mod... | closed | completed | false | 3 | [
"bug"
] | [] | 2024-10-24T13:17:16Z | 2024-11-01T03:50:21Z | 2024-11-01T03:50:21Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | wangyanhui666 | 48,179,299 | MDQ6VXNlcjQ4MTc5Mjk5 | User | false |
huggingface/diffusers | 2,613,024,872 | I_kwDOHa8MBc6bv5ho | 9,767 | https://github.com/huggingface/diffusers/issues/9767 | https://api.github.com/repos/huggingface/diffusers/issues/9767 | cuDNN Frontend error running LoRA on FluxImg2ImgPipeLine | ### Describe the bug
Trying to apply a lora to image in FluxImg2ImgPipeline but keep receiving the following error: RuntimeError: cuDNN Frontend error: [cudnn_frontend] Error: No execution plans support the graph.
### Reproduction
Deployed this on Modal
Using an image to replicate environment: flux_dev_image ... | closed | completed | false | 3 | [
"bug"
] | [] | 2024-10-25T04:11:32Z | 2024-10-25T18:25:45Z | 2024-10-25T18:25:45Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | shridharathi | 60,528,857 | MDQ6VXNlcjYwNTI4ODU3 | User | false |
huggingface/diffusers | 2,613,078,947 | I_kwDOHa8MBc6bwGuj | 9,768 | https://github.com/huggingface/diffusers/issues/9768 | https://api.github.com/repos/huggingface/diffusers/issues/9768 | torch 2.5 CuDNN backend for SDPA NaN error | ### Describe the bug
When using the recently released PyTorch 2.5, the default SDPA backend is CUDNN_ATTENTION. In the example's CogVideoX-lora [training script](https://github.com/huggingface/diffusers/tree/main/examples/cogvideo), NaN gradients occur right at the first step. However, using other SDPA backends, such ... | closed | completed | false | 7 | [
"bug"
] | [] | 2024-10-25T04:58:59Z | 2024-11-02T06:42:46Z | 2024-11-02T06:42:45Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | wtyuan96 | 38,099,141 | MDQ6VXNlcjM4MDk5MTQx | User | false |
huggingface/diffusers | 2,614,746,996 | I_kwDOHa8MBc6b2d90 | 9,772 | https://github.com/huggingface/diffusers/issues/9772 | https://api.github.com/repos/huggingface/diffusers/issues/9772 | Support ControlNetPlus Union if not already supported | It's not clear if ControlNetPlus is already supported by diffusers https://github.com/xinsir6/ControlNetPlus/tree/main/pipeline which consists of union controlnet for SDXL. This model seems to support the only SDXL segmentation that I'm aware of. If not already supported, it should be!
https://github.com/xinsir6/Con... | closed | completed | false | 5 | [
"help wanted",
"Good second issue",
"contributions-welcome"
] | [
"yiyixuxu"
] | 2024-10-25T17:43:43Z | 2024-12-11T17:07:54Z | 2024-12-11T17:07:54Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | jloveric | 5,200,721 | MDQ6VXNlcjUyMDA3MjE= | User | false |
huggingface/diffusers | 2,615,377,723 | I_kwDOHa8MBc6b4387 | 9,773 | https://github.com/huggingface/diffusers/issues/9773 | https://api.github.com/repos/huggingface/diffusers/issues/9773 | text_input_ids_list not part of encode_prompt definition | Description
code is trying to pass `text_input_ids_list` as argument for `encode_prompt`, but, `text_input_ids_list` isn't supposed to be part of argument as per the definition.
