repo stringclasses 1
value | github_id int64 1.27B 4.63B | 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 2026-06-10 12:14:31 | updated_at stringdate 2022-06-12 22:18:01 2026-06-10 22:20:39 | closed_at stringdate 2022-06-12 22:18:01 2026-06-10 12:07:08 ⌀ | author_association stringclasses 3
values | milestone_title stringclasses 0
values | snapshot_id stringclasses 70
values | extracted_at stringdate 2026-04-07 13:34:13 2026-06-10 22:35:20 | 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 | 2,905,516,807 | I_kwDOHa8MBc6tLqsH | 11,014 | https://github.com/huggingface/diffusers/issues/11014 | https://api.github.com/repos/huggingface/diffusers/issues/11014 | Getting CUDA out of memory error even with Colab A100 high RAM | ### Describe the bug
I am trying to fine-tune Flux.1-dev with Lora on the Google Colab A100 runtime environment. It has 80 GB system RAM and 40 GB VRAM. I followed the recommended steps from this [link](https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/README_flux.md). I am still getting "CUDA o... | open | null | false | 6 | [
"bug",
"stale"
] | [] | 2025-03-09T16:08:03Z | 2025-05-05T15:03:51Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | gurselnaziroglu | 3,117,731 | MDQ6VXNlcjMxMTc3MzE= | User | false |
huggingface/diffusers | 2,906,053,278 | I_kwDOHa8MBc6tNtqe | 11,016 | https://github.com/huggingface/diffusers/issues/11016 | https://api.github.com/repos/huggingface/diffusers/issues/11016 | Inconsistent results of Stable Diffusion when batch size is different | ### Describe the bug
I observed inconsistent results when running Stable Diffusion (v1.4, v2.0-base) multiple times with changes of batch size. Please see the code and log below. I am not sure whether it is a bug or a model property.
```
from typing import List
from six import iteritems
import torch
from torch impor... | closed | completed | false | 3 | [
"bug",
"stale"
] | [] | 2025-03-10T04:31:20Z | 2025-04-09T15:13:51Z | 2025-04-09T15:13:50Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | clarken92 | 1,445,226 | MDQ6VXNlcjE0NDUyMjY= | User | false |
huggingface/diffusers | 2,906,903,436 | I_kwDOHa8MBc6tQ9OM | 11,019 | https://github.com/huggingface/diffusers/issues/11019 | https://api.github.com/repos/huggingface/diffusers/issues/11019 | controlnet pipeline and training script for sana model | Controlnet training script for sana model
https://github.com/NVlabs/Sana | open | null | false | 2 | [
"stale"
] | [] | 2025-03-10T11:07:37Z | 2025-05-04T15:03:33Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | universewill | 6,790,730 | MDQ6VXNlcjY3OTA3MzA= | User | false |
huggingface/diffusers | 2,906,997,748 | I_kwDOHa8MBc6tRUP0 | 11,020 | https://github.com/huggingface/diffusers/issues/11020 | https://api.github.com/repos/huggingface/diffusers/issues/11020 | Multi-gpus Context Parallel training support? | Nowadays, the number of parameters in video generation models is increasing, and the video length is increasing. When training video models, it is difficult to fit a complete video sequence(200k~ tokens) on a single GPU. Some sequence parallel training technologies can solve this problem, such as the [fastvideo](https:... | closed | completed | false | 3 | [
"stale"
] | [
"a-r-r-o-w"
] | 2025-03-10T11:45:30Z | 2026-01-09T17:56:51Z | 2026-01-09T17:56:51Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | yinian-lw | 125,951,313 | U_kgDOB4HdUQ | User | false |
huggingface/diffusers | 2,907,938,445 | I_kwDOHa8MBc6tU56N | 11,022 | https://github.com/huggingface/diffusers/issues/11022 | https://api.github.com/repos/huggingface/diffusers/issues/11022 | Lack of Quanto support for transforming a WAN 2.1 model | ### Describe the bug
The WAN transformer doesn't work. It spits out garbage (basically all the dict items in the model) resulting in an empty transformer which can't be quantized.
`transformer = WanTransformer3DModel.from_pretrained( #<<<< FAILS
base_model,
torch_dtype=dtype,
use_safetensors=... | closed | completed | false | 2 | [
"bug"
] | [] | 2025-03-10T17:17:02Z | 2025-03-10T22:13:12Z | 2025-03-10T22:13:12Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ukaprch | 107,368,096 | U_kgDOBmZOoA | User | false |
huggingface/diffusers | 2,908,296,103 | I_kwDOHa8MBc6tWROn | 11,023 | https://github.com/huggingface/diffusers/issues/11023 | https://api.github.com/repos/huggingface/diffusers/issues/11023 | Getting blur images on playground v2.5 model when used with 'lpw_stable_diffusion_xl' custom pipeline | ### Describe the bug
I am getting blurred images on playground v2.5 model when used with 'lpw_stable_diffusion_xl'. I see Playground v2.5 uses the same architecture as SD-XL.
Please help me fix this @sayakpaul @hlky
Thank you!
### Reproduction
from diffusers import DiffusionPipeline, StableDiffusionPipeline
import... | closed | completed | false | 5 | [
"bug"
] | [] | 2025-03-10T20:03:35Z | 2025-03-31T21:32:30Z | 2025-03-31T21:32:30Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | kotlasaicharan | 156,230,258 | U_kgDOCU_icg | User | false |
huggingface/diffusers | 2,908,766,213 | I_kwDOHa8MBc6tYEAF | 11,024 | https://github.com/huggingface/diffusers/issues/11024 | https://api.github.com/repos/huggingface/diffusers/issues/11024 | SiglipImageProcessor has no .to() method (SD3IPAdapterMixin issue) | ### Describe the bug
The `SD3IPAdapterMixin` incorrectly applies `.to(self.device, dtype=self.dtype)` to `SiglipImageProcessor`, which does not have a `.to()` method.
https://github.com/huggingface/diffusers/blob/b88fef47851059ce32f161d17f00cd16d94af96a/src/diffusers/loaders/ip_adapter.py#L807-L809
This causes an `... | closed | completed | false | 2 | [
"bug"
] | [] | 2025-03-11T01:09:24Z | 2025-03-11T16:29:28Z | 2025-03-11T16:29:28Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Johnny-Katsa | 147,752,553 | U_kgDOCM6GaQ | User | false |
huggingface/diffusers | 2,911,120,195 | I_kwDOHa8MBc6thCtD | 11,032 | https://github.com/huggingface/diffusers/issues/11032 | https://api.github.com/repos/huggingface/diffusers/issues/11032 | [Quantization] Support TRT as a backend | Nice improvements here in https://github.com/NVIDIA/TensorRT-Model-Optimizer/tree/main/examples/diffusers/quantization.
Would it make sense to support this?
Cc: @SunMarc @DN6 | closed | completed | false | 14 | [
"contributions-welcome",
"quantization"
] | [] | 2025-03-11T16:08:01Z | 2025-11-07T12:53:31Z | 2025-11-07T12:53:31Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sayakpaul | 22,957,388 | MDQ6VXNlcjIyOTU3Mzg4 | User | false |
huggingface/diffusers | 2,911,560,289 | I_kwDOHa8MBc6tiuJh | 11,033 | https://github.com/huggingface/diffusers/issues/11033 | https://api.github.com/repos/huggingface/diffusers/issues/11033 | SD1.5 Unet from_single_file loading does not work | ### Describe the bug
SD1.5 Unet from_single_file loading does not work (either from safetensor or GGUF)
### Reproduction
`
import torch
from diffusers import UNet2DConditionModel
config = UNet2DConditionModel.load_config("SimianLuo/LCM_Dreamshaper_v7", subfolder="unet")
unet = UNet2DConditionModel.from_single_file(... | closed | completed | false | 11 | [
"bug"
] | [] | 2025-03-11T18:51:44Z | 2025-04-02T19:41:17Z | 2025-04-02T19:41:16Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | AbhinavGopal | 43,016,805 | MDQ6VXNlcjQzMDE2ODA1 | User | false |
huggingface/diffusers | 2,912,084,102 | I_kwDOHa8MBc6tkuCG | 11,035 | https://github.com/huggingface/diffusers/issues/11035 | https://api.github.com/repos/huggingface/diffusers/issues/11035 | Support symlink=False for from_single_file download. | **Is your feature request related to a problem? Please describe.**
When downloading from single file, take flux-dev-fp8 for example, the other pipeline components are loaded, but with a multi directory layout including symlinks (eg snapshots, blobs, etc).
This is a problem when trying to compress the downloaded comp... | closed | completed | false | 5 | [] | [] | 2025-03-11T23:08:36Z | 2025-04-17T19:27:57Z | 2025-04-17T19:27:57Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | jlonge4 | 91,354,480 | MDQ6VXNlcjkxMzU0NDgw | User | false |
huggingface/diffusers | 2,912,422,019 | I_kwDOHa8MBc6tmAiD | 11,036 | https://github.com/huggingface/diffusers/issues/11036 | https://api.github.com/repos/huggingface/diffusers/issues/11036 | Why perform the following operations on the latent condition? | in the code :https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/wan/pipeline_wan_i2v.py
line 395-404:
```
latents_mean = (
torch.tensor(self.vae.config.latents_mean)
.view(1, self.vae.config.z_dim, 1, 1, 1)
.to(latents.device, latents.dtype)
)
latents_std = 1.0 / torch.tensor(self.va... | closed | completed | false | 2 | [] | [] | 2025-03-12T02:32:09Z | 2025-03-15T02:40:13Z | 2025-03-15T02:40:12Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | trouble-maker007 | 73,164,596 | MDQ6VXNlcjczMTY0NTk2 | User | false |
huggingface/diffusers | 2,913,202,430 | I_kwDOHa8MBc6to_D- | 11,041 | https://github.com/huggingface/diffusers/issues/11041 | https://api.github.com/repos/huggingface/diffusers/issues/11041 | WAN2.1 apply_group_offloading **ERROR** result | ### Describe the bug
I am attempting to use the WAN 2.1 model from the diffusers library to complete an image-to-video task on an NVIDIA RTX 4090. To optimize memory usage, I chose the group offload method and intended to compare resource consumption across different configurations. However, during testing, I encounte... | closed | completed | false | 6 | [
"bug"
] | [
"a-r-r-o-w"
] | 2025-03-12T08:49:48Z | 2025-03-18T09:14:11Z | 2025-03-18T09:14:11Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Passenger12138 | 47,623,700 | MDQ6VXNlcjQ3NjIzNzAw | User | false |
huggingface/diffusers | 2,913,298,532 | I_kwDOHa8MBc6tpWhk | 11,042 | https://github.com/huggingface/diffusers/issues/11042 | https://api.github.com/repos/huggingface/diffusers/issues/11042 | ZeroDivisionError when performing forward pass with UNet3DConditionModel | ### Describe the bug
# ZeroDivisionError when performing forward pass with UNet3DConditionModel
I'm encountering a ZeroDivisionError when attempting to perform a forward pass with the UNet3DConditionModel. This seems to be related to the num_attention_heads parameter being None, which causes self.inner_dim to be 0.
