repo stringclasses 1
value | github_id int64 1.27B 4.42B | github_node_id stringlengths 18 24 | number int64 8 13.7k | 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 2026-05-11 21:29:10 | updated_at stringdate 2022-06-12 22:18:01 2026-05-13 10:44:12 | closed_at stringdate 2022-06-12 22:18:01 2026-05-13 10:44:12 ⌀ | author_association stringclasses 3
values | milestone_title stringclasses 0
values | snapshot_id stringclasses 42
values | extracted_at stringdate 2026-04-07 13:34:13 2026-05-13 11:35:24 | author_login stringlengths 3 28 | author_id int64 1.54k 282M | author_node_id stringlengths 12 20 | author_type stringclasses 3
values | author_site_admin bool 1
class |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
huggingface/diffusers | 3,207,399,600 | I_kwDOHa8MBc6_LQiw | 11,874 | https://github.com/huggingface/diffusers/issues/11874 | https://api.github.com/repos/huggingface/diffusers/issues/11874 | [lora] `exclude_modules` won't consider the modules names in `target_modules` | In https://github.com/huggingface/diffusers/pull/11806, we introduced automatic extraction of `exclude_modules` and passing it to the following function to prepare `LoraConfig` kwargs:
https://github.com/huggingface/diffusers/blob/425a715e35479338c06b2a68eb3a95790c1db3c5/src/diffusers/utils/peft_utils.py#L153
The pro... | closed | completed | false | 6 | [] | [] | 2025-07-07T03:07:54Z | 2025-08-08T03:52:50Z | 2025-08-08T03:52:50Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sayakpaul | 22,957,388 | MDQ6VXNlcjIyOTU3Mzg4 | User | false |
huggingface/diffusers | 3,208,615,449 | I_kwDOHa8MBc6_P5YZ | 11,878 | https://github.com/huggingface/diffusers/issues/11878 | https://api.github.com/repos/huggingface/diffusers/issues/11878 | WanVACETransformer3DModel with GGUF not working for 1.3B model | ### Describe the bug
The support for GGUF in WanVace was added in this PR
https://github.com/huggingface/diffusers/pull/11807
This maybe working for 14B model (not tested) but not working for 1.3B. Didn't posted the issue earlier but now confirmed it's not only me who is facing issue.
https://github.com/huggingface/d... | closed | completed | false | 4 | [
"bug"
] | [] | 2025-07-07T11:22:41Z | 2025-07-12T11:30:38Z | 2025-07-12T11:30:38Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | nitinmukesh | 2,102,186 | MDQ6VXNlcjIxMDIxODY= | User | false |
huggingface/diffusers | 1,439,917,859 | I_kwDOHa8MBc5V02cj | 1,188 | https://github.com/huggingface/diffusers/issues/1188 | https://api.github.com/repos/huggingface/diffusers/issues/1188 | [Community] Add pipeline for CLIPSeg x Stable Diffusion | ### Model/Pipeline/Scheduler description
We've just added [CLIPSeg](https://huggingface.co/docs/transformers/main/en/model_doc/clipseg) to the 🤗 Transformers library, making it possible to use CLIPSeg in a few lines of code as shown in [this notebook](https://github.com/NielsRogge/Transformers-Tutorials/blob/master... | closed | completed | false | 14 | [
"good first issue",
"community-examples"
] | [] | 2022-11-08T10:28:49Z | 2022-12-05T07:59:09Z | 2022-11-24T09:54:18Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | NielsRogge | 48,327,001 | MDQ6VXNlcjQ4MzI3MDAx | User | false |
huggingface/diffusers | 3,211,586,926 | I_kwDOHa8MBc6_bO1u | 11,885 | https://github.com/huggingface/diffusers/issues/11885 | https://api.github.com/repos/huggingface/diffusers/issues/11885 | `load_lora_weights` bug in FLUX.1 `diffusers-0.35.0.dev0` | ### Describe the bug
I used the latest version of `diffusers-0.35.0.dev0` to fine-tune FLUX.1-Kontext with LoRA. However, when I attempted to load the saved LoRA weights, I noticed it abnormally logged that `Loading adapter weights from state_dict led to unexpected keys found in the model`.
Subsequently, this same ve... | closed | completed | false | 4 | [
"bug",
"stale"
] | [] | 2025-07-08T08:37:05Z | 2026-01-10T03:25:56Z | 2026-01-10T03:25:56Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Chenzzzzzz217 | 74,236,002 | MDQ6VXNlcjc0MjM2MDAy | User | false |
huggingface/diffusers | 3,212,809,091 | I_kwDOHa8MBc6_f5OD | 11,886 | https://github.com/huggingface/diffusers/issues/11886 | https://api.github.com/repos/huggingface/diffusers/issues/11886 | Issue with FluxKontextPipeline | ### Describe the bug
If we pass width/height in inference no changes in output image.
If we do not pass width/height, it defaults to 1024 x 1024 and crop the image.
Source image used (832 x 1216)
<img width="832" height="1216" alt="Image" src="https://github.com/user-attachments/assets/3e04a88c-3917-4214-b6d3-b8b1c8... | closed | completed | false | 4 | [
"bug"
] | [] | 2025-07-08T14:38:54Z | 2025-07-12T10:33:38Z | 2025-07-12T10:33:38Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | nitinmukesh | 2,102,186 | MDQ6VXNlcjIxMDIxODY= | User | false |
huggingface/diffusers | 3,213,452,703 | I_kwDOHa8MBc6_iWWf | 11,889 | https://github.com/huggingface/diffusers/issues/11889 | https://api.github.com/repos/huggingface/diffusers/issues/11889 | Error during batch inference using FluxControlNetInpaintingPipeline | Hi @sayakpaul I am working on batchinferencing of [flux_controlnet_inpainting_pipeline](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/flux/pipeline_flux_controlnet_inpainting.py), but I'm encountering the following error:,
Traceback (most recent call last):
File "/home/ubuntu/dev_anand/... | open | null | false | 1 | [
"stale"
] | [] | 2025-07-08T18:41:33Z | 2026-01-09T15:17:21Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | anandguptacv | 90,339,262 | MDQ6VXNlcjkwMzM5MjYy | User | false |
huggingface/diffusers | 3,215,285,698 | I_kwDOHa8MBc6_pV3C | 11,898 | https://github.com/huggingface/diffusers/issues/11898 | https://api.github.com/repos/huggingface/diffusers/issues/11898 | stabilityai/stable-diffusion-2 is missing fp16 files | null | closed | completed | false | 0 | [] | [] | 2025-07-09T09:53:18Z | 2025-07-09T09:53:30Z | 2025-07-09T09:53:30Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Parimala-15 | 133,242,784 | U_kgDOB_EfoA | User | false |
huggingface/diffusers | 3,215,305,154 | I_kwDOHa8MBc6_panC | 11,899 | https://github.com/huggingface/diffusers/issues/11899 | https://api.github.com/repos/huggingface/diffusers/issues/11899 | stabilityai/stable-diffusion-2 is missing fp16 files | null | closed | completed | false | 2 | [
"stale"
] | [] | 2025-07-09T09:58:03Z | 2026-01-09T21:49:25Z | 2026-01-09T21:49:25Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Parimala-15 | 133,242,784 | U_kgDOB_EfoA | User | false |
huggingface/diffusers | 3,215,306,480 | I_kwDOHa8MBc6_pa7w | 11,900 | https://github.com/huggingface/diffusers/issues/11900 | https://api.github.com/repos/huggingface/diffusers/issues/11900 | stabilityai/stable-diffusion-2 is missing fp16 files | null | closed | completed | false | 1 | [] | [] | 2025-07-09T09:58:17Z | 2025-07-09T14:09:07Z | 2025-07-09T14:09:07Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Parimala-15 | 133,242,784 | U_kgDOB_EfoA | User | false |
huggingface/diffusers | 3,216,045,276 | I_kwDOHa8MBc6_sPTc | 11,901 | https://github.com/huggingface/diffusers/issues/11901 | https://api.github.com/repos/huggingface/diffusers/issues/11901 | can not load Wan2.1-Fun-14B-InP-MPS_reward_lora_comfy.safetensors to wan2.1-14b-720p-diffusers | hello, i want to add Wan2.1-Fun-14B-InP-MPS_reward_lora_comfy.safetensors as a lora module into wan2.1-14b-720p-diffusers, but get errors:
```
import torch
import numpy as np
from diffusers import AutoencoderKLWan, WanImageToVideoPipeline, DiffusionPipeline
from diffusers.schedulers.scheduling_unipc_multistep import U... | open | null | false | 6 | [
"stale"
] | [] | 2025-07-09T13:53:57Z | 2026-02-03T15:20:48Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | fzuo1230 | 84,018,094 | MDQ6VXNlcjg0MDE4MDk0 | User | false |
huggingface/diffusers | 3,216,248,747 | I_kwDOHa8MBc6_tA-r | 11,902 | https://github.com/huggingface/diffusers/issues/11902 | https://api.github.com/repos/huggingface/diffusers/issues/11902 | SD3.5-Controlnet-8b does not use encoder_hidden_states | In https://github.com/huggingface/diffusers/blob/main/examples/controlnet/train_controlnet_sd3.py#L1338
the encoder_hidden_states are set to prompt_embeds, which is incorrect if we are finetuning sd3.5-controlnet-8b since it does not use encoder_hidden_states as stated in https://github.com/huggingface/diffusers/blob/... | open | null | false | 4 | [
"stale"
] | [] | 2025-07-09T15:00:42Z | 2026-02-03T15:20:45Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | viv92 | 24,729,584 | MDQ6VXNlcjI0NzI5NTg0 | User | false |
huggingface/diffusers | 3,217,008,166 | I_kwDOHa8MBc6_v6Ym | 11,903 | https://github.com/huggingface/diffusers/issues/11903 | https://api.github.com/repos/huggingface/diffusers/issues/11903 | Fused QKV projections incompatible with training | ### Describe the bug
I've enabled fused qkv projections in SimpleTuner, but it took quite a bit of investigation and effort.