Call
https://github.com/huggingface/diffusers/blob/298ab6eb01f3ef475c15218ea87de1494e1250aa/examples/dreambooth/train_dreambooth_sd3.py#L... | closed | completed | false | 2 | [] | [] | 2024-10-26T00:02:09Z | 2024-11-06T01:08:51Z | 2024-11-06T01:08:51Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Vatsal-Malaviya | 77,090,011 | MDQ6VXNlcjc3MDkwMDEx | User | false |
huggingface/diffusers | 2,616,031,670 | I_kwDOHa8MBc6b7Xm2 | 9,778 | https://github.com/huggingface/diffusers/issues/9778 | https://api.github.com/repos/huggingface/diffusers/issues/9778 | Run tutorial bug | ### Describe the bug
when I follow the tutorial in dreambooth folder, I encounter this error:
accelerate launch train_dreambooth_lora_sdxl.py \
> --pretrained_model_name_or_path=$MODEL_NAME \
> --instance_data_dir=$INSTANCE_DIR \
> --pretrained_vae_model_name_or_path=$VAE_PATH \
> --output_dir=$OUTPUT... | closed | completed | false | 2 | [
"bug"
] | [] | 2024-10-26T17:47:45Z | 2024-11-17T07:34:17Z | 2024-11-17T07:34:17Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | jiqibuaixuexi | 102,913,931 | U_kgDOBiJXiw | User | false |
huggingface/diffusers | 2,616,228,171 | I_kwDOHa8MBc6b8HlL | 9,780 | https://github.com/huggingface/diffusers/issues/9780 | https://api.github.com/repos/huggingface/diffusers/issues/9780 | Value error: has no attribute norm_added_k --- while trying to load stable-diffusion-3.5-large using StableDiffusion3Pipeline | ### Describe the bug
After the model files being finished with the downloads, I received two warnings:
_**1. You set `add_prefix_space`. The tokenizer needs to be converted from the slow tokenizers
2. The config attributes {'qk_norm': 'rms_norm'} were passed to SD3Transformer2DModel, but are not expected and will b... | closed | completed | false | 2 | [
"bug"
] | [] | 2024-10-27T00:19:02Z | 2024-10-27T15:02:12Z | 2024-10-27T15:02:12Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Hasnat79 | 32,153,220 | MDQ6VXNlcjMyMTUzMjIw | User | false |
huggingface/diffusers | 2,617,567,095 | I_kwDOHa8MBc6cBOd3 | 9,783 | https://github.com/huggingface/diffusers/issues/9783 | https://api.github.com/repos/huggingface/diffusers/issues/9783 | Support SOTA Difussion Based talking head repos like Echomimic if not already supported | I have seen that Hugging Face doesnt have much of the Talking Head Repositories under its hood. Would love to contribute and integrate it within the diffusers pipeline.
Link to the original github repo - https://github.com/BadToBest/EchoMimic . | open | null | false | 2 | [
"stale"
] | [] | 2024-10-28T07:35:34Z | 2024-11-27T15:02:54Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | souvikg544 | 63,863,911 | MDQ6VXNlcjYzODYzOTEx | User | false |
huggingface/diffusers | 2,620,277,723 | I_kwDOHa8MBc6cLkPb | 9,790 | https://github.com/huggingface/diffusers/issues/9790 | https://api.github.com/repos/huggingface/diffusers/issues/9790 | FLUX.1-dev dreambooth training problem on multigpu | ### Describe the bug
I tried to use accelerate+deepspeed to train flux, but every time after a dozen steps, an error occurred and the program crashed. Can anyone provide some help?
### Reproduction
accelerate launch --config_file config.yaml train_flux.py
--pretrained_model_name_or_path="./FLUX.1-dev"
--resolution... | closed | completed | false | 6 | [
"bug",
"stale"
] | [] | 2024-10-29T06:37:11Z | 2024-11-28T15:06:45Z | 2024-11-28T15:06:45Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | jyy-1998 | 55,531,840 | MDQ6VXNlcjU1NTMxODQw | User | false |
huggingface/diffusers | 2,620,494,424 | I_kwDOHa8MBc6cMZJY | 9,792 | https://github.com/huggingface/diffusers/issues/9792 | https://api.github.com/repos/huggingface/diffusers/issues/9792 | Cannot load the cache when mapping the dataset | ### Describe the bug
I'm training the flux controlnet. The train_dataset.map() takes long time to finish. However, when I killed one training process and want to restart a new training with the same dataset. I can't reuse the mapped result even I defined the cache dir for the dataset.