... | closed | completed | false | 2 | [
"bug"
] | [] | 2025-03-12T09:26:01Z | 2025-03-13T02:00:12Z | 2025-03-13T02:00:11Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | txz32102 | 99,666,094 | U_kgDOBfDIrg | User | false |
huggingface/diffusers | 2,913,849,545 | I_kwDOHa8MBc6trdDJ | 11,043 | https://github.com/huggingface/diffusers/issues/11043 | https://api.github.com/repos/huggingface/diffusers/issues/11043 | When will we be getting Quanto support for Wan 2.1? | The diffusers library for quantizers currently doesn't contain an entry for Quantro:
https://github.com/huggingface/diffusers/tree/main/src/diffusers/quantizers
Isn't this needed to perform requantization on a quantized Transformer for WAN 2.1?
Currently we can't do this due to missing Quanto quantizer after we've q... | closed | completed | false | 2 | [] | [] | 2025-03-12T12:43:59Z | 2025-03-23T18:17:53Z | 2025-03-23T18:17:52Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ukaprch | 107,368,096 | U_kgDOBmZOoA | User | false |
huggingface/diffusers | 2,914,295,551 | I_kwDOHa8MBc6ttJ7_ | 11,045 | https://github.com/huggingface/diffusers/issues/11045 | https://api.github.com/repos/huggingface/diffusers/issues/11045 | Crash when loading Flux Schnell 1 model with train_dreambooth_lora_flux | ### Describe the bug
When using the `Diffusers/example/dreambooth/train_dreambooth_lora_flux` script with the Flux Schnell 1 model, the process consistently crashes during the transformer shard loading at 33% (1/3), causing my entire Google JupyterLab kernel to crash.
**Question:** Is this related to using the Flux S... | closed | completed | false | 4 | [
"bug",
"stale"
] | [] | 2025-03-12T15:08:11Z | 2025-05-07T15:18:15Z | 2025-05-07T15:18:14Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | rleygonie | 36,225,744 | MDQ6VXNlcjM2MjI1NzQ0 | User | false |
huggingface/diffusers | 2,915,063,075 | I_kwDOHa8MBc6twFUj | 11,046 | https://github.com/huggingface/diffusers/issues/11046 | https://api.github.com/repos/huggingface/diffusers/issues/11046 | flux pipeline inference with controlnet, inpainting, plus ip-adapter | ### Describe the bug
Hi, I would like to utilize flux pipeline. But for now, I have gpu issues to use origin flux pipeline.
If I would like to use nf4 version, How can I set up the inference file on controlnet, inpainting, ip-adapter?
Do I use Fluxcontrol depth or canny and mask, ip-adapter model? or fluxcontrol, flu... | open | null | false | 1 | [
"bug",
"stale"
] | [] | 2025-03-12T20:14:01Z | 2025-04-12T15:02:52Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | john09282922 | 114,594,549 | U_kgDOBtSS9Q | User | false |
huggingface/diffusers | 2,915,284,248 | I_kwDOHa8MBc6tw7UY | 11,047 | https://github.com/huggingface/diffusers/issues/11047 | https://api.github.com/repos/huggingface/diffusers/issues/11047 | No controlnet pag inpaint model in pipeline | ### Describe the bug
Hi, I would like to utilize pag version on sdxl controlnet inpaint model. But there is no pipeline. Is it just copy and past utilizing sdxl pag model?
### Reproduction
na
### Logs
```shell
na
```
###... | closed | completed | false | 3 | [
"New pipeline/model",
"contributions-welcome"
] | [] | 2025-03-12T22:16:10Z | 2025-03-14T01:02:59Z | 2025-03-14T01:02:59Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | john09282922 | 114,594,549 | U_kgDOBtSS9Q | User | false |
huggingface/diffusers | 2,915,932,081 | I_kwDOHa8MBc6tzZex | 11,048 | https://github.com/huggingface/diffusers/issues/11048 | https://api.github.com/repos/huggingface/diffusers/issues/11048 | add LTX-Video 0.9.5 diffusers support | please add LTX-Video 0.9.5 diffusers support
thanks | closed | completed | false | 1 | [
"New pipeline/model",
"wip"
] | [] | 2025-03-13T05:44:32Z | 2025-03-18T02:43:37Z | 2025-03-18T02:43:36Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | zhaoyun0071 | 35,762,050 | MDQ6VXNlcjM1NzYyMDUw | User | false |
huggingface/diffusers | 2,916,876,594 | I_kwDOHa8MBc6t3AEy | 11,049 | https://github.com/huggingface/diffusers/issues/11049 | https://api.github.com/repos/huggingface/diffusers/issues/11049 | Training Wan2.1 LoRAs | Hi :)
We are trying to train our own Wan2.1 LoRAs that are compatible with the code on diffusers, but we couldn't find a training script in the code.
Can anyone please direct us at the training script used for the LoRAs on diffusers?
Also, does anyone know the compute requirements for Wan2.1 14B vs 1.3B and how long ... | closed | completed | false | 3 | [] | [] | 2025-03-13T11:47:36Z | 2025-04-03T07:25:00Z | 2025-03-13T14:13:36Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | hila-chefer | 38,704,554 | MDQ6VXNlcjM4NzA0NTU0 | User | false |
huggingface/diffusers | 2,916,905,630 | I_kwDOHa8MBc6t3HKe | 11,050 | https://github.com/huggingface/diffusers/issues/11050 | https://api.github.com/repos/huggingface/diffusers/issues/11050 | [examples/controlnet/train_controlnet_sd3.py] prompt_embeds and pooled_prompt_embeds not cast to weight_dtype in bf16/fp16 training | ### Describe the bug
When training with --mixed_precision bf16 or fp16, the prompt_embeds and pooled_prompt_embeds tensors in the compute_text_embeddings function are not cast to the appropriate weight_dtype (matching the rest of the model inputs and parameters), causing a mismatch error during training.
Specifically... | closed | completed | false | 2 | [
"bug"
] | [] | 2025-03-13T11:58:41Z | 2025-03-14T12:03:17Z | 2025-03-14T12:03:17Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | andjoer | 60,151,338 | MDQ6VXNlcjYwMTUxMzM4 | User | false |
huggingface/diffusers | 2,918,357,550 | I_kwDOHa8MBc6t8pou | 11,055 | https://github.com/huggingface/diffusers/issues/11055 | https://api.github.com/repos/huggingface/diffusers/issues/11055 | Training on unconditional image generation creates colorized images | ### Describe the bug
Hi, I'm trying to follow the tutorial from unconditional image generation on my own dataset, and I'm getting weirdly colored images. I originally thought it was due to RGB/BGR channel order, but I've switched it around and got the same result. Do you have any suggestions of how to fix it?
### Re... | open | null | false | 1 | [
"bug",
"stale"
] | [] | 2025-03-13T20:47:22Z | 2025-04-13T15:02:53Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | esizikova-fda | 113,468,039 | U_kgDOBsNihw | User | false |
huggingface/diffusers | 2,918,852,741 | I_kwDOHa8MBc6t-iiF | 11,056 | https://github.com/huggingface/diffusers/issues/11056 | https://api.github.com/repos/huggingface/diffusers/issues/11056 | Output of randn_tensor changes based on dtype, making seeds non-reproducible if the model precision is changed | ### Describe the bug
Using `diffusers.utils.torch_utils.randn_tensor` to create noise will create different random noise depending on the tensor dtype even if a generator with a manual seed is passed, meaning seeds are not reproducible between dtypes.
For example, the [prepare_latents](https://github.com/huggingface/... | open | null | false | 5 | [
"bug",
"stale"
] | [] | 2025-03-14T02:37:41Z | 2025-05-09T15:03:37Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | city96 | 125,218,114 | U_kgDOB3atQg | User | false |
huggingface/diffusers | 1,433,474,985 | I_kwDOHa8MBc5VcRep | 1,106 | https://github.com/huggingface/diffusers/issues/1106 | https://api.github.com/repos/huggingface/diffusers/issues/1106 | PNDM repeatedly calls the unet model | ### Describe the bug
https://github.com/huggingface/diffusers/blob/0b61cea347e9b464fb03506cb78a49d38e1c74ee/src/diffusers/schedulers/scheduling_pndm.py#L168-L175
plms_timesteps uses a repeated time step at the second and the third step, which leads to the pipeline computing the unet once again:
https://github.... | closed | completed | false | 4 | [
"bug",
"stale"
] | [] | 2022-11-02T16:51:18Z | 2022-12-30T15:09:56Z | 2022-12-30T15:09:56Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | LuChengTHU | 25,171,708 | MDQ6VXNlcjI1MTcxNzA4 | User | false |
huggingface/diffusers | 2,919,841,776 | I_kwDOHa8MBc6uCT_w | 11,060 | https://github.com/huggingface/diffusers/issues/11060 | https://api.github.com/repos/huggingface/diffusers/issues/11060 | `prepare_image` in Kandinsky pipelines doesn't support `torch.Tensor` | Hi, I want to report a bug in Kandinsky pipelines.
https://github.com/huggingface/diffusers/blob/2f0f281b0d808c05bc7a974e68d298a006dd120a/src/diffusers/pipelines/kandinsky/pipeline_kandinsky_img2img.py#L413-L420
According to the above contents, elements in `image` can be either `PIL.Image.Image` or `torch.Tensor`.