1. any PEFT LoRAs become fused as well. we have to adjust the lora_target to include `to_qkv` instead of the split target layer names.
2. the `fuse_qkv_projections` method on the Attention clas... | open | null | false | 4 | [
"bug",
"stale"
] | [] | 2025-07-09T20:05:54Z | 2026-01-09T15:17:06Z | null | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | bghira | 59,658,056 | MDQ6VXNlcjU5NjU4MDU2 | User | false |
huggingface/diffusers | 3,218,275,357 | I_kwDOHa8MBc6_0vwd | 11,907 | https://github.com/huggingface/diffusers/issues/11907 | https://api.github.com/repos/huggingface/diffusers/issues/11907 | flux pipeline not work with DPMSolverMultistepScheduler and UniPCMultistepScheduler | ### Describe the bug
FluxPipeline is not compatible with two advanced schedulers: `DPMSolverMultistepScheduler` and `UniPCMultistepScheduler`
### Reproduction
`python repro.py dpm` or unipc, euler
```python
import sys
import torch
from diffusers.schedulers.scheduling_dpmsolver_multistep import \
DPMSolverMulti... | closed | completed | false | 0 | [
"bug"
] | [] | 2025-07-10T07:44:46Z | 2025-07-16T03:49:58Z | 2025-07-16T03:49:58Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | gameofdimension | 32,255,912 | MDQ6VXNlcjMyMjU1OTEy | User | false |
huggingface/diffusers | 3,225,513,446 | I_kwDOHa8MBc7AQW3m | 11,914 | https://github.com/huggingface/diffusers/issues/11914 | https://api.github.com/repos/huggingface/diffusers/issues/11914 | Loading multiple LoRAs to 1 pipeline in parallel, 1 LoRA to 2-pipelines on 2-GPUs | Hi everyone,
I have the following scenario.
I have a machine with 2-GPUs and a running service that keep has two pipelines loaded to their corresponding devices. Also I have a list of LoRAs (say 10). On each request I split the batch into 2 parts (request also has the corresponding information about LoRA), load LoRA... | closed | completed | true | 5 | [] | [] | 2025-07-12T15:54:44Z | 2025-07-15T19:40:11Z | 2025-07-15T19:40:11Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | vahe-toffee | 192,042,540 | U_kgDOC3JWLA | User | false |
huggingface/diffusers | 3,225,527,094 | I_kwDOHa8MBc7AQaM2 | 11,915 | https://github.com/huggingface/diffusers/issues/11915 | https://api.github.com/repos/huggingface/diffusers/issues/11915 | Create modular pipeline from existing pipeline | new concept of modular pipelines added via #9672 is very flexible way of creating custom pipelines
and one of the best early use-cases is new concept of modular guiders added via #11311
however, this would require a complete rewrite of the existing user apps/codebases to use new concepts
and would likely signifi... | closed | completed | false | 6 | [] | [] | 2025-07-12T16:08:30Z | 2025-08-28T08:18:08Z | 2025-08-28T08:18:08Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | vladmandic | 57,876,960 | MDQ6VXNlcjU3ODc2OTYw | User | false |
huggingface/diffusers | 3,227,270,484 | I_kwDOHa8MBc7AXD1U | 11,917 | https://github.com/huggingface/diffusers/issues/11917 | https://api.github.com/repos/huggingface/diffusers/issues/11917 | SD3ControlNet | ### Describe the bug
line 1333 of examples/train_conrolnet_sd3.py
controlnet_image = controlnet_image * vae.config.scaling_factor
missing shift factor
should be controlnet_image = (controlnet_image - vae.config.shift_factor) * vae.config.scaling_factor
### Reproduction
controlnet_image = (controlnet_image - vae.con... | closed | completed | false | 1 | [
"bug"
] | [] | 2025-07-14T03:53:16Z | 2025-07-14T16:59:25Z | 2025-07-14T16:59:25Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ZonglinL | 97,364,054 | U_kgDOBc2oVg | User | false |
huggingface/diffusers | 3,228,675,794 | I_kwDOHa8MBc7Aca7S | 11,922 | https://github.com/huggingface/diffusers/issues/11922 | https://api.github.com/repos/huggingface/diffusers/issues/11922 | FLUX.1-Fill-dev running problem | ### Describe the bug
I don't know why, but the loading has remained stationary at this step. I'm running it on a 4090-24G graphics card. But it seems that it has been stuck since loading the pipe
(worldgen) lqz27@rise:~/.../WorldGen/flux$ python flux.py
Loading checkpoint shards: 100%|█████████████████| 2/2 [00:00<00... | open | null | false | 17 | [
"bug",
"stale"
] | [] | 2025-07-14T12:52:46Z | 2026-02-03T15:20:39Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | pique2233 | 168,385,644 | U_kgDOCglcbA | User | false |
huggingface/diffusers | 3,231,501,395 | I_kwDOHa8MBc7AnMxT | 11,928 | https://github.com/huggingface/diffusers/issues/11928 | https://api.github.com/repos/huggingface/diffusers/issues/11928 | Support for Large attn_bias via Sparse Tensors or On‑The‑Fly Construction (seq_len ≈ 12 288) | I’m working with a Transformer model that routinely processes sequences up to 12 288 tokens.
For attention bias I currently create a dense attn_bias of shape 12 288 × 12 288.
Right now, I have problems with memory for my multi-head att, because of that large tensors.
I could create smaller blocks on fly from sparse at... | closed | completed | true | 1 | [] | [] | 2025-07-15T09:21:56Z | 2025-07-15T19:40:31Z | 2025-07-15T19:40:31Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | maciek-wisniewski | 152,893,708 | U_kgDOCRz5DA | User | false |
huggingface/diffusers | 3,232,891,653 | I_kwDOHa8MBc7AsgMF | 11,930 | https://github.com/huggingface/diffusers/issues/11930 | https://api.github.com/repos/huggingface/diffusers/issues/11930 | how to run convert_cosmos_to_diffusers.py correctly? | ### Describe the bug
hi. I have tried to convert the cosmos-transfer1's base model to diffuers using "convert_cosmos_to_diffusers.py" code with options --transformer_type Cosmo
s-1.0-Diffusion-7B-Video2World --vae_type CV8x8x8-1.0 --transformer_ckpt_path ../fsdp_edge_v1/iter_000016000_ema_model_only.pt --output_path .... | closed | completed | false | 2 | [
"bug",
"stale"
] | [] | 2025-07-15T16:20:09Z | 2026-01-09T21:53:35Z | 2026-01-09T21:53:35Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | dedoogong | 12,013,568 | MDQ6VXNlcjEyMDEzNTY4 | User | false |
huggingface/diffusers | 3,234,286,440 | I_kwDOHa8MBc7Ax0to | 11,938 | https://github.com/huggingface/diffusers/issues/11938 | https://api.github.com/repos/huggingface/diffusers/issues/11938 | RMSNorm's weight not registered as submodules when initializing | ### Describe the bug
RMSNorm() look like this, which is in models/normalization.py
```python
class RMSNorm(nn.Module):
def __init__(self, dim, eps: float, elementwise_affine: bool = True, bias: bool = False):
super().__init__()
self.eps = eps
self.elementwise_affine = elementwise_affine
... | open | null | false | 3 | [
"bug",
"stale"
] | [] | 2025-07-16T03:06:33Z | 2026-01-17T15:05:23Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Darkbblue | 65,117,884 | MDQ6VXNlcjY1MTE3ODg0 | User | false |
huggingface/diffusers | 3,234,290,538 | I_kwDOHa8MBc7Ax1tq | 11,939 | https://github.com/huggingface/diffusers/issues/11939 | https://api.github.com/repos/huggingface/diffusers/issues/11939 | train flux_controlnet code need to update | This code is just for instantx controlnet, not suitable xlabs
https://github.com/huggingface/diffusers/blob/aa14f090f86c9641abf2da761f39a88133b49f09/examples/controlnet/train_controlnet_flux.py#L1258-L1266
reference FluxControlNetPipeline code:
https://github.com/huggingface/diffusers/blob/aa14f090f86c9641abf2da761f... | open | null | false | 1 | [
"stale"
] | [] | 2025-07-16T03:08:23Z | 2026-01-09T15:16:52Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Johnson-yue | 10,268,274 | MDQ6VXNlcjEwMjY4Mjc0 | User | false |
huggingface/diffusers | 3,236,757,212 | I_kwDOHa8MBc7A7P7c | 11,942 | https://github.com/huggingface/diffusers/issues/11942 | https://api.github.com/repos/huggingface/diffusers/issues/11942 | Kontext pipeline resizing image even if it is within PREFERRED_KONTEXT_RESOLUTIONS | ### Describe the bug
My assumption is, if resolution of image matches one of PREFERRED_KONTEXT_RESOLUTIONS, pipeline should not resize the input image.