### Reproduction
train flux con... | closed | completed | false | 2 | [
"bug"
] | [] | 2024-10-29T08:21:07Z | 2024-11-01T03:23:21Z | 2024-11-01T03:22:50Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | zhangn77 | 43,033,959 | MDQ6VXNlcjQzMDMzOTU5 | User | false |
huggingface/diffusers | 2,620,755,140 | I_kwDOHa8MBc6cNYzE | 9,793 | https://github.com/huggingface/diffusers/issues/9793 | https://api.github.com/repos/huggingface/diffusers/issues/9793 | Progress bar does not update | ### Describe the bug
The progress bars do not update.
### Reproduction
```
from diffusers import DiffusionPipeline, DPMSolverMultistepScheduler
import torch
repo_id = "stabilityai/stable-diffusion-2-base"
pipe = DiffusionPipeline.from_pretrained(repo_id, torch_dtype=torch.float16, variant="fp16")
pipe.sched... | closed | completed | false | 3 | [
"bug",
"stale"
] | [] | 2024-10-29T10:02:30Z | 2024-11-28T18:14:36Z | 2024-11-28T18:14:36Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sorenwacker | 3,391,614 | MDQ6VXNlcjMzOTE2MTQ= | User | false |
huggingface/diffusers | 2,620,999,034 | I_kwDOHa8MBc6cOUV6 | 9,794 | https://github.com/huggingface/diffusers/issues/9794 | https://api.github.com/repos/huggingface/diffusers/issues/9794 | Add support to Meissonic | Meissonic is a non-autoregressive mask image modeling text-to-image synthesis model that can generate high-resolution images. It is designed to run on consumer graphics cards.
The model checkpoint can be found in https://huggingface.co/MeissonFlow/Meissonic
The inference code can be found in https://github.com/viii... | open | null | false | 7 | [
"stale",
"contributions-welcome"
] | [] | 2024-10-29T11:46:45Z | 2024-12-01T15:03:12Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | viiika | 39,821,659 | MDQ6VXNlcjM5ODIxNjU5 | User | false |
huggingface/diffusers | 2,621,745,283 | I_kwDOHa8MBc6cRKiD | 9,797 | https://github.com/huggingface/diffusers/issues/9797 | https://api.github.com/repos/huggingface/diffusers/issues/9797 | Loading pipeline in precision it was saved in | **Is your feature request related to a problem? Please describe.**
Currently, if `torch_dtype` is not specified, the pipeline defaults to loading in `float32`. This behavior causes `float16` or `bfloat16` weights to be upcast to `float32` when the model is saved in lower precision, leading to increased memory usage. I... | open | null | false | 10 | [
"enhancement"
] | [] | 2024-10-29T16:17:02Z | 2024-12-01T22:11:13Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | mvafin | 16,835,333 | MDQ6VXNlcjE2ODM1MzMz | User | false |
huggingface/diffusers | 2,621,788,428 | I_kwDOHa8MBc6cRVEM | 9,798 | https://github.com/huggingface/diffusers/issues/9798 | https://api.github.com/repos/huggingface/diffusers/issues/9798 | Replace shutil.copy with shutil.copyfile | shutil.copy copies permission bits which fails when the user who's running the script is trying to use a common cache that was generated by another user, even though the first user has read & write permissions over the cache (through Group permission for example). A real case scenario: submitting jobs on a GPU cluster ... | closed | completed | false | 4 | [] | [] | 2024-10-29T16:35:48Z | 2024-11-08T18:32:33Z | 2024-11-08T18:32:33Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | almarouk | 72,821,992 | MDQ6VXNlcjcyODIxOTky | User | false |
huggingface/diffusers | 2,621,855,622 | I_kwDOHa8MBc6cRleG | 9,799 | https://github.com/huggingface/diffusers/issues/9799 | https://api.github.com/repos/huggingface/diffusers/issues/9799 | RuntimeError: The size of tensor a (4173) must match the size of tensor b (16461) at non-singleton dimension 2 | ### Describe the bug
I am trying to work on the flux lora quantization example as per the link
https://github.com/huggingface/diffusers/tree/main/examples/research_projects/flux_lora_quantization
but facing RuntimeError: The size of tensor a (4173) must match the size of tensor b (16461) at non-singleton dimension... | closed | completed | false | 3 | [
"bug"
] | [] | 2024-10-29T17:03:47Z | 2024-11-01T03:18:28Z | 2024-11-01T03:18:28Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | TharunSivamani | 77,732,018 | MDQ6VXNlcjc3NzMyMDE4 | User | false |