h... | closed | completed | false | 1 | [
"good first issue",
"help wanted"
] | [] | 2025-03-14T10:34:30Z | 2025-04-21T18:41:10Z | 2025-04-21T18:41:10Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | dk-hong | 40,478,740 | MDQ6VXNlcjQwNDc4NzQw | User | false |
huggingface/diffusers | 2,920,853,770 | I_kwDOHa8MBc6uGLEK | 11,062 | https://github.com/huggingface/diffusers/issues/11062 | https://api.github.com/repos/huggingface/diffusers/issues/11062 | Error in loading Civit AI Lora: LCMTurboMix_Euler_A_fix | ### Describe the bug
[This CIVITAI Lora](https://civitai.com/models/216190/lora) has over 20k downloads and doesn't work with SDXL Pipeline. It is giving `lora_unet_down_blocks_0_downsamplers_0_conv.alpha` not supported error. I have uploaded the model on hugging face. Error appears on `load_lora_weights()` function
... | closed | completed | false | 1 | [
"bug",
"lora"
] | [
"sayakpaul"
] | 2025-03-14T17:22:09Z | 2025-04-14T11:40:00Z | 2025-04-14T11:40:00Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ali-afridi26 | 188,188,833 | U_kgDOCzeIoQ | User | false |
huggingface/diffusers | 2,921,068,252 | I_kwDOHa8MBc6uG_bc | 11,063 | https://github.com/huggingface/diffusers/issues/11063 | https://api.github.com/repos/huggingface/diffusers/issues/11063 | prepare_attention_mask - incorrect padding? | ### Describe the bug
I'm experimenting with attention masking in Stable Diffusion (so that padding tokens aren't considered for cross attention), and I found that UNet2DConditionModel doesn't work when given an `attention_mask`.
https://github.com/huggingface/diffusers/blob/8ead643bb786fe6bc80c9a4bd1730372d410a9df/sr... | open | null | false | 2 | [
"bug",
"stale"
] | [] | 2025-03-14T19:01:01Z | 2025-04-14T15:03:14Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | cheald | 9,830 | MDQ6VXNlcjk4MzA= | User | false |
huggingface/diffusers | 2,921,943,889 | I_kwDOHa8MBc6uKVNR | 11,068 | https://github.com/huggingface/diffusers/issues/11068 | https://api.github.com/repos/huggingface/diffusers/issues/11068 | [examples/controlnet/train_controlnet_sd3.py] training SD3.5 large controlnets from stability | It is currently not possible to fine-tune controlnets for SD3.5 large from stability ai like stabilityai/stable-diffusion-3.5-large-controlnet-blur due to the handling of positional embeddings.
I have extended and testet the script accordingly since I need it for my use case. Would this be interesting and should I do... | open | null | false | 2 | [
"stale"
] | [] | 2025-03-15T09:39:01Z | 2025-04-14T15:03:07Z | null | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | andjoer | 60,151,338 | MDQ6VXNlcjYwMTUxMzM4 | User | false |
huggingface/diffusers | 2,922,543,799 | I_kwDOHa8MBc6uMnq3 | 11,069 | https://github.com/huggingface/diffusers/issues/11069 | https://api.github.com/repos/huggingface/diffusers/issues/11069 | Remote VAE encode | ### Describe the bug
Remote VAE encode was added via PR ##11017
I've added remote-vae feature for img2img and inpaint workflows for sd15, sdxl and flux.1
- sdxl: works for both img2img and inpaint
- sd15: works for img2img, but fails for inpaint
- binary mask (typically a pil image type l) cannot be combined... | closed | completed | false | 2 | [
"bug",
"stale"
] | [] | 2025-03-15T20:58:33Z | 2025-04-15T15:26:09Z | 2025-04-15T15:26:07Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | vladmandic | 57,876,960 | MDQ6VXNlcjU3ODc2OTYw | User | false |
huggingface/diffusers | 2,923,112,149 | I_kwDOHa8MBc6uOybV | 11,071 | https://github.com/huggingface/diffusers/issues/11071 | https://api.github.com/repos/huggingface/diffusers/issues/11071 | AutoencoderKLWan - support grandient_checkpointing | Do you have a plan for supporting gradient checkpointing for AutoencoderKLWan?
Thank you for always working hard for open source :pray::pray: | open | null | false | 4 | [
"stale",
"contributions-welcome"
] | [] | 2025-03-16T14:49:43Z | 2025-04-15T15:03:04Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | agwmon | 96,934,018 | U_kgDOBccYgg | User | false |
huggingface/diffusers | 2,923,648,089 | I_kwDOHa8MBc6uQ1RZ | 11,077 | https://github.com/huggingface/diffusers/issues/11077 | https://api.github.com/repos/huggingface/diffusers/issues/11077 | About VAE convert script | ### Describe the bug
When I used `scripts/convert_vae_pt_to_diffusers.py` to convert my trained VAE to Diffusers, I found that the current script cannot handle downsampling and upsampling blocks with attention layers. Therefore, I had to manually convert it using another method.
### Reproduction
```
import argparse
... | closed | completed | false | 1 | [
"bug",
"stale"
] | [] | 2025-03-17T03:33:15Z | 2025-04-17T01:54:13Z | 2025-04-17T01:54:13Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | lavinal712 | 98,888,959 | U_kgDOBeTs_w | User | false |
huggingface/diffusers | 2,924,074,053 | I_kwDOHa8MBc6uSdRF | 11,083 | https://github.com/huggingface/diffusers/issues/11083 | https://api.github.com/repos/huggingface/diffusers/issues/11083 | Running diffusers with torch-DDP leads to warning on stride mismatch | ### Describe the bug
This issue has already been reported at https://github.com/huggingface/diffusers/issues/3809 Through an analysis of the mismatched network parameters, it was found that modifying line 369 in `diffusers/models/resnet.py` from:
```python
input_tensor = self.conv_shortcut(input_tensor)
```
to
``... | closed | completed | false | 1 | [
"bug"
] | [] | 2025-03-17T07:51:25Z | 2025-03-18T07:38:18Z | 2025-03-18T07:38:17Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | jinc7461 | 126,931,906 | U_kgDOB5DTwg | User | false |
huggingface/diffusers | 2,924,657,287 | I_kwDOHa8MBc6uUrqH | 11,086 | https://github.com/huggingface/diffusers/issues/11086 | https://api.github.com/repos/huggingface/diffusers/issues/11086 | RuntimeError after using apply_group_offloading on diffusers: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same | Can anyone help me?
I used WanX's diffusers and used apply_group_offloading according to url: https://huggingface.co/docs/diffusers/main/en/optimization/memory.
The code is as follows:
```
image_encoder = CLIPVisionModel.from_pretrained(local_model_path, subfolder="image_encoder", torch_dtype=torch.float32)
vae = Auto... | open | null | false | 5 | [
"stale"
] | [] | 2025-03-17T11:03:48Z | 2025-04-16T15:03:36Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | tiga-dudu | 139,452,550 | U_kgDOCE_ghg | User | false |
huggingface/diffusers | 2,924,915,043 | I_kwDOHa8MBc6uVqlj | 11,088 | https://github.com/huggingface/diffusers/issues/11088 | https://api.github.com/repos/huggingface/diffusers/issues/11088 | wan2.1 transformer gguf load error | ### Describe the bug
I am testing the performance of the Wan2.1 image-to-video generation on an RTX 4090 using Diffusers' Wan2.1 model【https://huggingface.co/Wan-AI/Wan2.1-I2V-14B-480P-Diffusers】 and the City96 quantized GGUF model【https://huggingface.co/city96/Wan2.1-I2V-14B-480P-gguf/tree/main】. I referred to this d... | closed | completed | false | 8 | [
"bug",
"stale"
] | [] | 2025-03-17T12:32:53Z | 2025-05-13T01:31:55Z | 2025-05-13T01:31:55Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Passenger12138 | 47,623,700 | MDQ6VXNlcjQ3NjIzNzAw | User | false |
huggingface/diffusers | 1,433,555,596 | I_kwDOHa8MBc5VclKM | 1,109 | https://github.com/huggingface/diffusers/issues/1109 | https://api.github.com/repos/huggingface/diffusers/issues/1109 | [Community] Only half of my cpu cores are being used? | ### Describe the bug
Only about half of my cpus are used.
I have 8 cores, but only 4 are used. Is there way to fix this?
![image](https://user-images.githubusercontent.com/55401336/199564675-cee7a67b-53... | open | null | false | 2 | [
"bug",
"Good second issue"
] | [] | 2022-11-02T17:53:53Z | 2026-05-22T07:10:46Z | null | NONE | null | 20260522T073512Z | 2026-05-22T07:35:12Z | BasToTheMax | 55,401,336 | MDQ6VXNlcjU1NDAxMzM2 | User | false |
huggingface/diffusers | 2,926,152,625 | I_kwDOHa8MBc6uaYux | 11,092 | https://github.com/huggingface/diffusers/issues/11092 | https://api.github.com/repos/huggingface/diffusers/issues/11092 | Empty timesteps in StableDiffusionXLImg2ImgPipeline with SDXL-Turbo using strength > 0.5 and 1 inference step | ### Describe the bug
When using StableDiffusionXLImg2ImgPipeline with the SDXL-Turbo model, setting num_inference_steps=1 and certain strength values (typically > 0.5), the pipeline crashes with a tensor error during generation. The issue occurs specifically in the get_timesteps method, where an empty tensor is create... | open | null | false | 3 | [
"bug",
"stale"
] | [] | 2025-03-17T19:28:41Z | 2026-02-03T15:23:35Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | himmelroman | 79,290,761 | MDQ6VXNlcjc5MjkwNzYx | User | false |
huggingface/diffusers | 2,926,607,734 | I_kwDOHa8MBc6ucH12 | 11,093 | https://github.com/huggingface/diffusers/issues/11093 | https://api.github.com/repos/huggingface/diffusers/issues/11093 | TorchAO example broken | ### Describe the bug
The torchao example https://huggingface.co/docs/diffusers/v0.32.2/en/quantization/torchao#torchao errors with
```Traceback (most recent call last):
File "/home/ec2-user/miniconda3/envs/comfy/lib/python3.11/site-packages/transformers/utils/import_utils.py", line 1863, in _get_module
return i... | closed | completed | false | 5 | [
"bug",
"stale"
] | [] | 2025-03-17T23:15:41Z | 2025-04-17T15:04:08Z | 2025-04-17T15:04:07Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | AbhinavGopal | 43,016,805 | MDQ6VXNlcjQzMDE2ODA1 | User | false |
huggingface/diffusers | 2,928,037,450 | I_kwDOHa8MBc6uhk5K | 11,103 | https://github.com/huggingface/diffusers/issues/11103 | https://api.github.com/repos/huggingface/diffusers/issues/11103 | Which repo should I use for LTX-Video 0.9.5 diffusers | I see the changes are merged
Checked repo and it is empty
https://huggingface.co/Lightricks/LTX-Video-0.9.5/tree/main
Noticed in test pipeline it is
repo = "YiYiXu/ltx-95"
So can I safely assume that the above can be used?