Even though I use the image with one of the preferred resolution, the pipeline is resizing the image.
PREFERRED_KONTEXT_RESOLUTIONS = [
**(672, 1568),**
(688... | closed | completed | false | 4 | [
"bug"
] | [] | 2025-07-16T17:26:11Z | 2025-08-02T13:38:46Z | 2025-08-02T13:38:46Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | nitinmukesh | 2,102,186 | MDQ6VXNlcjIxMDIxODY= | User | false |
huggingface/diffusers | 3,237,556,499 | I_kwDOHa8MBc7A-TET | 11,943 | https://github.com/huggingface/diffusers/issues/11943 | https://api.github.com/repos/huggingface/diffusers/issues/11943 | ValueError when loading Lycoris safetensors LoRA: keys not correctly renamed | ### Describe the bug
Hello,
I am trying to load a Lycoris LoRA safetensors file (e.g. `de-anime-er_v10.safetensors`) into the Diffusers pipeline using the following code:
safetensors download url : https://civitai.com/api/download/models/111997?type=Model&format=SafeTensor
lora_path = "/content/de-anime-er_v10.safe... | closed | completed | false | 3 | [
"bug",
"stale"
] | [] | 2025-07-16T22:52:54Z | 2026-01-09T21:55:12Z | 2026-01-09T21:55:12Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sss-okayu-korone | 115,227,116 | U_kgDOBt457A | User | false |
huggingface/diffusers | 3,237,943,840 | I_kwDOHa8MBc7A_xog | 11,945 | https://github.com/huggingface/diffusers/issues/11945 | https://api.github.com/repos/huggingface/diffusers/issues/11945 | Floating point exception with nightly PyTorch and CUDA | ### Describe the bug
When running any code snippet using diffusers it fails with floating point exception, and doesn't print any traceback.
For example this one would cause the issue (the example of Stable Diffusion 3.5 medium):
```
import torch
from diffusers import StableDiffusion3Pipeline
pipe = StableDiffusion3... | closed | completed | false | 3 | [
"bug",
"stale"
] | [] | 2025-07-17T03:16:02Z | 2026-01-09T21:56:04Z | 2026-01-09T21:56:04Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | MxtAppz | 121,626,118 | U_kgDOBz_eBg | User | false |
huggingface/diffusers | 3,238,442,724 | I_kwDOHa8MBc7BBrbk | 11,946 | https://github.com/huggingface/diffusers/issues/11946 | https://api.github.com/repos/huggingface/diffusers/issues/11946 | closed | null | closed | completed | false | 0 | [
"bug"
] | [] | 2025-07-17T07:21:24Z | 2025-07-23T07:14:16Z | 2025-07-23T07:14:16Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | SS-snap | 143,506,979 | U_kgDOCI2-Iw | User | false |
huggingface/diffusers | 3,239,350,249 | I_kwDOHa8MBc7BFI_p | 11,948 | https://github.com/huggingface/diffusers/issues/11948 | https://api.github.com/repos/huggingface/diffusers/issues/11948 | Impossible to load WanTransformer3DModel when offline using the 'from_pretrained' function. | ### Describe the bug
Hi!
I am using the Wan2.1-1.3B-Diffusers transformer for a project. When I load the transformer (but not the complete pipeline), I receive a ConnectionError, even when I pass the local_files_only=True input. However, when I load the complete pipeline, I do not receive an error. Please note that th... | closed | completed | false | 16 | [
"bug",
"stale"
] | [] | 2025-07-17T12:11:56Z | 2026-01-09T21:56:52Z | 2026-01-09T21:56:52Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | guillaumejs2403 | 25,442,525 | MDQ6VXNlcjI1NDQyNTI1 | User | false |
huggingface/diffusers | 1,440,364,833 | I_kwDOHa8MBc5V2jkh | 1,195 | https://github.com/huggingface/diffusers/issues/1195 | https://api.github.com/repos/huggingface/diffusers/issues/1195 | Memory efficient attention not working with fp16 weights | ### Describe the bug
Following the example code with available in the `0.7.0` release
```py
from diffusers import StableDiffusionPipeline
import torch
pipe = StableDiffusionPipeline.from_pretrained(
"runwayml/stable-diffusion-v1-5",
revision="fp16",
torch_dtype=torch.float16,
).to("cuda")
p... | closed | completed | false | 2 | [
"bug"
] | [
"patil-suraj"
] | 2022-11-08T15:01:10Z | 2022-11-08T16:15:25Z | 2022-11-08T16:15:25Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | apolinario | 788,417 | MDQ6VXNlcjc4ODQxNw== | User | false |
huggingface/diffusers | 3,241,705,468 | I_kwDOHa8MBc7BOH_8 | 11,951 | https://github.com/huggingface/diffusers/issues/11951 | https://api.github.com/repos/huggingface/diffusers/issues/11951 | Kontext model loading quantization problem | Hello, can kontext be loaded quantitatively at present? Because I only have a 4090 with 24g video memory, the current fp16 loading method will cause OOM. Like flux, can it be loaded with torchao or gguf, so that this model can run on 4090? | closed | completed | false | 2 | [] | [] | 2025-07-18T03:20:48Z | 2025-07-18T05:39:28Z | 2025-07-18T05:39:27Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | babyta | 37,787,795 | MDQ6VXNlcjM3Nzg3Nzk1 | User | false |
huggingface/diffusers | 3,244,149,950 | I_kwDOHa8MBc7BXcy- | 11,956 | https://github.com/huggingface/diffusers/issues/11956 | https://api.github.com/repos/huggingface/diffusers/issues/11956 | Frequency-Decoupled Guidance (FDG) for diffusion models | FDG is a new method for applying CFG in the frequency domain. It improves generation quality at low CFG scales while inherently avoiding the harmful effects of high CFG values. It could be a nice addition to the guiders part of diffusers. The implementation details for FDG are available on page 19 of the paper.
https:... | closed | completed | false | 5 | [
"help wanted",
"Good second issue",
"contributions-welcome",
"advanced",
"consider-for-modular-diffusers"
] | [] | 2025-07-18T19:12:50Z | 2025-08-07T05:51:03Z | 2025-08-07T05:51:03Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Msadat97 | 71,992,433 | MDQ6VXNlcjcxOTkyNDMz | User | false |
huggingface/diffusers | 3,244,453,789 | I_kwDOHa8MBc7BYm-d | 11,957 | https://github.com/huggingface/diffusers/issues/11957 | https://api.github.com/repos/huggingface/diffusers/issues/11957 | Flash/Sage varlen does not work with torch.compile | @a-r-r-o-w @DN6 @tolgacangoz
I got a bit excited about this PR and wanted to give it a go. I love the syntax, both the setter function and the context, great work!