huggingface/diffusers | 2,622,236,543 | I_kwDOHa8MBc6cTCd_ | 9,802 | https://github.com/huggingface/diffusers/issues/9802 | https://api.github.com/repos/huggingface/diffusers/issues/9802 | Multidiffusion (panorama pipeline) is missing segmentation inputs? | I'm looking at the multidiffusion panorama pipeline page (https://huggingface.co/docs/diffusers/en/api/pipelines/panorama). It looks like there is no way to specify the segmentation and associated prompts as in the original paper https://multidiffusion.github.io/ . If the code only has the panorama capability and not t... | open | null | false | 5 | [
"stale"
] | [] | 2024-10-29T20:15:15Z | 2024-12-24T15:03:30Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | jloveric | 5,200,721 | MDQ6VXNlcjUyMDA3MjE= | User | false |
huggingface/diffusers | 2,622,588,401 | I_kwDOHa8MBc6cUYXx | 9,803 | https://github.com/huggingface/diffusers/issues/9803 | https://api.github.com/repos/huggingface/diffusers/issues/9803 | Flux LoRA loading failure | ### Describe the bug
When loading in [this model](https://civitai.com/models/631986) you get a `ValueError(f"Incompatible keys detected: \n\n {', '.join(remaining_keys)}")` error
It looks like it's a LoRA type that doesn't fall into any of the classic conversion script paths.
It looks like a Flux diffusers ch... | open | null | false | 7 | [
"bug"
] | [] | 2024-10-30T00:34:13Z | 2025-04-28T09:32:55Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | jonluca | 13,029,040 | MDQ6VXNlcjEzMDI5MDQw | User | false |
huggingface/diffusers | 2,622,604,144 | I_kwDOHa8MBc6cUcNw | 9,804 | https://github.com/huggingface/diffusers/issues/9804 | https://api.github.com/repos/huggingface/diffusers/issues/9804 | ControlNetModel.from_pretrained doesn't load sdxl controlnet model from local | ### Describe the bug
can't load model after train
the train script is
https://github.com/huggingface/diffusers/blob/main/examples/controlnet/README_sdxl.md
<img width="379" alt="image" src="https://github.com/user-attachments/assets/d33b5716-5ea9-4fb7-b5e6-e12f3c528fb1">
### Reproduction
```
impor... | closed | completed | false | 1 | [
"bug"
] | [] | 2024-10-30T00:52:22Z | 2024-10-30T01:15:14Z | 2024-10-30T01:15:14Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | crapthings | 1,147,704 | MDQ6VXNlcjExNDc3MDQ= | User | false |
huggingface/diffusers | 2,623,670,454 | I_kwDOHa8MBc6cYgi2 | 9,810 | https://github.com/huggingface/diffusers/issues/9810 | https://api.github.com/repos/huggingface/diffusers/issues/9810 | Lora does not support module_to_save() | ### Describe the bug
When we set
```
LoraConfig(
r=args.rank,
lora_alpha=args.lora_alpha,
init_lora_weights=True,
target_modules=["to_k", "to_q", "to_v", "to_out.0"],
modules_to_save=["other"],
)
```
if `modules_to_save` is not none, then we load the adapter,`diff... | open | null | false | 2 | [
"bug",
"stale"
] | [] | 2024-10-30T10:50:30Z | 2024-11-29T15:02:55Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | liuchaohu | 41,250,082 | MDQ6VXNlcjQxMjUwMDgy | User | false |
huggingface/diffusers | 2,624,398,831 | I_kwDOHa8MBc6cbSXv | 9,813 | https://github.com/huggingface/diffusers/issues/9813 | https://api.github.com/repos/huggingface/diffusers/issues/9813 | `prepare_ip_adapter_image_embeds` Bug Causes Feature Mixing During Batch Processing in IP-Adapter | ### Describe the bug
**The `prepare_ip_adapter_image_embeds ` function has a bug that results in unintended feature mixing across images during batch processing. This issue causes the generated images to combine features from multiple reference images, instead of maintaining a one-to-one correspondence with each ref... | open | null | false | 4 | [
"bug"
] | [
"asomoza",
"yiyixuxu"
] | 2024-10-30T15:01:55Z | 2025-05-15T15:59:14Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | heiheizwplus | 175,671,121 | U_kgDOCniHUQ | User | false |
huggingface/diffusers | 2,626,621,312 | I_kwDOHa8MBc6cjw-A | 9,819 | https://github.com/huggingface/diffusers/issues/9819 | https://api.github.com/repos/huggingface/diffusers/issues/9819 | add skip layer guidance to sd 3.5 | stabilityai recently added support to skip_layer_guidance in their official implementation via this pr:
<https://github.com/Stability-AI/sd3.5/pull/11/files>
slg adds extra guidance to the original cfg during specific steps.