@yiyixuxu | closed | completed | false | 2 | [] | [] | 2025-03-18T10:50:41Z | 2025-03-18T11:00:34Z | 2025-03-18T11:00:32Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | nitinmukesh | 2,102,186 | MDQ6VXNlcjIxMDIxODY= | User | false |
huggingface/diffusers | 2,928,221,565 | I_kwDOHa8MBc6uiR19 | 11,104 | https://github.com/huggingface/diffusers/issues/11104 | https://api.github.com/repos/huggingface/diffusers/issues/11104 | LTX-Video-0.9.5 both offload methods are broken | ### Describe the bug
Both memory optimizations are broken after integration of 0.9.5. I can confirm it was working till 0.9.1
1. pipe.enable_sequential_cpu_offload()
2. pipe.enable_model_cpu_offload()
### Reproduction
```python
import torch
from diffusers import LTXPipeline
from diffusers.utils import export_to_vi... | closed | completed | false | 4 | [
"bug"
] | [] | 2025-03-18T11:44:16Z | 2025-03-18T19:07:02Z | 2025-03-18T19:07:01Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | nitinmukesh | 2,102,186 | MDQ6VXNlcjIxMDIxODY= | User | false |
huggingface/diffusers | 2,928,571,566 | I_kwDOHa8MBc6ujnSu | 11,108 | https://github.com/huggingface/diffusers/issues/11108 | https://api.github.com/repos/huggingface/diffusers/issues/11108 | Is there a way to generate a single image using multiple GPUs? | This is related to #2977 and #3392, but I would like to know how to generate a single image using multiple GPUs. If such a method does not exist, I would also like to know if Accelerate's [Memory-efficient pipeline parallelism](https://huggingface.co/docs/accelerate/usage_guides/distributed_inference#memory-efficient-p... | closed | completed | true | 12 | [
"stale"
] | [] | 2025-03-18T13:43:05Z | 2025-05-02T21:00:31Z | 2025-05-02T21:00:31Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | suzukimain | 131,413,573 | U_kgDOB9U2RQ | User | false |
huggingface/diffusers | 2,928,885,786 | I_kwDOHa8MBc6uk0Aa | 11,109 | https://github.com/huggingface/diffusers/issues/11109 | https://api.github.com/repos/huggingface/diffusers/issues/11109 | depth lora causing transformer layers mismatch | ### Describe the bug
When the depth lora is injected into the base dev model, peft changes the transformer's linear layer self.x_embedder from (3072, 64) to (3072, 128). These peft changes persist even after unloading the depth lora , causing subsequent requests to throw a layer mismatch error.
### Reproduction
``... | closed | completed | false | 2 | [
"bug"
] | [] | 2025-03-18T15:15:22Z | 2025-03-18T15:23:38Z | 2025-03-18T15:23:38Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | mshehzadkhan | 169,471,449 | U_kgDOChnt2Q | User | false |
huggingface/diffusers | 1,433,772,887 | I_kwDOHa8MBc5VdaNX | 1,111 | https://github.com/huggingface/diffusers/issues/1111 | https://api.github.com/repos/huggingface/diffusers/issues/1111 | A checkpoint trained only on non-copyrighted images. | **Is your feature request related to a problem? Please describe.**
I would like to be able to respect the copyright of artists whose work was used to train the model.
**Describe the solution you'd like**
I'm not sure if this is exactly the place, but if possible, I would like a checkpoint that is trained only on n... | closed | completed | false | 2 | [] | [] | 2022-11-02T20:50:25Z | 2022-11-03T21:07:39Z | 2022-11-03T21:07:39Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | MatthewWaller | 5,520,521 | MDQ6VXNlcjU1MjA1MjE= | User | false |
huggingface/diffusers | 2,930,796,247 | I_kwDOHa8MBc6usGbX | 11,114 | https://github.com/huggingface/diffusers/issues/11114 | https://api.github.com/repos/huggingface/diffusers/issues/11114 | channel inconsistency in cogvideo Lora training example | ### Describe the bug
while using the training script in (https://github.com/huggingface/diffusers/blob/main/examples/cogvideo/train_cogvideox_image_to_video_lora.py)
I made a dataset as described in readme and run training.
but a bug occurred at the forward pass process.It is because the model in-channel is 16 but m... | open | null | false | 2 | [
"bug",
"stale"
] | [] | 2025-03-19T07:55:00Z | 2025-04-18T15:02:52Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | MrTom34 | 120,256,016 | U_kgDOByr2EA | User | false |
huggingface/diffusers | 2,930,961,447 | I_kwDOHa8MBc6usuwn | 11,116 | https://github.com/huggingface/diffusers/issues/11116 | https://api.github.com/repos/huggingface/diffusers/issues/11116 | ControlNets for Multi-GPU | Currently most of our documentation has the following way of using ControlNets
```
from diffusers import SanaControlNetModel, SanaControlNetPipeline
import torch
controlnet = SanaControlNetModel.from_pretrained(
"ishan24/Sana_600M_1024px_ControlNet_diffusers",
torch_dtype=torch.float16
)
pipe = SanaControlNe... | closed | completed | false | 8 | [] | [] | 2025-03-19T08:56:12Z | 2025-04-07T22:30:23Z | 2025-04-07T22:30:22Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ishan-modi | 54,568,147 | MDQ6VXNlcjU0NTY4MTQ3 | User | false |
huggingface/diffusers | 2,931,170,274 | I_kwDOHa8MBc6uthvi | 11,117 | https://github.com/huggingface/diffusers/issues/11117 | https://api.github.com/repos/huggingface/diffusers/issues/11117 | StableDiffusion3Pipeline RuntimeError: c10::Half != c10::BFloat16 | ### Describe the bug
I tried to use Stable Diffusion 3.5 medium to generate images and copied their official script.
When I tried to run it, I got an error:
` RuntimeError: expected mat1 and mat2 to have the same dtype, but got: c10::Half != c10::BFloat16. `
When I changed `torch_dtype=torch.bfloat16` in the script... | closed | completed | false | 5 | [
"bug",
"stale"
] | [] | 2025-03-19T09:58:48Z | 2025-05-02T20:58:41Z | 2025-05-02T20:58:40Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | DataSailor | 39,119,487 | MDQ6VXNlcjM5MTE5NDg3 | User | false |
huggingface/diffusers | 2,931,975,677 | I_kwDOHa8MBc6uwmX9 | 11,118 | https://github.com/huggingface/diffusers/issues/11118 | https://api.github.com/repos/huggingface/diffusers/issues/11118 | HunyuanVideoImageToVideoPipeline failures | ### Describe the bug
pipline `HunyuanVideoImageToVideoPipeline` fails with latest combination of `diffusers` and `transformers` libraries
first, minor issue is with offloading - this snipped updates `pipeline_hunyuan_video_image2video.py` to add explicit `.to(device)` so two torch.cat operations do not fail.
```py
... | closed | completed | false | 5 | [
"bug"
] | [
"DN6"
] | 2025-03-19T14:18:33Z | 2025-04-21T07:14:09Z | 2025-04-21T07:14:07Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | vladmandic | 57,876,960 | MDQ6VXNlcjU3ODc2OTYw | User | false |
huggingface/diffusers | 1,434,017,076 | I_kwDOHa8MBc5VeV00 | 1,112 | https://github.com/huggingface/diffusers/issues/1112 | https://api.github.com/repos/huggingface/diffusers/issues/1112 | [Community] Add aspect ratio bucketing | Add the possibility of finetuning SD and dreambooth using [aspect ratio bucketing](https://github.com/NovelAI/novelai-aspect-ratio-bucketing). Instead of using crops, training with aspect ratio bucketing can greatly improve the quality of outputs. NovelAI has released this method under MIT license, so adding it should... | closed | completed | false | 5 | [
"community-examples"
] | [] | 2022-11-03T02:29:55Z | 2023-04-27T13:11:22Z | 2022-12-14T15:03:10Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Inkorak | 52,286,717 | MDQ6VXNlcjUyMjg2NzE3 | User | false |
huggingface/diffusers | 2,934,568,827 | I_kwDOHa8MBc6u6fd7 | 11,121 | https://github.com/huggingface/diffusers/issues/11121 | https://api.github.com/repos/huggingface/diffusers/issues/11121 | support SageAttention | Sage Attention provide low-bit quantization of attention. https://github.com/thu-ml/SageAttention
Seems like SageAttention supports plug-and-play way.. Will Diffusers plan to support sage attention by options?
```
import torch.nn.functional as F
+ from sageattention import sageattn
+ F.scaled_dot_product_attentio... | closed | completed | false | 2 | [
"stale"
] | [] | 2025-03-20T08:49:33Z | 2025-07-05T21:25:31Z | 2025-07-05T21:25:31Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | a120092009 | 33,205,509 | MDQ6VXNlcjMzMjA1NTA5 | User | false |
huggingface/diffusers | 2,935,315,776 | I_kwDOHa8MBc6u9V1A | 11,123 | https://github.com/huggingface/diffusers/issues/11123 | https://api.github.com/repos/huggingface/diffusers/issues/11123 | Pipelines loaded with `dtype=torch.float16` cannot run with `cpu` device. | Hi there!
I am using a image to video for lipsync through Sonic and suddenly get this error. When I do shorter videos it does not seem to cause any issues I assume this is due to length?
The thing is I want to create an animated avatar for youtube videos and they are generally 10 minutes long but when I work with 1 m... | open | null | false | 4 | [
"stale"
] | [] | 2025-03-20T12:27:22Z | 2025-04-19T15:02:42Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | r3KoR | 201,253,271 | U_kgDOC_7hlw | User | false |
huggingface/diffusers | 2,936,155,221 | I_kwDOHa8MBc6vAixV | 11,127 | https://github.com/huggingface/diffusers/issues/11127 | https://api.github.com/repos/huggingface/diffusers/issues/11127 | Civit AI flux model razor-8step-rapid-real not working with diffusers single file | ### Describe the bug
We have this civit AI model: https://civitai.com/models/849864/razor-8step-rapid-real which we want to run using `from_single_file`, but it errors out
### Reproduction
1) First create your CivitAI API key by logging into civit ai and navigating to https://civitai.com/user/account
Then go to "API... | open | null | false | 7 | [
"bug",
"stale"
] | [] | 2025-03-20T17:11:49Z | 2025-05-16T15:03:30Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ali-afridi26 | 188,188,833 | U_kgDOCzeIoQ | User | false |
huggingface/diffusers | 2,938,139,957 | I_kwDOHa8MBc6vIHU1 | 11,133 | https://github.com/huggingface/diffusers/issues/11133 | https://api.github.com/repos/huggingface/diffusers/issues/11133 | bug while using cogvideox image to video pipeline | ### Describe the bug
while using the script at "https://huggingface.co/docs/diffusers/using-diffusers/text-img2vid" of cogvideox pipeline,an error occured:
RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 16 but got size 8 for tensor number 1 in the list.