I wanted to also see if it would still compile but got the following error logs:
```
[t+28s648ms] 0%| | 0/30 [00:00<?, ?it/s]/inferencesh/... | open | null | false | 4 | [
"stale"
] | [
"a-r-r-o-w"
] | 2025-07-18T21:57:42Z | 2026-01-09T15:16:34Z | null | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | a-r-r-o-w | 72,266,394 | MDQ6VXNlcjcyMjY2Mzk0 | User | false |
huggingface/diffusers | 3,245,390,831 | I_kwDOHa8MBc7BcLvv | 11,959 | https://github.com/huggingface/diffusers/issues/11959 | https://api.github.com/repos/huggingface/diffusers/issues/11959 | Passing enable_gqa flag in attention dispatcher should be version guarded | Just realized this is potentially breaking: https://github.com/huggingface/diffusers/blob/cde02b061b6f13012dfefe76bc8abf5e6ec6d3f3/src/diffusers/models/attention_dispatch.py#L229
I believe enable_gqa was first released in PT 2.5.0. For versions before that, this flag should not be passed. Additionally, some user code ... | closed | completed | false | 0 | [
"bug"
] | [
"a-r-r-o-w"
] | 2025-07-19T15:36:55Z | 2025-07-22T15:17:46Z | 2025-07-22T15:17:46Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | a-r-r-o-w | 72,266,394 | MDQ6VXNlcjcyMjY2Mzk0 | User | false |
huggingface/diffusers | 3,245,889,780 | I_kwDOHa8MBc7BeFj0 | 11,961 | https://github.com/huggingface/diffusers/issues/11961 | https://api.github.com/repos/huggingface/diffusers/issues/11961 | New Adapter/Pipeline Request: IT-Blender for Creative Conceptual Blending | ## Model/Pipeline/Scheduler description
### Name of the model/pipeline/scheduler
"Image-and-Text Concept Blender" (IT-Blender), a diffusion adapter that blends visual concepts from a real reference image with textual concepts from a prompt in a disentangled manner. The goal is to enhance human creativity in design tas... | open | null | false | 1 | [
"stale"
] | [] | 2025-07-20T03:07:38Z | 2026-01-09T15:16:31Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | WonwoongCho | 20,943,085 | MDQ6VXNlcjIwOTQzMDg1 | User | false |
huggingface/diffusers | 3,246,119,771 | I_kwDOHa8MBc7Be9tb | 11,962 | https://github.com/huggingface/diffusers/issues/11962 | https://api.github.com/repos/huggingface/diffusers/issues/11962 | FLUX.1-Kontext-dev Support for GGUF Quantized Model | ### Model/Pipeline/Scheduler description
The original model weights and pipeline are available at:
https://huggingface.co/black-forest-labs/FLUX.1-Kontext-dev/tree/main
Quantized (GGUF) versions of the model can be found here:
https://huggingface.co/QuantStack/FLUX.1-Kontext-dev-GGUF/tree/main
Using the code below, I... | closed | completed | false | 1 | [] | [] | 2025-07-20T07:46:49Z | 2025-08-02T13:45:04Z | 2025-08-02T13:45:04Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sahandkh1419 | 52,316,446 | MDQ6VXNlcjUyMzE2NDQ2 | User | false |
huggingface/diffusers | 3,247,231,562 | I_kwDOHa8MBc7BjNJK | 11,964 | https://github.com/huggingface/diffusers/issues/11964 | https://api.github.com/repos/huggingface/diffusers/issues/11964 | KeyError when loading LoRA for Flux model: missing lora_unet_final_layer_adaLN_modulation_1 weights | I'm trying to run Overlay-Kontext-Dev-LoRA locally by loading the LoRA weights using the pipe.load_lora_weights() function. However, I encountered the following error during execution:
> KeyError: 'lora_unet_final_layer_adaLN_modulation_1.lora_down.weight'
```
import torch
from diffusers import DiffusionPipeline
fro... | open | null | false | 4 | [
"stale"
] | [] | 2025-07-21T05:16:34Z | 2026-02-03T15:20:25Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | NEWbie0709 | 81,673,708 | MDQ6VXNlcjgxNjczNzA4 | User | false |
huggingface/diffusers | 3,247,234,057 | I_kwDOHa8MBc7BjNwJ | 11,965 | https://github.com/huggingface/diffusers/issues/11965 | https://api.github.com/repos/huggingface/diffusers/issues/11965 | Highdream multiple gpu to offload llama | An option to offload clip and llama to a secondary gpu would be great allowing primary GPU to focus on Lora training would be great! | closed | completed | false | 2 | [
"stale"
] | [] | 2025-07-21T05:18:01Z | 2026-01-09T22:02:19Z | 2026-01-09T22:02:19Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | grio43 | 1,619,512 | MDQ6VXNlcjE2MTk1MTI= | User | false |
huggingface/diffusers | 3,247,718,019 | I_kwDOHa8MBc7BlD6D | 11,966 | https://github.com/huggingface/diffusers/issues/11966 | https://api.github.com/repos/huggingface/diffusers/issues/11966 | How about forcing the first and last block on device when groupoffloading is used? | **Is your feature request related to a problem? Please describe.**
When group offloading is enabled, the offload and onload cannot be streamed between steps and this is really a big time comsuming problem.
**Describe the solution you'd like.**
Is it possible to add an option that could make the first and last block fo... | open | null | false | 16 | [
"stale",
"contributions-welcome",
"group-offloading"
] | [
"a-r-r-o-w"
] | 2025-07-21T08:38:30Z | 2026-02-03T15:20:22Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | seed93 | 4,570,032 | MDQ6VXNlcjQ1NzAwMzI= | User | false |
huggingface/diffusers | 1,440,629,396 | I_kwDOHa8MBc5V3kKU | 1,197 | https://github.com/huggingface/diffusers/issues/1197 | https://api.github.com/repos/huggingface/diffusers/issues/1197 | [Community] OpenAI Diffusion Pipeline | **Is your feature request related to a problem? Please describe.**
It would be cool to be able to use the many OpenAI Guided Diffusion models within diffusers itself.
**Describe the solution you'd like**
An OpenAIGuidedDiffusionPipeline class to load and handle OpenAI Guided Diffusion Models
**Describe alterna... | open | null | false | 8 | [
"New pipeline/model"
] | [] | 2022-11-08T17:36:41Z | 2025-10-06T14:57:45Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | WASasquatch | 1,151,589 | MDQ6VXNlcjExNTE1ODk= | User | false |
huggingface/diffusers | 3,251,540,760 | I_kwDOHa8MBc7BzpMY | 11,971 | https://github.com/huggingface/diffusers/issues/11971 | https://api.github.com/repos/huggingface/diffusers/issues/11971 | What is the minimum memory requirement for model training? | Hello, I would like to try training an SDXL model using my own dataset. What is the minimum memory size required for the model? | closed | completed | true | 1 | [] | [] | 2025-07-22T07:52:28Z | 2025-07-22T08:26:27Z | 2025-07-22T08:26:27Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | WWWPPPGGG | 124,946,162 | U_kgDOB3KG8g | User | false |
huggingface/diffusers | 3,251,695,764 | I_kwDOHa8MBc7B0PCU | 11,973 | https://github.com/huggingface/diffusers/issues/11973 | https://api.github.com/repos/huggingface/diffusers/issues/11973 | train_dreambooth_lora_flux_kontext.py: error: the following arguments are required: --instance_prompt | ### Describe the bug
train_dreambooth_lora_flux_kontext.py: error: the following arguments are required: --instance_prompt
### Reproduction
accelerate launch --main_process_port 29513 train_dreambooth_lora_flux_kontext.py \
--pretrained_model_name_or_path=black-forest-labs/FLUX.1-Kontext-dev \
--output_dir="kon... | open | null | false | 4 | [
"bug",
"stale"
] | [] | 2025-07-22T08:36:20Z | 2026-02-03T15:20:17Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | huassaue | 107,062,689 | U_kgDOBmGloQ | User | false |
huggingface/diffusers | 3,256,073,911 | I_kwDOHa8MBc7CE763 | 11,977 | https://github.com/huggingface/diffusers/issues/11977 | https://api.github.com/repos/huggingface/diffusers/issues/11977 | how to load a finetuned model especially during validation phase | <img width="1034" height="743" alt="Image" src="https://github.com/user-attachments/assets/c4e9318f-10aa-4b91-9d60-e28a3be38f8a" />
As the above, I have finetuned the model and want to validate it, but the given demo which is train_dreambooth_sd3.py still uses
"pipeline = StableDiffusion3Pipeline.from_pretrained(
... | closed | completed | true | 6 | [
"stale"
] | [] | 2025-07-23T11:54:16Z | 2026-01-09T22:03:59Z | 2026-01-09T22:03:59Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | micklexqg | 13,776,012 | MDQ6VXNlcjEzNzc2MDEy | User | false |
huggingface/diffusers | 3,256,249,258 | I_kwDOHa8MBc7CFmuq | 11,978 | https://github.com/huggingface/diffusers/issues/11978 | https://api.github.com/repos/huggingface/diffusers/issues/11978 | Add Bria 3.2 pipeline - next-generation commercial-ready text-to-image model | ### Model/Pipeline/Scheduler description
**TL;DR**
Bria 3.2 is the next-generation commercial-ready text-to-image model. With just 4 billion parameters, it provides exceptional aesthetics and text rendering, evaluated to provide on par results to leading open-source models, and outperforming other licensed models.
I... | closed | completed | false | 3 | [] | [] | 2025-07-23T12:43:54Z | 2025-08-26T05:30:48Z | 2025-08-26T05:30:48Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | galbria | 158,810,732 | U_kgDOCXdCbA | User | false |
huggingface/diffusers | 3,256,422,691 | I_kwDOHa8MBc7CGREj | 11,981 | https://github.com/huggingface/diffusers/issues/11981 | https://api.github.com/repos/huggingface/diffusers/issues/11981 | Groupoffloading introduce bad results | ### Describe the bug
I am using groupoffloading for saving gpu memory. I got worse results with a cosine similarity aboud 0.934 on A800, which is unexpected. And I got results with a cosine similarity about 0.78 on 4090, which is worse.
Could anyone give me any suggestions to fix the precision?