its calculated as positive output from original model minus positive model output from a m... | closed | completed | false | 3 | [] | [] | 2024-10-31T11:28:02Z | 2024-11-19T23:02:21Z | 2024-11-19T23:02:21Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | vladmandic | 57,876,960 | MDQ6VXNlcjU3ODc2OTYw | User | false |
huggingface/diffusers | 1,422,862,815 | I_kwDOHa8MBc5Uzynf | 982 | https://github.com/huggingface/diffusers/issues/982 | https://api.github.com/repos/huggingface/diffusers/issues/982 | Dance Diffusion in FP16 | At the moment dance diffusion in FP16 seems to perform relatively bad even though it was trained in FP16 with AMP autocast -> the reason might just be because the layernorms and/or attention scores have to be done in fp32. | closed | completed | false | 2 | [
"stale"
] | [] | 2022-10-25T17:58:41Z | 2022-12-03T15:03:13Z | 2022-12-03T15:03:13Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/diffusers | 2,627,502,481 | I_kwDOHa8MBc6cnIGR | 9,822 | https://github.com/huggingface/diffusers/issues/9822 | https://api.github.com/repos/huggingface/diffusers/issues/9822 | Loading SDXL loras into Flux | ### Describe the bug
Currently it's possible to load SDXL loras without warning into Flux.
### Reproduction
Is it possible for you to implement a raise a warning (and an error when a boolean is active) when the list of layers here is zero:
https://github.com/huggingface/diffusers/blob/41e4779d988ead99e7acd78dc8e7... | closed | completed | false | 8 | [
"bug"
] | [
"sayakpaul"
] | 2024-10-31T18:01:29Z | 2024-12-10T14:37:32Z | 2024-12-10T14:37:31Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | christopher5106 | 6,875,375 | MDQ6VXNlcjY4NzUzNzU= | User | false |
huggingface/diffusers | 2,627,978,082 | I_kwDOHa8MBc6co8Ni | 9,825 | https://github.com/huggingface/diffusers/issues/9825 | https://api.github.com/repos/huggingface/diffusers/issues/9825 | Support IPAdapters for FLUX pipelines | ### Model/Pipeline/Scheduler description
IPAdapter for FLUX is available now, do you have any plans to add IPAdapter to FLUX pipelines?
### Open source status
- [X] The model implementation is available.
- [X] The model weights are available (Only relevant if addition is not a scheduler).
### Provide useful links f... | closed | completed | false | 10 | [
"help wanted",
"wip",
"contributions-welcome",
"IPAdapter"
] | [
"asomoza"
] | 2024-10-31T23:07:32Z | 2024-12-21T17:49:59Z | 2024-12-21T17:49:59Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | chenxiao111222 | 154,797,505 | U_kgDOCToFwQ | User | false |
huggingface/diffusers | 2,628,503,075 | I_kwDOHa8MBc6cq8Yj | 9,831 | https://github.com/huggingface/diffusers/issues/9831 | https://api.github.com/repos/huggingface/diffusers/issues/9831 | diffusers库能支持20B、30B的DIT视频生成模型训练么 | 想了解下 ，diffusers库能支持20B、30B的DIT视频生成模型训练么 | closed | completed | false | 3 | [] | [] | 2024-11-01T07:31:06Z | 2024-11-17T07:32:12Z | 2024-11-17T07:32:12Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | QingQingS | 150,359,873 | U_kgDOCPZPQQ | User | false |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.