### Reproduction
import torch
f... | open | null | false | 1 | [
"bug",
"stale"
] | [] | 2025-03-21T11:41:22Z | 2025-04-20T15:02:34Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | MrTom34 | 120,256,016 | U_kgDOByr2EA | User | false |
huggingface/diffusers | 2,939,095,992 | I_kwDOHa8MBc6vLwu4 | 11,134 | https://github.com/huggingface/diffusers/issues/11134 | https://api.github.com/repos/huggingface/diffusers/issues/11134 | Implement caching on LTX and WAN video models | `CacheConfig` is used to enable **FasterCache** and **PyramidAttentionBroadcast**
CacheConfig is present in Hunyuan, Mochi, Latte, Allegro, Cog transformer modules
but its not present in WAN or LTX transformer modules:
- `WanTransformer3DModel`
- `LTXVideoTransformer3DModel`
ask is to enable caching functionality... | closed | completed | false | 3 | [
"wip"
] | [
"a-r-r-o-w"
] | 2025-03-21T17:48:50Z | 2025-07-18T13:06:47Z | 2025-07-18T13:06:47Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | vladmandic | 57,876,960 | MDQ6VXNlcjU3ODc2OTYw | User | false |
huggingface/diffusers | 2,939,675,538 | I_kwDOHa8MBc6vN-OS | 11,135 | https://github.com/huggingface/diffusers/issues/11135 | https://api.github.com/repos/huggingface/diffusers/issues/11135 | [New Pipeline]: SmoothCache: A Universal Inference Acceleration Technique for Diffusion Transformers | ### Model/Pipeline/Scheduler description
Repo: https://github.com/Roblox/SmoothCache
Paper: https://huggingface.co/papers/2411.10510
This is a training-free acceleration technique for DiT pipelines, that controls caching behavior of individual components and works across different pipelines and modalities.
There is... | open | null | false | 4 | [
"stale"
] | [] | 2025-03-21T22:45:02Z | 2025-05-10T15:03:34Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ziyu-guo | 40,365,354 | MDQ6VXNlcjQwMzY1MzU0 | User | false |
huggingface/diffusers | 2,940,269,595 | I_kwDOHa8MBc6vQPQb | 11,136 | https://github.com/huggingface/diffusers/issues/11136 | https://api.github.com/repos/huggingface/diffusers/issues/11136 | SanaSprintPipeline: Intermediate timesteps for SCM is not supported when num_inference_steps != 2 | ### Describe the bug
Documentation says 1-4 inference steps, but using 4 steps throws error
> achieving high-quality generation with 1-4 inference steps
https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/sana/pipeline_sana_sprint.py#L413
```python
if intermediate_timesteps is not None and n... | closed | completed | false | 1 | [
"bug"
] | [] | 2025-03-22T09:37:35Z | 2025-03-22T09:40:42Z | 2025-03-22T09:40:41Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | nitinmukesh | 2,102,186 | MDQ6VXNlcjIxMDIxODY= | User | false |
huggingface/diffusers | 2,940,386,314 | I_kwDOHa8MBc6vQrwK | 11,137 | https://github.com/huggingface/diffusers/issues/11137 | https://api.github.com/repos/huggingface/diffusers/issues/11137 | LattePipeline fails with dtype mismatch | ### Describe the bug
LattePipeline fails with dtype mismatch in transformer block during attention phase if `enable_temporal_attentions` is enabled (which is the default). Disabling temporal attention skips the affected block of code, so model works.
### Reproduction
load `diffusers.LattePipeline` with specific `to... | closed | completed | false | 0 | [
"bug"
] | [] | 2025-03-22T13:47:23Z | 2025-03-29T14:52:57Z | 2025-03-29T14:52:57Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | vladmandic | 57,876,960 | MDQ6VXNlcjU3ODc2OTYw | User | false |
huggingface/diffusers | 2,941,311,249 | I_kwDOHa8MBc6vUNkR | 11,143 | https://github.com/huggingface/diffusers/issues/11143 | https://api.github.com/repos/huggingface/diffusers/issues/11143 | Generated video quality is not up to the mark LTX-Video 0.9.5 | Is it that this model can be used with LTXConditionPipeline only, which mandatorily requires image/video? [The documentation says https://huggingface.co/Lightricks/LTX-Video-0.9.5. LTX Video is compatible with the [Diffusers Python library](https://huggingface.co/docs/diffusers/main/en/index). It supports both text-to-... | closed | completed | false | 4 | [] | [] | 2025-03-23T16:43:46Z | 2025-03-24T09:40:26Z | 2025-03-24T09:40:26Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | nitinmukesh | 2,102,186 | MDQ6VXNlcjIxMDIxODY= | User | false |
huggingface/diffusers | 2,941,361,125 | I_kwDOHa8MBc6vUZvl | 11,144 | https://github.com/huggingface/diffusers/issues/11144 | https://api.github.com/repos/huggingface/diffusers/issues/11144 | FlaxUNet2DConditionModel is not initialized with correct dtypes | ### Describe the bug
The FlaxUNet2DConditionModel allows specifying the dtype of the weights. Supplying a dtype different from float32 does not seem to be propagated to the actual model. This is imo different from https://github.com/huggingface/diffusers/issues/2068 since the afaik the code has correct dtype initializ... | open | null | false | 9 | [
"bug",
"stale"
] | [] | 2025-03-23T17:47:33Z | 2025-06-02T15:04:25Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | wittenator | 9,154,515 | MDQ6VXNlcjkxNTQ1MTU= | User | false |
huggingface/diffusers | 2,941,882,223 | I_kwDOHa8MBc6vWY9v | 11,145 | https://github.com/huggingface/diffusers/issues/11145 | https://api.github.com/repos/huggingface/diffusers/issues/11145 | Flux FIll lora / dreambooth finetune support | I would like to ask if there are plans to support LoRA fine-tuning for Flux Fill (inpainting)? I see that the community has already supported related code, and I would like to ask if this can be directly integrated? Thank you
https://github.com/huggingface/diffusers/blob/7a350cc8da60ee67f12a5bf5d3b3fcb4c06ffdb0/exampl... | open | null | false | 5 | [
"stale"
] | [] | 2025-03-24T03:32:35Z | 2025-07-28T06:11:05Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sanbuphy | 96,160,062 | U_kgDOBbtJPg | User | false |
huggingface/diffusers | 2,942,677,819 | I_kwDOHa8MBc6vZbM7 | 11,147 | https://github.com/huggingface/diffusers/issues/11147 | https://api.github.com/repos/huggingface/diffusers/issues/11147 | [LTX0.9.5] make LTX0.9.5 works with text-to-video | see more context here https://github.com/huggingface/diffusers/issues/11143#issuecomment-2747390564 | closed | completed | false | 9 | [
"help wanted"
] | [
"yiyixuxu"
] | 2025-03-24T09:56:47Z | 2025-04-04T14:43:16Z | 2025-04-04T14:43:16Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | yiyixuxu | 12,631,849 | MDQ6VXNlcjEyNjMxODQ5 | User | false |
huggingface/diffusers | 2,944,344,252 | I_kwDOHa8MBc6vfyC8 | 11,148 | https://github.com/huggingface/diffusers/issues/11148 | https://api.github.com/repos/huggingface/diffusers/issues/11148 | [PEFT] Loading a LoRA in PeftAdapterMixin does not set _hf_peft_config_loaded to True; causes issues with {enable,disable}_adapters | ### Describe the bug
Certain adapter functions in PeftAdapterMixin such as `enable_adapters` and `disable_adapters` do not work if the only adapters that were added were LoRAs loaded with `load_lora_adapter`, instead failing with a ValueError message indicating that no adapters are loaded:
```
>>> transformer.enable_a... | closed | completed | false | 1 | [
"bug"
] | [] | 2025-03-24T20:08:40Z | 2025-03-26T17:31:29Z | 2025-03-26T17:31:29Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | kentdan3msu | 46,754,450 | MDQ6VXNlcjQ2NzU0NDUw | User | false |
huggingface/diffusers | 2,945,309,739 | I_kwDOHa8MBc6vjdwr | 11,149 | https://github.com/huggingface/diffusers/issues/11149 | https://api.github.com/repos/huggingface/diffusers/issues/11149 | TypeError: BnB4BitDiffusersQuantizer.create_quantized_param() got an unexpected keyword argument 'dtype' | ### Describe the bug
```txt
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
Cell In[5], line 13
10 model_path = "/mnt/models/AI-ModelScope/stable-diffusion-3.5-large-turbo"
11 torch_dtype = torch.float16
-... | closed | completed | false | 4 | [
"bug"
] | [] | 2025-03-25T05:23:17Z | 2025-05-06T02:38:37Z | 2025-05-06T02:36:16Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Anopoke | 113,891,083 | U_kgDOBsnXCw | User | false |
huggingface/diffusers | 2,946,408,653 | I_kwDOHa8MBc6vnqDN | 11,151 | https://github.com/huggingface/diffusers/issues/11151 | https://api.github.com/repos/huggingface/diffusers/issues/11151 | Stable Virtual Camera | ### Model/Pipeline/Scheduler description
Hello,
does anyone have plan about implementing Diffusers style of [Stable Virtual Camera](https://github.com/Stability-AI/stable-virtual-camera)?
If so, it would be very helpful for me.
Thank you!
Best,
Daehyeon
### Open source status
- [ ] The model implementation is a... | closed | completed | false | 7 | [
"stale",
"contributions-welcome"
] | [] | 2025-03-25T12:43:19Z | 2025-06-16T09:35:47Z | 2025-05-08T05:45:24Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | choidaedae | 105,369,646 | U_kgDOBkfQLg | User | false |
huggingface/diffusers | 2,951,425,088 | I_kwDOHa8MBc6v6yxA | 11,160 | https://github.com/huggingface/diffusers/issues/11160 | https://api.github.com/repos/huggingface/diffusers/issues/11160 | Inconsistent Inference Results with Diffusers’ Implementation of WAN 2.1 14B I2V | ### Describe the bug
When using the diffusers library to run inference with WAN 2.1 14B I2V, the generated results do not align with those produced by the official inference code or DiffSynth-Studio. This discrepancy occurs even when using the same GPU, seed, precision, and prompt.
diffusers result：
https://github.co... | closed | completed | false | 7 | [
"bug"
] | [] | 2025-03-27T03:54:13Z | 2025-04-04T14:18:26Z | 2025-04-04T14:15:04Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | matabear-wyx | 128,615,473 | U_kgDOB6qEMQ | User | false |
huggingface/diffusers | 2,953,349,575 | I_kwDOHa8MBc6wCInH | 11,162 | https://github.com/huggingface/diffusers/issues/11162 | https://api.github.com/repos/huggingface/diffusers/issues/11162 | WanImageToVideoPipeline update to work with offloading | ### Describe the bug
Currently WAN I2V fails with offloaded model since `image_processor` returns image tensors on cpu while `image_encoder` gets moved to gpu by accelerate hook according to execution_device.