### Reproduction
```... | closed | completed | false | 10 | [
"bug"
] | [] | 2025-07-23T13:31:58Z | 2025-11-18T08:40:51Z | 2025-08-06T15:41:01Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | seed93 | 4,570,032 | MDQ6VXNlcjQ1NzAwMzI= | User | false |
huggingface/diffusers | 3,259,150,182 | I_kwDOHa8MBc7CQq9m | 11,984 | https://github.com/huggingface/diffusers/issues/11984 | https://api.github.com/repos/huggingface/diffusers/issues/11984 | A compatibility issue when using custom Stable Diffusion with pre-trained ControlNets | I have successfully fine-tuned a Stable Diffusion v1.5 model using the Dreambooth script, and the results are excellent. However, I've encountered a compatibility issue when using this custom model with pre-trained ControlNets. Since the Dreambooth process modifies the U-Net weights, the original ControlNet is no longe... | closed | completed | false | 6 | [] | [] | 2025-07-24T09:16:55Z | 2025-07-24T15:15:20Z | 2025-07-24T15:15:20Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ScienceLi1125 | 99,795,063 | U_kgDOBfLAdw | User | false |
huggingface/diffusers | 3,261,485,275 | I_kwDOHa8MBc7CZlDb | 11,989 | https://github.com/huggingface/diffusers/issues/11989 | https://api.github.com/repos/huggingface/diffusers/issues/11989 | The inference image is pure black. This seems to be related to weight, but I have been unable to pinpoint where the problem lies. | ### Describe the bug
I occasionally encounter situations where, when I restart my inference script again, the resulting image turns completely black. I haven't made any changes.
My intuition tells me that this is related to weight. I am currently a combination of SD and ControlNet, which I have trained on my own dat... | closed | completed | false | 8 | [
"bug"
] | [] | 2025-07-24T23:08:13Z | 2025-07-29T06:42:33Z | 2025-07-29T06:42:33Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ShanZard | 87,146,029 | MDQ6VXNlcjg3MTQ2MDI5 | User | false |
huggingface/diffusers | 1,440,739,767 | I_kwDOHa8MBc5V3_G3 | 1,199 | https://github.com/huggingface/diffusers/issues/1199 | https://api.github.com/repos/huggingface/diffusers/issues/1199 | Download stable_diffusion on aws | ### Describe the bug
Hi,
I want to run stable diffusion model on aws. and I am downloading model on aws with following code:
`git lfs install`
`git clone https://huggingface.co/runwayml/stable-diffusion-v1-5`
However during downloading 'stable-diffusion-v1-5' the downloading interrupts and gives following erro... | closed | completed | false | 6 | [
"bug"
] | [] | 2022-11-08T18:59:49Z | 2022-11-11T11:12:10Z | 2022-11-11T11:12:09Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | hamzafar | 6,842,017 | MDQ6VXNlcjY4NDIwMTc= | User | false |
huggingface/diffusers | 3,262,772,203 | I_kwDOHa8MBc7CefPr | 11,992 | https://github.com/huggingface/diffusers/issues/11992 | https://api.github.com/repos/huggingface/diffusers/issues/11992 | Problems with Kontext quantization | I have some problems with Kontext quantization model.
1. FP8 quantization
This is my code:
```
import torch
from diffusers import FluxKontextPipeline, FluxTransformer2DModel, AutoencoderKL, FlowMatchEulerDiscreteScheduler, GGUFQuantizationConfig
from transformers import CLIPTextModel, CLIPTokenizer, T5EncoderModel, T... | open | null | false | 12 | [
"stale"
] | [] | 2025-07-25T10:35:45Z | 2026-02-03T15:20:03Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | MinhVD-ZenAI | 191,195,078 | U_kgDOC2Vnxg | User | false |
huggingface/diffusers | 3,267,151,659 | I_kwDOHa8MBc7CvMcr | 11,996 | https://github.com/huggingface/diffusers/issues/11996 | https://api.github.com/repos/huggingface/diffusers/issues/11996 | Using AutoModel.from_pretrained for wan2.1 crash my notebook | Hi, i been trying to use the diffusers library to use the Wan2.1 model, and i was following the official documentation to use it with the diffusers library from here: https://huggingface.co/docs/diffusers/main/en/api/pipelines/wan
But everytime I try to load the vae my jupyter notebook, It keep crashing when the model... | closed | completed | false | 2 | [
"stale"
] | [] | 2025-07-27T14:22:20Z | 2026-01-09T22:06:51Z | 2026-01-09T22:06:51Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Javierxx05 | 42,848,068 | MDQ6VXNlcjQyODQ4MDY4 | User | false |
huggingface/diffusers | 3,267,939,180 | I_kwDOHa8MBc7CyMts | 11,998 | https://github.com/huggingface/diffusers/issues/11998 | https://api.github.com/repos/huggingface/diffusers/issues/11998 | Support training variance learning (Improved DDPM) with VLB loss. | Issues has been previously discussed here, but was closed unresolved. https://github.com/huggingface/diffusers/issues/3287. Diffusers only support variance learning in inference, which is weird if they did not support the training in the first place. Improved DDPM has been since used in lot of other repositories like A... | closed | not_planned | false | 1 | [] | [] | 2025-07-28T03:40:22Z | 2025-08-02T13:43:55Z | 2025-08-02T13:43:55Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | bhosalems | 10,846,405 | MDQ6VXNlcjEwODQ2NDA1 | User | false |
huggingface/diffusers | 1,272,010,652 | I_kwDOHa8MBc5L0Vec | 12 | https://github.com/huggingface/diffusers/issues/12 | https://api.github.com/repos/huggingface/diffusers/issues/12 | Collection of nitpicks | See below for a small collection of nitpicks; I suppose those will be adressed before the first release, but wanted to write them down somewhere:
- Documentation seems to be in a different format so will need to be refactored to work with the `doc-builder`
- The `logging` class still has mentions to `🤗 Transformers`... | closed | completed | false | 5 | [] | [] | 2022-06-15T10:21:24Z | 2022-07-21T18:54:29Z | 2022-07-21T18:54:29Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | LysandreJik | 30,755,778 | MDQ6VXNlcjMwNzU1Nzc4 | User | false |
huggingface/diffusers | 3,269,498,493 | I_kwDOHa8MBc7C4JZ9 | 12,003 | https://github.com/huggingface/diffusers/issues/12003 | https://api.github.com/repos/huggingface/diffusers/issues/12003 | Cannot copy out of meta tensor; no data! Please use torch.nn.Module.to_empty() instead of torch.nn.Module.to() when moving module from meta to a different device. | I tried to load `Flux_Turbo_Alpha` into FluxKontextPipeline by this code, and save_pretrained to local.
```
import torch
from diffusers import FluxKontextPipeline
from diffusers.utils import load_image
pipe = FluxKontextPipeline.from_pretrained(
"black-forest-labs/FLUX.1-Kontext-dev", torch_dtype=torch.bfloat16
)... | open | null | false | 9 | [
"stale"
] | [
"sayakpaul"
] | 2025-07-28T11:27:24Z | 2026-03-14T05:20:21Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | MinhVD-ZenAI | 191,195,078 | U_kgDOC2Vnxg | User | false |
huggingface/diffusers | 3,271,947,098 | I_kwDOHa8MBc7DBfNa | 12,007 | https://github.com/huggingface/diffusers/issues/12007 | https://api.github.com/repos/huggingface/diffusers/issues/12007 | The fuse_lora() func in class PeftAdapterMixin did not take effect. | ### Describe the bug
I use the fuse_lora function， but the weights before and after the fuse_lora stay the same in FluxTransformer2DModel.
Codes in TensorRT demo:
```shell
def merge_loras(model, lora_loader):
import copy
model_transformer_blocks_bak = copy.deepcopy(model.transformer_blocks)
paths, weight... | closed | completed | false | 0 | [
"bug"
] | [] | 2025-07-29T04:07:00Z | 2025-07-29T13:21:17Z | 2025-07-29T13:21:17Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | dujifeng | 42,289,967 | MDQ6VXNlcjQyMjg5OTY3 | User | false |
huggingface/diffusers | 3,272,524,463 | I_kwDOHa8MBc7DDsKv | 12,009 | https://github.com/huggingface/diffusers/issues/12009 | https://api.github.com/repos/huggingface/diffusers/issues/12009 | WanTransformer3DModel.from_single_file wont load Wan2.2 GGUF (NotImplementedError: Cannot copy out of meta tensor; no data) | ### Describe the bug
I am trying to load Wan2.2 gguf transformers but unfortunately they yield a criptic error when I try to use them
```python
repo_id = "QuantStack/Wan2.2-I2V-A14B-GGUF"
filename = "HighNoise/Wan2.2-I2V-A14B-HighNoise-Q2_K.gguf"
gguf_path = hf_hub_download(repo_id=repo_id, filename=filename)
trans... | closed | completed | false | 3 | [
"bug"
] | [] | 2025-07-29T07:42:58Z | 2025-08-07T12:56:10Z | 2025-08-07T12:56:10Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | luke14free | 166,602 | MDQ6VXNlcjE2NjYwMg== | User | false |
huggingface/diffusers | 3,273,266,009 | I_kwDOHa8MBc7DGhNZ | 12,011 | https://github.com/huggingface/diffusers/issues/12011 | https://api.github.com/repos/huggingface/diffusers/issues/12011 | Wan 2.2 a14b i2v OOM | ### Describe the bug
@a-r-r-o-w @DN6 @asomoza @jlonge4
sorry to tag you all, however after @yiyixuxu 's merge of Wan2.2 PR, one of your commit's is causing the model to OOM. I was only able to narrow it down to this.
Wan2.2 PR, working
commit: #a6d9f6a1a9a9ede2c64972d83ccee192b801c4a0
<img width="902" height="47... | open | null | false | 6 | [
"bug",
"stale"
] | [] | 2025-07-29T11:31:12Z | 2026-02-03T15:19:56Z | null | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | okaris | 1,448,702 | MDQ6VXNlcjE0NDg3MDI= | User | false |
huggingface/diffusers | 3,273,392,175 | I_kwDOHa8MBc7DHAAv | 12,012 | https://github.com/huggingface/diffusers/issues/12012 | https://api.github.com/repos/huggingface/diffusers/issues/12012 | apply_first_block_cache with Wan 2.2 causes ValueError: No context is set. Please set a context before retrieving the state | ### Describe the bug
Following @a-r-r-o-w guide here https://huggingface.co/posts/a-r-r-o-w/278025275110164
I tried both
`apply_first_block_cache(pipe.transformer, FirstBlockCacheConfig(threshold=0.2))`
and
`pipe.transformer.enable_cache(FirstBlockCacheConfig(threshold=input_data.cache_threshold))`
but both yield... | closed | completed | false | 0 | [
"bug"
] | [] | 2025-07-29T12:11:54Z | 2025-07-29T13:29:26Z | 2025-07-29T12:55:10Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | luke14free | 166,602 | MDQ6VXNlcjE2NjYwMg== | User | false |
huggingface/diffusers | 3,274,218,294 | I_kwDOHa8MBc7DKJs2 | 12,017 | https://github.com/huggingface/diffusers/issues/12017 | https://api.github.com/repos/huggingface/diffusers/issues/12017 | Dream 7B | ### Model/Pipeline/Scheduler description
https://hkunlp.github.io/blog/2025/dream/
> In short, Dream 7B:
>
> - consistently outperforms existing diffusion language models by a large margin;
> - matches or exceeds top-tier Autoregressive (AR) language models of similar size on the general, math, and coding > abilitie... | open | null | false | 2 | [
"stale"
] | [] | 2025-07-29T16:15:30Z | 2026-01-09T15:15:35Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ntoxeg | 823,693 | MDQ6VXNlcjgyMzY5Mw== | User | false |
huggingface/diffusers | 3,274,615,521 | I_kwDOHa8MBc7DLqrh | 12,019 | https://github.com/huggingface/diffusers/issues/12019 | https://api.github.com/repos/huggingface/diffusers/issues/12019 | Wan 2.2 First vs Second stage | Wan 2.2 consists of two transformer stages.
Diffusers currently allows to run either fully on first stage or at the boundary switch to second stage (that concept somewhat resembles old-sdxl-refiner).
However, after some testing of Wan2.2, its far more useful to be able to skip first stage and run on second stage only ... | closed | completed | false | 7 | [] | [] | 2025-07-29T18:42:18Z | 2025-08-01T15:29:53Z | 2025-08-01T15:29:53Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | vladmandic | 57,876,960 | MDQ6VXNlcjU3ODc2OTYw | User | false |
huggingface/diffusers | 1,441,161,894 | I_kwDOHa8MBc5V5mKm | 1,202 | https://github.com/huggingface/diffusers/issues/1202 | https://api.github.com/repos/huggingface/diffusers/issues/1202 | Obtaining the image iterations before final image has been generated StableDiffusionPipeline.pretrained() | Hey all. I am currently using the StableDiffusionPipeline to generate AI images with a discord bot which I use with my friends. I was wondering if it was possible to get a preview of the image being generated before it is finished?
For example, if an image takes 20 seconds to generate, since it is using diffusion it... | closed | completed | false | 4 | [
"stale"
] | [] | 2022-11-09T01:02:43Z | 2022-12-18T15:03:15Z | 2022-12-18T15:03:15Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | MonkeeMan1 | 98,274,841 | U_kgDOBduOGQ | User | false |
huggingface/diffusers | 3,276,799,103 | I_kwDOHa8MBc7DT_x_ | 12,022 | https://github.com/huggingface/diffusers/issues/12022 | https://api.github.com/repos/huggingface/diffusers/issues/12022 | _flash_attention_3 in dispatch_attention_fn is not compatible with the latest flash-atten interface. | ### Describe the bug
[FA3] Don't return lse:
https://github.com/Dao-AILab/flash-attention/commit/ed209409acedbb2379f870bbd03abce31a7a51b7
but in the current diffuser version, it is not updated.
https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_dispatch.py#L608
when use fa3 backend, di... | closed | completed | false | 8 | [
"bug",
"stale"
] | [] | 2025-07-30T12:13:50Z | 2026-02-26T12:04:38Z | 2026-02-26T12:04:38Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | hmzjwhmzjw | 23,023,261 | MDQ6VXNlcjIzMDIzMjYx | User | false |
huggingface/diffusers | 3,278,427,878 | I_kwDOHa8MBc7DaNbm | 12,025 | https://github.com/huggingface/diffusers/issues/12025 | https://api.github.com/repos/huggingface/diffusers/issues/12025 | Invalid API call in Cosmos VAE Encoder | ### Describe the bug
At this line, the reshape appears to be called incorrectly. I believe it should either be `hidden_states.reshape(*)` or `torch.reshape(hidden_states, *)`
https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/autoencoders/autoencoder_kl_cosmos.py#L171
It appears like this part o... | closed | completed | false | 4 | [
"bug"
] | [] | 2025-07-30T21:00:07Z | 2025-08-02T14:54:02Z | 2025-08-02T14:54:02Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | akhilg-nv | 165,961,486 | U_kgDOCeRfDg | User | false |
huggingface/diffusers | 3,279,891,593 | I_kwDOHa8MBc7DfyyJ | 12,029 | https://github.com/huggingface/diffusers/issues/12029 | https://api.github.com/repos/huggingface/diffusers/issues/12029 | X-Omni-En model won't load | I am running this notebook:
https://colab.research.google.com/#scrollTo=rRaMa09zUh66&fileId=https%3A//huggingface.co/X-Omni/X-Omni-En.ipynb
And I am getting the following error. That notebook says to post an issue here.
---------------------------------------------------------------------------
HTTPError ... | closed | completed | false | 6 | [
"stale"
] | [] | 2025-07-30T17:38:34Z | 2026-01-09T22:08:19Z | 2026-01-09T22:08:19Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Tylersuard | 41,713,505 | MDQ6VXNlcjQxNzEzNTA1 | User | false |
huggingface/diffusers | 3,280,676,514 | I_kwDOHa8MBc7Diyai | 12,034 | https://github.com/huggingface/diffusers/issues/12034 | https://api.github.com/repos/huggingface/diffusers/issues/12034 | Wan 2.2 5b i2v results poor quality compared to official Wan HF Space | ### Describe the bug
# diffusers result:
<img width="1292" height="772" alt="Image" src="https://github.com/user-attachments/assets/35c30cdc-6c38-48bb-9a8d-71faa1d7896d" />
video links
https://github.com/user-attachments/assets/4dd6d342-9dfb-4946-a714-641f5a4cc98d
(this video might not play in the browser due to a ... | closed | completed | false | 8 | [
"bug"
] | [
"yiyixuxu"
] | 2025-07-31T14:24:30Z | 2025-11-15T20:12:32Z | 2025-08-01T09:43:43Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | okaris | 1,448,702 | MDQ6VXNlcjE0NDg3MDI= | User | false |
huggingface/diffusers | 3,280,959,121 | I_kwDOHa8MBc7Dj3aR | 12,037 | https://github.com/huggingface/diffusers/issues/12037 | https://api.github.com/repos/huggingface/diffusers/issues/12037 | Wan 2.2 WanTransformer3DModel not compatible with Lightx2v self-forcing guidance distilled loras | ### Describe the bug
The latest advancement in Wan has been self forcing loras, which allow to get extremely good results in just 4 steps.
Although the comfy community was successful in using the Lightx2v cfg step distill Wan2.1 loras on Wan2.2, I can't apply them to the trasnformers in any way.
The suggested comb... | closed | completed | false | 2 | [
"bug"
] | [
"sayakpaul"
] | 2025-07-31T15:51:28Z | 2025-08-02T06:13:27Z | 2025-08-02T06:13:27Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | luke14free | 166,602 | MDQ6VXNlcjE2NjYwMg== | User | false |
huggingface/diffusers | 3,281,015,033 | I_kwDOHa8MBc7DkFD5 | 12,038 | https://github.com/huggingface/diffusers/issues/12038 | https://api.github.com/repos/huggingface/diffusers/issues/12038 | Dataset structure for train_text_to_image_lora.py | Hello. I am trying to use **train_text_to_image_lora.py** script following the instructions https://github.com/huggingface/diffusers/tree/main/examples/text_to_image
I get errors on data structure and don't know what is the issue on my side.
I have a folder **data** where I have folder **image** and **csv** file.