This simple patch fixes it by setting device explicitly
in `pipelines/wan/pipeline_wan_i2v.py`:
```py
def... | closed | completed | false | 1 | [
"bug"
] | [] | 2025-03-27T15:04:54Z | 2025-03-28T18:06:25Z | 2025-03-28T18:06:24Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | vladmandic | 57,876,960 | MDQ6VXNlcjU3ODc2OTYw | User | false |
huggingface/diffusers | 2,953,373,134 | I_kwDOHa8MBc6wCOXO | 11,163 | https://github.com/huggingface/diffusers/issues/11163 | https://api.github.com/repos/huggingface/diffusers/issues/11163 | WanImageToVideoPipeline broken math when preparing latents | ### Describe the bug
WAN 2.1 I2V models `prepare_latents` method has an issue when num_frames is not at default 81 frames.
### Reproduction
Set width=832 height=480 num_frames=15
### Logs
```shell
│ /home/vlado/dev/sdnext/venv/lib/python3.12/site-packages/diffusers/pipelines/wan/pipeline_wan_i2v.py:611 in __cal... | closed | completed | false | 1 | [
"bug"
] | [] | 2025-03-27T15:10:31Z | 2025-03-31T08:03:30Z | 2025-03-31T08:03:29Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | vladmandic | 57,876,960 | MDQ6VXNlcjU3ODc2OTYw | User | false |
huggingface/diffusers | 2,957,167,999 | I_kwDOHa8MBc6wQs1_ | 11,168 | https://github.com/huggingface/diffusers/issues/11168 | https://api.github.com/repos/huggingface/diffusers/issues/11168 | Sage Attention for diffuser library | **Is your feature request related to a problem? No
**Describe the solution you'd like.**
A clear and concise description of what you want to happen.
Incorporate a way to add sage attention to the diffusers library: Flux pipeline, Wan pipeline, etc.
**Describe alternatives you've considered.**
None
**Additional conte... | open | null | false | 12 | [
"wip"
] | [] | 2025-03-28T20:39:30Z | 2025-06-23T05:59:27Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ukaprch | 107,368,096 | U_kgDOBmZOoA | User | false |
huggingface/diffusers | 2,957,613,010 | I_kwDOHa8MBc6wSZfS | 11,169 | https://github.com/huggingface/diffusers/issues/11169 | https://api.github.com/repos/huggingface/diffusers/issues/11169 | Failed to import diffusers.schedulers.scheduling_ddim because of the following error (look up to see its traceback): numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject | ### Describe the bug
Google Colab upgraded to Python 3.11 and now this basic code fails. I don't get why
### Reproduction
```
import torch
from diffusers import StableDiffusionPipeline
# Path to your trained model
model_path = '/content/stable_diffusion_weights/ohwx/800' # Update if needed
pipe = StableDiff... | closed | completed | false | 8 | [
"bug"
] | [] | 2025-03-29T01:25:02Z | 2025-06-10T07:48:22Z | 2025-05-05T17:32:07Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | FurkanGozukara | 19,240,467 | MDQ6VXNlcjE5MjQwNDY3 | User | false |
huggingface/diffusers | 1,434,593,580 | I_kwDOHa8MBc5Vgiks | 1,117 | https://github.com/huggingface/diffusers/issues/1117 | https://api.github.com/repos/huggingface/diffusers/issues/1117 | Create tests for multiple negative prompts | As discussed in https://github.com/huggingface/diffusers/pull/1035#issuecomment-1301919727 and https://github.com/huggingface/diffusers/pull/1002#pullrequestreview-1157249188. | closed | completed | false | 1 | [
"stale"
] | [
"pcuenca"
] | 2022-11-03T12:39:26Z | 2022-12-11T15:03:02Z | 2022-12-11T15:03:02Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | pcuenca | 1,177,582 | MDQ6VXNlcjExNzc1ODI= | User | false |
huggingface/diffusers | 2,958,172,660 | I_kwDOHa8MBc6wUiH0 | 11,171 | https://github.com/huggingface/diffusers/issues/11171 | https://api.github.com/repos/huggingface/diffusers/issues/11171 | UNet1DModel does not converge | ### Describe the bug
I tried to train a UNet1DModel, DDPMScheduler Diffusion Pipeline using AdamW optimizer and mse_loss. No matter what I tried, I never got the model to produce a loss below `0.5`. As a sanity check, I also tried to replace the UNet1DModel with a UNet2DModel, which performed significantly better. Bot... | open | null | false | 1 | [
"bug",
"stale"
] | [] | 2025-03-29T16:00:12Z | 2025-11-18T17:46:05Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | MrInformatic | 15,707,532 | MDQ6VXNlcjE1NzA3NTMy | User | false |
huggingface/diffusers | 2,958,979,261 | I_kwDOHa8MBc6wXnC9 | 11,176 | https://github.com/huggingface/diffusers/issues/11176 | https://api.github.com/repos/huggingface/diffusers/issues/11176 | How to use attention_mask and encoder_attention_mask or apply prompts to specific areas in the image? | Hi, I'm aware of the attention_mask and encoder_attention_mask that exist in the forward function of the UNet2DConditionModel yet there are no examples on how to use this
I would appreciate some help on that, thank you in advance
@patrickvonplaten @Birch-san | open | null | false | 3 | [
"stale"
] | [] | 2025-03-30T16:56:40Z | 2025-04-30T15:03:34Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | alexblattner | 15,870,094 | MDQ6VXNlcjE1ODcwMDk0 | User | false |
huggingface/diffusers | 2,958,983,244 | I_kwDOHa8MBc6wXoBM | 11,177 | https://github.com/huggingface/diffusers/issues/11177 | https://api.github.com/repos/huggingface/diffusers/issues/11177 | a | null | closed | completed | false | 0 | [] | [] | 2025-03-30T17:04:56Z | 2025-04-30T02:12:34Z | 2025-04-30T02:12:34Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | felix900888 | 205,437,020 | U_kgDODD64XA | User | false |
huggingface/diffusers | 2,958,987,166 | I_kwDOHa8MBc6wXo-e | 11,178 | https://github.com/huggingface/diffusers/issues/11178 | https://api.github.com/repos/huggingface/diffusers/issues/11178 | Can't load MistoLine rank_256 ControlNet with from_single_file | ### Describe the bug
Incomprehensible error when attempting to load the MistoLine rank256 ControlNet model. There are already numerous issues documenting the same bug, and it was apparently fixed, but still occurs for me here.
### Reproduction
```
from diffusers.models.controlnet import ControlNetModel
import torch
... | closed | completed | false | 5 | [
"bug",
"stale"
] | [] | 2025-03-30T17:12:57Z | 2025-04-30T15:43:27Z | 2025-04-30T15:43:26Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | oxysoft | 3,945,277 | MDQ6VXNlcjM5NDUyNzc= | User | false |
huggingface/diffusers | 2,959,704,075 | I_kwDOHa8MBc6waYAL | 11,181 | https://github.com/huggingface/diffusers/issues/11181 | https://api.github.com/repos/huggingface/diffusers/issues/11181 | train flux controlnet with train_controlnet_flux.py script get a bad result with toy-dataset(fill50k)!! | ### Describe the bug
I using script: examples/controlnet/train_controlnet_flux.py want to train a new controlnet with my own dataset.
First I try to train toy-dataset (fill50k) as examples/controlnet/README_flux.md and the parameters as the same to README_flux.md
The only difference is train controlnet with Flux.1... | closed | completed | false | 48 | [
"bug"
] | [] | 2025-03-31T07:39:05Z | 2025-05-26T08:14:32Z | 2025-04-25T07:30:05Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Johnson-yue | 10,268,274 | MDQ6VXNlcjEwMjY4Mjc0 | User | false |
huggingface/diffusers | 2,960,946,708 | I_kwDOHa8MBc6wfHYU | 11,182 | https://github.com/huggingface/diffusers/issues/11182 | https://api.github.com/repos/huggingface/diffusers/issues/11182 | Unnecessary download when loading BF16 variant | ### Describe the bug
Loading the BF16 variant of SANA leads to downloading and loading non-BF16 files.
### Reproduction
```python
import torch
from diffusers import SanaPipeline
pipe = SanaPipeline.from_pretrained(
"Efficient-Large-Model/Sana_1600M_1024px_BF16_diffusers",
torch_dtype=torch.bfloat16,
va... | closed | completed | false | 2 | [
"bug"
] | [] | 2025-03-31T16:12:55Z | 2025-04-01T08:19:00Z | 2025-04-01T08:19:00Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | francois-rozet | 37,352,336 | MDQ6VXNlcjM3MzUyMzM2 | User | false |
huggingface/diffusers | 2,962,389,951 | I_kwDOHa8MBc6wknu_ | 11,183 | https://github.com/huggingface/diffusers/issues/11183 | https://api.github.com/repos/huggingface/diffusers/issues/11183 | Support custom_pipeline argument for Diffuser T2V Community Pipelines | **What API design would you like to have changed or added to the library? Why?**
In the diffusers community pipelines, the T2V pipelines such as [Spatiotemporal Skip Guidance (STG)](https://github.com/huggingface/diffusers/tree/main/examples/community#spatiotemporal-skip-guidance) or [CogVideoX DDIM Inversion Pipeline... | closed | not_planned | false | 2 | [
"stale"
] | [] | 2025-04-01T06:27:16Z | 2025-07-07T14:57:08Z | 2025-07-07T14:57:08Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ParagEkbote | 69,567,729 | MDQ6VXNlcjY5NTY3NzI5 | User | false |
huggingface/diffusers | 2,962,615,999 | I_kwDOHa8MBc6wle6_ | 11,185 | https://github.com/huggingface/diffusers/issues/11185 | https://api.github.com/repos/huggingface/diffusers/issues/11185 | CogView4 generate images Error | When I use diffusers to generate images with CogView4, sometimes error occured like this:
```
File "/home/igp/cogview/gradio_api_demo.py", line 189, in infer
images = pipe(
File "/home/igp/miniconda3/envs/muse/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
return func... | closed | completed | false | 0 | [] | [] | 2025-04-01T07:54:54Z | 2025-04-01T10:35:07Z | 2025-04-01T10:35:07Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | colacool | 33,508,870 | MDQ6VXNlcjMzNTA4ODcw | User | false |
huggingface/diffusers | 2,966,787,194 | I_kwDOHa8MBc6w1ZR6 | 11,199 | https://github.com/huggingface/diffusers/issues/11199 | https://api.github.com/repos/huggingface/diffusers/issues/11199 | Failed to import diffusers.pipelines.onnx_utils | ### Describe the bug
Running error:
```
diffusers\utils\import_utils.py", line 922, in _get_module
raise RuntimeError(
RuntimeError: Failed to import diffusers.pipelines.onnx_utils because of the following error (look up to see its traceback):
DLL load failed while importing onnxruntime_pybind11_state
```
It se... | open | null | false | 2 | [
"bug",
"stale"
] | [] | 2025-04-02T15:29:11Z | 2025-05-03T15:02:47Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | juntaosun | 22,288,197 | MDQ6VXNlcjIyMjg4MTk3 | User | false |
huggingface/diffusers | 2,971,942,973 | I_kwDOHa8MBc6xJEA9 | 11,207 | https://github.com/huggingface/diffusers/issues/11207 | https://api.github.com/repos/huggingface/diffusers/issues/11207 | Issue of train flux control | null | closed | completed | false | 0 | [] | [] | 2025-04-04T10:22:49Z | 2025-04-04T10:22:59Z | 2025-04-04T10:22:59Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | xduzhangjiayu | 73,270,275 | MDQ6VXNlcjczMjcwMjc1 | User | false |
huggingface/diffusers | 2,972,268,767 | I_kwDOHa8MBc6xKTjf | 11,208 | https://github.com/huggingface/diffusers/issues/11208 | https://api.github.com/repos/huggingface/diffusers/issues/11208 | MultiControlNetModel is not supported for SD3ControlNetInpaintingPipeline | ### Describe the bug
When using `StableDiffusion3ControlNetInpaintingPipeline` with `SD3MultiControlNetModel`, I receive an error:
`NotImplementedError: MultiControlNetModel is not supported for SD3ControlNetInpaintingPipeline.`
### Reproduction
Example reproduction code:
```python
import os
import torch
from dif... | closed | completed | false | 6 | [
"bug",
"help wanted",
"stale",
"Good Example PR",
"contributions-welcome"
] | [] | 2025-04-04T12:39:10Z | 2026-01-09T18:55:17Z | 2026-01-09T18:55:17Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | DanilaAniva | 128,606,792 | U_kgDOB6piSA | User | false |
huggingface/diffusers | 2,973,156,023 | I_kwDOHa8MBc6xNsK3 | 11,209 | https://github.com/huggingface/diffusers/issues/11209 | https://api.github.com/repos/huggingface/diffusers/issues/11209 | Error in loading Lora | ### Describe the bug
`lora_unet_down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k.alpha` not supported error. I have uploaded the model on hugging face. Error appears on `load_lora_weights()` function
### Reproduction
```
from diffusers import DiffusionPipeline
pipe = DiffusionPipeline.from_pretrained("sta... | open | null | false | 2 | [
"bug",
"stale",
"lora"
] | [] | 2025-04-04T19:01:02Z | 2025-05-05T15:03:09Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | D1-3105 | 65,292,437 | MDQ6VXNlcjY1MjkyNDM3 | User | false |
huggingface/diffusers | 2,973,752,742 | I_kwDOHa8MBc6xP92m | 11,213 | https://github.com/huggingface/diffusers/issues/11213 | https://api.github.com/repos/huggingface/diffusers/issues/11213 | scheduler UniPCMultistepScheduler get_velocity() | Hello, everyone,
I want to train WanTransformer3D model, but not familiar with the flow matching, the training I think I have to get this function? get_velocity() | open | null | false | 4 | [
"stale"
] | [] | 2025-04-05T02:06:26Z | 2025-05-24T15:03:24Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | JustinKai0527 | 103,019,623 | U_kgDOBiP0Zw | User | false |
huggingface/diffusers | 2,974,814,383 | I_kwDOHa8MBc6xUBCv | 11,214 | https://github.com/huggingface/diffusers/issues/11214 | https://api.github.com/repos/huggingface/diffusers/issues/11214 | TypeError: UVit2DModel.forward.<locals>.layer_() got an unexpected keyword argument 'encoder_hidden_states' | ### Describe the bug
When I run the[ amused example](https://github.com/huggingface/diffusers/tree/main/examples/amused), I got this error: TypeError: UVit2DModel.forward.<locals>.layer_() got an unexpected keyword argument 'encoder_hidden_states'.
Looks like the transformer layer does not have this encore_hidden_sta... | open | null | false | 5 | [
"bug",
"stale"
] | [] | 2025-04-06T09:04:51Z | 2025-05-06T15:03:30Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | RongKaiWeskerMA | 33,888,801 | MDQ6VXNlcjMzODg4ODAx | User | false |
huggingface/diffusers | 2,974,933,372 | I_kwDOHa8MBc6xUeF8 | 11,215 | https://github.com/huggingface/diffusers/issues/11215 | https://api.github.com/repos/huggingface/diffusers/issues/11215 | Quantization is slow with FLUX.1-dev AND no effective LoRA support | Hi everyone,
I have the following use-case. I have one base model e.g. flux-dev and I would like to dynamically load/unload different LoRAs.
I have tried different quantization methods - and to be honest it was a pain use them.
`Quanto` was generating pure noise half of the time and not reliable at all. Also when I ... | closed | completed | false | 12 | [] | [] | 2025-04-06T12:45:18Z | 2025-04-16T12:29:02Z | 2025-04-09T01:36:44Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | cryptexis | 1,158,677 | MDQ6VXNlcjExNTg2Nzc= | User | false |
huggingface/diffusers | 2,975,061,050 | I_kwDOHa8MBc6xU9Q6 | 11,216 | https://github.com/huggingface/diffusers/issues/11216 | https://api.github.com/repos/huggingface/diffusers/issues/11216 | PixArt Sigma PEFT LoRA loader support | **Is your feature request related to a problem? Please describe.**
Currently, the upstream PixArt trainer does this:
```py
transformer = get_peft_model(transformer, lora_config)
if args.mixed_precision == "fp16":
# only upcast trainable parameters (LoRA) into fp32
cast_training_params(transfo... | open | null | false | 15 | [
"stale",
"contributions-welcome"
] | [] | 2025-04-06T16:38:01Z | 2026-01-09T15:23:32Z | null | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | bghira | 59,658,056 | MDQ6VXNlcjU5NjU4MDU2 | User | false |
huggingface/diffusers | 2,975,498,831 | I_kwDOHa8MBc6xWoJP | 11,218 | https://github.com/huggingface/diffusers/issues/11218 | https://api.github.com/repos/huggingface/diffusers/issues/11218 | ImportError: cannot import name 'CogView4LoraLoaderMixin' from 'diffusers.loaders' | ### Describe the bug
Traceback (most recent call last):
File "/root/mambaforge/envs/gwb_kohya/bin/cogkit", line 5, in <module>
from cogkit.cli import cli
File "/root/mambaforge/envs/gwb_kohya/lib/python3.10/site-packages/cogkit/__init__.py", line 4, in <module>
from cogkit.python import generate_image, gen... | closed | completed | false | 4 | [
"bug",
"stale"
] | [] | 2025-04-07T03:04:57Z | 2025-05-07T15:19:14Z | 2025-05-07T15:19:13Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | godwenbin | 52,455,335 | MDQ6VXNlcjUyNDU1MzM1 | User | false |
huggingface/diffusers | 2,976,402,953 | I_kwDOHa8MBc6xaE4J | 11,220 | https://github.com/huggingface/diffusers/issues/11220 | https://api.github.com/repos/huggingface/diffusers/issues/11220 | Unconditional image generation documentation page not working as expected | ### Describe the bug
When consulting the documentation for [unconditional image generation](https://huggingface.co/docs/diffusers/using-diffusers/unconditional_image_generation), the last embedded page seems to contain an error that blocks it from being shown (see image below). This is @stevhliu's model stored in [thi... | closed | completed | false | 2 | [
"bug"
] | [] | 2025-04-07T10:32:45Z | 2025-04-08T08:47:18Z | 2025-04-07T15:46:34Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | alvaro-mazcu | 102,028,776 | U_kgDOBhTV6A | User | false |
huggingface/diffusers | 2,976,778,475 | I_kwDOHa8MBc6xbgjr | 11,221 | https://github.com/huggingface/diffusers/issues/11221 | https://api.github.com/repos/huggingface/diffusers/issues/11221 | Question about non-convergence of training autoencoderkl. | ### Describe the bug
When training the Autoencoderkl model, its loss does not converge on the ImageNet dataset. Unlike
[this](https://github.com/huggingface/diffusers/pull/10605#issuecomment-2601776571).
### Reproduction
**Script**
```
accelerate launch --multi_gpu --num_processes=2 --gpu_ids=0,1 \
train_auto... | closed | completed | false | 4 | [
"bug",
"stale"
] | [] | 2025-04-07T12:56:31Z | 2025-05-07T15:20:40Z | 2025-05-07T15:20:39Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | xiaoli1996 | 52,903,099 | MDQ6VXNlcjUyOTAzMDk5 | User | false |
huggingface/diffusers | 2,977,050,721 | I_kwDOHa8MBc6xcjBh | 11,222 | https://github.com/huggingface/diffusers/issues/11222 | https://api.github.com/repos/huggingface/diffusers/issues/11222 | Add support for more WAN Loras | There are more loras for WAN now, some really popular ones, but as right now they don't work with diffusers.
T2V Example with this [lora](https://civitai.com/models/1404755/studio-ghibli-style-wan21-t2v-14b)
```python
import torch
from diffusers import WanPipeline
from diffusers.utils import export_to_video
model_... | closed | completed | false | 0 | [] | [] | 2025-04-07T14:24:17Z | 2025-04-10T04:20:23Z | 2025-04-10T04:20:23Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | asomoza | 5,442,875 | MDQ6VXNlcjU0NDI4NzU= | User | false |
huggingface/diffusers | 2,977,126,917 | I_kwDOHa8MBc6xc1oF | 11,223 | https://github.com/huggingface/diffusers/issues/11223 | https://api.github.com/repos/huggingface/diffusers/issues/11223 | Using Layer Wise Upcasting with WAN gives OOM with loras | ### Describe the bug
When using Layer Wise Upcasting with WAN it works as intended but if it's used with a lora, it OOMs, only tested this with a 24GB GPU.