C:/... | closed | completed | true | 2 | [
"stale"
] | [] | 2025-07-31T16:10:38Z | 2026-01-09T22:16:39Z | 2026-01-09T22:16:39Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | HripsimeS | 42,246,765 | MDQ6VXNlcjQyMjQ2NzY1 | User | false |
huggingface/diffusers | 3,282,505,199 | I_kwDOHa8MBc7Dpw3v | 12,039 | https://github.com/huggingface/diffusers/issues/12039 | https://api.github.com/repos/huggingface/diffusers/issues/12039 | Wan 2.2 VAE forward fails | Hey,
With Wan2.1 we was able to pass just a RGB PIl image. With 2.2 i get
``` python
def _conv_forward(self, input: Tensor, weight: Tensor, bias: Optional[Tensor]):
if self.padding_mode != "zeros":
return F.conv3d(
F.pad(
input, self._reversed_paddi... | closed | completed | false | 12 | [] | [
"yiyixuxu"
] | 2025-08-01T04:28:43Z | 2025-08-06T12:56:19Z | 2025-08-06T12:56:19Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | a-r-r-o-w | 72,266,394 | MDQ6VXNlcjcyMjY2Mzk0 | User | false |
huggingface/diffusers | 1,441,216,040 | I_kwDOHa8MBc5V5zYo | 1,204 | https://github.com/huggingface/diffusers/issues/1204 | https://api.github.com/repos/huggingface/diffusers/issues/1204 | [Community] Can we composite Dreambooth network training? | Very impressed with Dreambooth capabilities. I have what i think is a feature request - or perhaps a clarification on what is and is not possible in training networks with Dreambooth. In particular, i was wondering if there was a way to composite two networks to enable embedding of two instances (e.g. an sks dog >and< ... | closed | completed | false | 2 | [
"question",
"stale"
] | [
"pcuenca"
] | 2022-11-09T01:59:05Z | 2022-12-21T15:03:19Z | 2022-12-21T15:03:19Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | felgryn | 97,655,026 | U_kgDOBdIY8g | User | false |
huggingface/diffusers | 3,285,462,623 | I_kwDOHa8MBc7D1C5f | 12,044 | https://github.com/huggingface/diffusers/issues/12044 | https://api.github.com/repos/huggingface/diffusers/issues/12044 | AttributeError: 'bool' object has no attribute '__module__'. Did you mean: '__mod__'? | I am train the Flux.1-dev model and get this error. I found the solution to bring diffuser to version 0.21.0 but then it would beconflict with some other libraries. Is there any solution for this?
```
Traceback (most recent call last):
File "/home/quyetnv/t2i/ai-toolkit/run.py", line 120, in <module>
main()
Fi... | closed | completed | false | 3 | [] | [] | 2025-08-02T01:37:30Z | 2025-08-21T01:27:19Z | 2025-08-21T01:25:02Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | quyetnv98 | 47,714,907 | MDQ6VXNlcjQ3NzE0OTA3 | User | false |
huggingface/diffusers | 3,286,158,094 | I_kwDOHa8MBc7D3ssO | 12,047 | https://github.com/huggingface/diffusers/issues/12047 | https://api.github.com/repos/huggingface/diffusers/issues/12047 | Fusing Lightx2v lora on Wan2.2 GGUF fails | ### Describe the bug
Fusing works fine on non-gguf versions of Wan2.2, but yield issues when used with the GGUF transformer which is what most consumers have to use due to memory constraints (which also make fusing quite important).
### Reproduction
```python
import torch
from diffusers import WanImageToVideoPipelin... | closed | completed | false | 9 | [
"bug"
] | [] | 2025-08-02T17:51:06Z | 2025-08-06T15:22:57Z | 2025-08-06T15:19:24Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | luke14free | 166,602 | MDQ6VXNlcjE2NjYwMg== | User | false |
huggingface/diffusers | 1,441,466,789 | I_kwDOHa8MBc5V6wml | 1,205 | https://github.com/huggingface/diffusers/issues/1205 | https://api.github.com/repos/huggingface/diffusers/issues/1205 | converting SD model to onnx | ### Describe the bug
Hi - I am trying to convert stable-diffusion-v1.4 to onnx using the below code.
python convert_stable_diffusion_checkpoint_to_onnx.py --model_path="CompVis/stable-diffusion-v1-4" --output_path="./stable_diffusion_onnx" --fp16="True"
The output I am getting is a model.onnx (430MB) and weights... | closed | completed | false | 2 | [
"bug"
] | [
"anton-l"
] | 2022-11-09T06:30:14Z | 2022-11-30T12:18:23Z | 2022-11-30T12:18:23Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | harishprabhala | 26,444,096 | MDQ6VXNlcjI2NDQ0MDk2 | User | false |
huggingface/diffusers | 3,287,039,666 | I_kwDOHa8MBc7D7D6y | 12,050 | https://github.com/huggingface/diffusers/issues/12050 | https://api.github.com/repos/huggingface/diffusers/issues/12050 | `UNet2DConditionModel` : `qk_norm` setting in `config.json` is ignored | ### Describe the bug
Adding eg `"qk_norm": "rms_norm"` to config.json for a `UNet2DConditionModel` has no effect.
This is because the value is not propagated by the `UNet2DContionalModel` initialization logic through to `Attention.__init__` in `src/diffusers/models/attention_processor.py`.
### Reproduction
Default ... | open | null | false | 5 | [
"bug",
"stale"
] | [] | 2025-08-03T10:06:58Z | 2026-01-09T15:15:18Z | null | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | damian0815 | 144,366 | MDQ6VXNlcjE0NDM2Ng== | User | false |
huggingface/diffusers | 3,287,142,976 | I_kwDOHa8MBc7D7dJA | 12,052 | https://github.com/huggingface/diffusers/issues/12052 | https://api.github.com/repos/huggingface/diffusers/issues/12052 | Wan 2.2 with LightX2V offloading tries to multiply tensors from different devices and fails | ### Describe the bug
After @sayakpaul great work in https://github.com/huggingface/diffusers/pull/12040 LightX2V now works. However what doesn't work is adding both a lora and offloading to the transformer_2. I can get away with either (i.e. offload both transformers but add a lora only to transformer and NOT to trans... | closed | completed | false | 4 | [
"bug"
] | [] | 2025-08-03T12:43:13Z | 2025-08-11T15:53:41Z | 2025-08-08T07:51:48Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | luke14free | 166,602 | MDQ6VXNlcjE2NjYwMg== | User | false |
huggingface/diffusers | 3,287,287,219 | I_kwDOHa8MBc7D8AWz | 12,053 | https://github.com/huggingface/diffusers/issues/12053 | https://api.github.com/repos/huggingface/diffusers/issues/12053 | Flux1.Dev Kohya Loras text encoder layers no more supported | Hello,
I trained a Lora with Kohya SS and I have a problem of conversion, I thought it should have been managed by your conversion script ?
```
Loading adapter weights from state_dict led to unexpected keys found in the model: single_transformer_blocks.0.proj_out.lora_A.default_0.weight, single_transformer_blocks.0.p... | closed | completed | false | 22 | [] | [] | 2025-08-03T15:47:59Z | 2026-02-27T10:13:43Z | 2026-02-27T10:13:43Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | christopher5106 | 6,875,375 | MDQ6VXNlcjY4NzUzNzU= | User | false |
huggingface/diffusers | 3,288,731,263 | I_kwDOHa8MBc7EBg5_ | 12,060 | https://github.com/huggingface/diffusers/issues/12060 | https://api.github.com/repos/huggingface/diffusers/issues/12060 | Is there any DiT block defined in the huggingface/diffusers OR huggingface/transformers project? | **Is your feature request related to a problem? Please describe.**
I want to make some experiments about DiT based flow-matching model, I need an implementation of the common DiT block, but did not found it in both huggingface/diffusers and huggingface/transformers. Is there any implementation about it with just some ... | open | null | false | 3 | [
"stale"
] | [] | 2025-08-04T09:40:43Z | 2026-01-09T15:15:11Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | JohnHerry | 8,011,802 | MDQ6VXNlcjgwMTE4MDI= | User | false |
huggingface/diffusers | 3,290,456,768 | I_kwDOHa8MBc7EIGLA | 12,065 | https://github.com/huggingface/diffusers/issues/12065 | https://api.github.com/repos/huggingface/diffusers/issues/12065 | Qwen Image | Image-To-Image + Editing + Inpainting | # Qwen Image | Editing Capabilities
> When it comes to image editing, Qwen-Image goes far beyond simple adjustments. It enables advanced operations such as style transfer, object insertion or removal, detail enhancement, text ed... | closed | completed | false | 3 | [] | [] | 2025-08-04T18:50:38Z | 2025-08-13T14:45:18Z | 2025-08-04T19:10:16Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ghunkins | 12,562,057 | MDQ6VXNlcjEyNTYyMDU3 | User | false |
huggingface/diffusers | 3,290,566,900 | I_kwDOHa8MBc7EIhD0 | 12,066 | https://github.com/huggingface/diffusers/issues/12066 | https://api.github.com/repos/huggingface/diffusers/issues/12066 | Qwen Image incorrect device assignment during prompt encode | ### Describe the bug
In the new `QwenImagePipeline` method `_get_qwen_prompt_embeds` it does following:
```py
txt_tokens = self.tokenizer(
txt, max_length=self.tokenizer_max_length + drop_idx, padding=True, truncation=True, return_tensors="pt"
).to(self.device)
```
i assume this is a typo a... | closed | completed | false | 2 | [
"bug"
] | [] | 2025-08-04T19:38:02Z | 2025-08-07T22:27:40Z | 2025-08-07T22:27:40Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | vladmandic | 57,876,960 | MDQ6VXNlcjU3ODc2OTYw | User | false |
huggingface/diffusers | 1,441,551,773 | I_kwDOHa8MBc5V7FWd | 1,207 | https://github.com/huggingface/diffusers/issues/1207 | https://api.github.com/repos/huggingface/diffusers/issues/1207 | AttributeError: /opt/conda/bin/python: undefined symbol: cudaRuntimeGetVersion | ### Describe the bug
```
AttributeError: /opt/conda/bin/python: undefined symbol: cudaRuntimeGetVersion
```
### Reproduction
```
%cd /workspace
!git clone https://github.com/huggingface/diffusers.git
```
```
%cd /workspace/diffusers/examples/dreambooth
!pwd
```
```
pip install -U -r requirements.t... | closed | completed | false | 6 | [
"bug"
] | [] | 2022-11-09T07:22:27Z | 2022-12-24T02:32:25Z | 2022-11-15T11:53:55Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | 0xdevalias | 753,891 | MDQ6VXNlcjc1Mzg5MQ== | User | false |
huggingface/diffusers | 3,292,205,997 | I_kwDOHa8MBc7EOxOt | 12,071 | https://github.com/huggingface/diffusers/issues/12071 | https://api.github.com/repos/huggingface/diffusers/issues/12071 | flux kontext transformer single blocks forward behavior changed | ### Describe the bug
I observed that this line of code(https://github.com/huggingface/diffusers/blob/0454fbb30bfbe21aa4ea29c827c396bac57dc518/src/diffusers/models/transformers/transformer_flux.py#L88) was added for `First Block Cache`.