### Reproduction
Tested with this [lora](https://civitai.com/models/1375382/super-saiyan-transformation-wan21-i2v-lora):
```python
import numpy as np
import tor... | closed | completed | false | 2 | [
"bug"
] | [] | 2025-04-07T14:48:58Z | 2025-04-08T06:14:00Z | 2025-04-08T06:13:59Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | asomoza | 5,442,875 | MDQ6VXNlcjU0NDI4NzU= | User | false |
huggingface/diffusers | 2,977,665,138 | I_kwDOHa8MBc6xe5By | 11,224 | https://github.com/huggingface/diffusers/issues/11224 | https://api.github.com/repos/huggingface/diffusers/issues/11224 | Error in loading Lora | ### Describe the bug
`lora_unet_label_emb_0_0.alpha` not supported error. I have uploaded the model on hugging face. Error appears on `load_lora_weights()` function
### Reproduction
```
from diffusers import DiffusionPipeline
pipe = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0")
pipe... | open | null | false | 6 | [
"bug",
"stale",
"lora"
] | [] | 2025-04-07T18:31:08Z | 2025-05-17T15:03:10Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | D1-3105 | 65,292,437 | MDQ6VXNlcjY1MjkyNDM3 | User | false |
huggingface/diffusers | 2,977,784,460 | I_kwDOHa8MBc6xfWKM | 11,225 | https://github.com/huggingface/diffusers/issues/11225 | https://api.github.com/repos/huggingface/diffusers/issues/11225 | SD3 Controlnet Train Example, run out of memory on validation step | ### Describe the bug
On default settings provided in SD3 controlnet example, with 2 validation images training will error out with out of memory during validation on single A100 80GB.
```
04/07/2025 21:15:15 - INFO - __main__ - ***** Running training *****
04/07/2025 21:15:15 - INFO - __main__ - Num examples =... | closed | completed | false | 13 | [
"bug"
] | [] | 2025-04-07T19:20:23Z | 2025-04-09T08:03:26Z | 2025-04-09T06:42:54Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Bex0n | 64,657,877 | MDQ6VXNlcjY0NjU3ODc3 | User | false |
huggingface/diffusers | 2,979,423,022 | I_kwDOHa8MBc6xlmMu | 11,229 | https://github.com/huggingface/diffusers/issues/11229 | https://api.github.com/repos/huggingface/diffusers/issues/11229 | enable_attention_slicing give NaN results for SDXL on MPS | ### Describe the bug
If I call pipe.enable_attention_slicing I get NaN's returned when output type is 'latent' and a value error for image output.
The error is....
```py
/Volumes/SSD2TB/AI/Diffusers/lib/python3.11/site-packages/diffusers/image_processor.py:147: RuntimeWarning: invalid value encountered in cast
imag... | open | null | false | 4 | [
"bug",
"stale"
] | [] | 2025-04-08T11:00:04Z | 2025-05-28T15:03:25Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Vargol | 62,868 | MDQ6VXNlcjYyODY4 | User | false |
huggingface/diffusers | 2,979,721,541 | I_kwDOHa8MBc6xmvFF | 11,232 | https://github.com/huggingface/diffusers/issues/11232 | https://api.github.com/repos/huggingface/diffusers/issues/11232 | Noise residuals during Text-to-Image Fine-tuning | ### Describe the bug
Loss curve decreases but U-net denoising gets worse when fine-tuning stable-diffusion-2-1-base on self-built dataset.
> train_loss
> validation（Step 11,616）
![Image](https://github.com/user-attachments/asse... | closed | completed | true | 0 | [
"bug"
] | [] | 2025-04-08T12:58:37Z | 2025-04-08T13:01:24Z | 2025-04-08T13:01:24Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Yanlin2001 | 95,154,525 | U_kgDOBavxXQ | User | false |
huggingface/diffusers | 2,981,674,741 | I_kwDOHa8MBc6xuL71 | 11,241 | https://github.com/huggingface/diffusers/issues/11241 | https://api.github.com/repos/huggingface/diffusers/issues/11241 | Train Flux ControlNet Error: `Floating point exception (core dumped)` | ### Describe the bug
I got an error when i training `Flux ControlNet` with script `examples/controlnet/train_controlnet_flux.py`.
I have tried to check my datasets and there are no problems.
Here is my launch command:
```shell
accelerate launch train_flux_controlnet.py --pretrained_model_name_or_path="/root/model/flux... | closed | completed | false | 1 | [
"bug"
] | [] | 2025-04-09T06:28:14Z | 2025-04-10T03:09:37Z | 2025-04-10T03:09:37Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | paizero | 87,218,152 | MDQ6VXNlcjg3MjE4MTUy | User | false |
huggingface/diffusers | 2,982,087,944 | I_kwDOHa8MBc6xvw0I | 11,247 | https://github.com/huggingface/diffusers/issues/11247 | https://api.github.com/repos/huggingface/diffusers/issues/11247 | Error distributing FluxTransformer2DModel to multiple GPUs using controlnet | ### Describe the bug
In order to work with infiniteyou (https://huggingface.co/ByteDance/InfiniteYou) in 24GB vram gpus I'm distributing the model in several gpus.
After trying different device_map configurations and moving input tensors on different devices, I always get the same error in FluxTransformer2DModel. The ... | open | null | false | 2 | [
"bug",
"stale"
] | [] | 2025-04-09T09:00:20Z | 2025-05-09T15:03:03Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | maflx | 68,293,373 | MDQ6VXNlcjY4MjkzMzcz | User | false |
huggingface/diffusers | 1,435,021,416 | I_kwDOHa8MBc5ViLBo | 1,125 | https://github.com/huggingface/diffusers/issues/1125 | https://api.github.com/repos/huggingface/diffusers/issues/1125 | Runtime error when running Stable Diffusion Mega pipeline (inpaint) | When attempting to execute the Stable Diffusion Mega [pipeline](https://github.com/huggingface/diffusers/tree/main/examples/community#stable-diffusion-mega) snippet, the following runtime error was triggered.
This seems to be a datatype mismatch, a workaround was found by commenting out `torch_dtype` and `revision` ... | closed | completed | false | 3 | [
"bug"
] | [] | 2022-11-03T17:22:43Z | 2022-11-07T19:57:46Z | 2022-11-07T19:57:46Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | kovtcharov | 4,722,733 | MDQ6VXNlcjQ3MjI3MzM= | User | false |
huggingface/diffusers | 2,982,930,588 | I_kwDOHa8MBc6xy-ic | 11,256 | https://github.com/huggingface/diffusers/issues/11256 | https://api.github.com/repos/huggingface/diffusers/issues/11256 | Add ReNeg: An end-toend method designed to learn improved Negative embeddings (CVPR 2025 Highlight) |
### **Model/Pipeline/Scheduler description**
[ReNeg] is a reward-guided approach that directly learns Negative embeddings through gradient descent. The negative embedding learned within the same text embedding space exhibits strong generalization capabilities.
For example, using the same CLIP text encoder, the negat... | open | null | false | 2 | [
"stale"
] | [] | 2025-04-09T14:00:00Z | 2025-05-09T15:03:00Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | junpan19-new | 34,177,955 | MDQ6VXNlcjM0MTc3OTU1 | User | false |
huggingface/diffusers | 2,983,105,495 | I_kwDOHa8MBc6xzpPX | 11,258 | https://github.com/huggingface/diffusers/issues/11258 | https://api.github.com/repos/huggingface/diffusers/issues/11258 | LTX 0.95 Single file | ### Describe the bug
Using the documentation, single file loading doesn't seem to work.
### Reproduction
Combining the 0.95 weight and the from_single_file code from the documentation doesn't seem to work: https://huggingface.co/docs/diffusers/main/en/api/pipelines/ltx_video
```
import torch
from diffusers import A... | closed | completed | false | 3 | [
"bug"
] | [] | 2025-04-09T15:00:35Z | 2025-04-10T06:51:01Z | 2025-04-10T06:06:14Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | tin2tin | 1,322,593 | MDQ6VXNlcjEzMjI1OTM= | User | false |
huggingface/diffusers | 2,984,350,051 | I_kwDOHa8MBc6x4ZFj | 11,267 | https://github.com/huggingface/diffusers/issues/11267 | https://api.github.com/repos/huggingface/diffusers/issues/11267 | add `onnxruntime-qnn` to the list | https://github.com/huggingface/diffusers/blob/0706786e5393e28ecf8b669bdb9d0ee03239b019/src/diffusers/utils/import_utils.py#L102 | closed | completed | false | 1 | [] | [] | 2025-04-10T03:30:03Z | 2025-04-10T09:26:44Z | 2025-04-10T09:26:44Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | xieofxie | 2,876,650 | MDQ6VXNlcjI4NzY2NTA= | User | false |
huggingface/diffusers | 1,435,287,101 | I_kwDOHa8MBc5VjL49 | 1,127 | https://github.com/huggingface/diffusers/issues/1127 | https://api.github.com/repos/huggingface/diffusers/issues/1127 | Diffusers 0.7.0 - Torch Accelerator - "import OnnxStableDiffusionPipeline" results in Traceback Error (DmlExecutionProvider) | ### Intro
Diffusers provides a Stable Diffusion pipeline compatible with the ONNX Runtime. This allows you to run Stable Diffusion on any hardware that supports ONNX (including CPUs), and where an accelerated version of PyTorch is not available.
### Describe the bug
Calling "from diffusers import OnnxStableDiffu... | closed | completed | false | 10 | [
"bug"
] | [
"pcuenca",
"patrickvonplaten",
"anton-l"
] | 2022-11-03T21:36:22Z | 2022-11-07T22:22:42Z | 2022-11-04T15:41:22Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | averad | 640,619 | MDQ6VXNlcjY0MDYxOQ== | User | false |
huggingface/diffusers | 2,984,479,967 | I_kwDOHa8MBc6x44zf | 11,272 | https://github.com/huggingface/diffusers/issues/11272 | https://api.github.com/repos/huggingface/diffusers/issues/11272 | what is the difference between from diffusion import *** and from diffusers import ***? | I have installed diffusers and it runs ok, however the code gets wrong with " no module named diffusion "
when goes to from diffusion import ***?
What is the difference between from diffusion import *** and from diffusers import ***?
Need I install them all and what is the difference between diffusion and diffusers? | closed | completed | false | 4 | [] | [] | 2025-04-10T05:11:56Z | 2025-04-30T02:11:51Z | 2025-04-30T02:11:50Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | micklexqg | 13,776,012 | MDQ6VXNlcjEzNzc2MDEy | User | false |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.