However, it not only increases the computation, but also thoroughly changes the f... | closed | completed | false | 3 | [
"bug"
] | [] | 2025-08-05T08:55:14Z | 2025-08-05T11:47:53Z | 2025-08-05T11:47:53Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | WingEdge777 | 16,758,743 | MDQ6VXNlcjE2NzU4NzQz | User | false |
huggingface/diffusers | 3,293,162,769 | I_kwDOHa8MBc7ESa0R | 12,075 | https://github.com/huggingface/diffusers/issues/12075 | https://api.github.com/repos/huggingface/diffusers/issues/12075 | Qwen Image prompt encoding is not padding to max seq len | ### Describe the bug
The pipeline method for QwenImagePipeline.encode_prompts is not padding correctly; it's padding by the longest sequence length in the batch, which leaves very very short embeds that are out of distribution for the models' training set.
The padding should remain at 1024 tokens even after the syste... | open | null | false | 22 | [
"bug"
] | [] | 2025-08-05T13:38:09Z | 2026-02-03T11:13:06Z | null | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | bghira | 59,658,056 | MDQ6VXNlcjU5NjU4MDU2 | User | false |
huggingface/diffusers | 3,294,612,975 | I_kwDOHa8MBc7EX83v | 12,078 | https://github.com/huggingface/diffusers/issues/12078 | https://api.github.com/repos/huggingface/diffusers/issues/12078 | Problem with provided example validation input in the Flux Control finetuning example | ### Describe the bug
The help page for the Flux control finetuning example, https://github.com/huggingface/diffusers/blob/main/examples/flux-control/README.md, provides a sample validation input, a pose condition image
[<img src="https://huggingface.co/api/resolve-cache/models/Adapter/t2iadapter/3c291e0547a1b17bed9342... | open | null | false | 2 | [
"bug",
"stale"
] | [] | 2025-08-05T22:29:35Z | 2026-01-09T15:15:01Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | kzhang2 | 12,497,065 | MDQ6VXNlcjEyNDk3MDY1 | User | false |
huggingface/diffusers | 3,294,962,295 | I_kwDOHa8MBc7EZSJ3 | 12,079 | https://github.com/huggingface/diffusers/issues/12079 | https://api.github.com/repos/huggingface/diffusers/issues/12079 | API Suggestion: Expose Methods to Convert to Sample Prediction in Schedulers | **What API design would you like to have changed or added to the library? Why?**
My proposal is for schedulers to expose `convert_to_sample_prediction` and `convert_to_prediction_type` methods, which would do the following:
1. `convert_to_sample_prediction`: Converts from a given `prediction_type` to `sample_predicti... | open | null | false | 1 | [
"stale"
] | [] | 2025-08-06T02:24:46Z | 2026-01-09T15:14:57Z | null | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | dg845 | 58,458,699 | MDQ6VXNlcjU4NDU4Njk5 | User | false |
huggingface/diffusers | 3,295,077,009 | I_kwDOHa8MBc7EZuKR | 12,080 | https://github.com/huggingface/diffusers/issues/12080 | https://api.github.com/repos/huggingface/diffusers/issues/12080 | Qwen Image : Image Editing Inference | **Is your feature request related to a problem? Please describe.**
A clear and concise description of what the problem is. Ex. I'm always frustrated when [...].
The technical report and huggingface page describe qwen_image's image editing capabilities in detail, but the inference code does not have that capability.
*... | closed | completed | false | 1 | [] | [] | 2025-08-06T03:46:11Z | 2025-08-06T17:35:25Z | 2025-08-06T17:35:25Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | PranavAdlinge | 56,697,400 | MDQ6VXNlcjU2Njk3NDAw | User | false |
huggingface/diffusers | 3,295,630,526 | I_kwDOHa8MBc7Eb1S- | 12,082 | https://github.com/huggingface/diffusers/issues/12082 | https://api.github.com/repos/huggingface/diffusers/issues/12082 | wan2.1 vae take more gpu memory after compile | ### Describe the bug
After `torch.compile` wan2.1 vae consume more GPU memory than `no compilation`, which is unexpected in my opinion.
**compiled**
<img width="3006" height="1158" alt="Image" src="https://github.com/user-attachments/assets/3eff903c-af4f-422a-b407-9afdd77ef843" />
**no-compile**
<img width="2850" he... | open | null | false | 4 | [
"bug",
"stale"
] | [] | 2025-08-06T07:57:54Z | 2026-02-03T15:19:43Z | null | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | gameofdimension | 32,255,912 | MDQ6VXNlcjMyMjU1OTEy | User | false |
huggingface/diffusers | 3,295,650,521 | I_kwDOHa8MBc7Eb6LZ | 12,083 | https://github.com/huggingface/diffusers/issues/12083 | https://api.github.com/repos/huggingface/diffusers/issues/12083 | Qwen-Image long prompt will cause error | ### Describe the bug
When the token length is greater than 1024, it will be truncated to 1024. However, the length of RoPE is fixed at 1024 because the image takes up 32 len (for a 1024 width and height image). This causes the length of txt_freqs to be less than 1024. Therefore, x_rotated * freqs_cis will generate an ... | closed | completed | false | 2 | [
"bug"
] | [] | 2025-08-06T08:03:23Z | 2025-08-08T06:28:26Z | 2025-08-08T06:28:26Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | akk-123 | 98,469,560 | U_kgDOBd6GuA | User | false |
huggingface/diffusers | 3,296,306,911 | I_kwDOHa8MBc7Eeabf | 12,084 | https://github.com/huggingface/diffusers/issues/12084 | https://api.github.com/repos/huggingface/diffusers/issues/12084 | Will `cosmos-transfer1` be supported in diffusers in the future? |
Hi @a-r-r-o-w and @yiyixuxu :)
First of all, thank you for recently enabling cosmos-predict1 models (text2world and video2world) in the diffusers library — it's super exciting to see them integrated!
I was wondering if there are any plans to also support [cosmos-transfer1](https://github.com/nvidia-cosmos/cosmos-tr... | open | null | false | 4 | [
"stale"
] | [] | 2025-08-06T11:22:28Z | 2026-01-09T15:14:50Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | rebel-shshin | 87,514,200 | MDQ6VXNlcjg3NTE0MjAw | User | false |
huggingface/diffusers | 3,296,455,383 | I_kwDOHa8MBc7Ee-rX | 12,085 | https://github.com/huggingface/diffusers/issues/12085 | https://api.github.com/repos/huggingface/diffusers/issues/12085 | WanVACETransformer3DModel load gguf error | ### Describe the bug
When attempting to load a model using WanVACETransformer3DModel.from_single_file(), I encounter a ValueError indicating that the model type is not compatible with FromOriginalModelMixin. The error suggests that WanVACETransformer3DModel is not in the list of supported model types.
### Reproductio... | closed | completed | false | 0 | [
"bug"
] | [] | 2025-08-06T12:09:48Z | 2025-08-06T12:43:42Z | 2025-08-06T12:43:42Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | 00Neil | 76,234,364 | MDQ6VXNlcjc2MjM0MzY0 | User | false |
huggingface/diffusers | 3,300,136,658 | I_kwDOHa8MBc7EtBbS | 12,094 | https://github.com/huggingface/diffusers/issues/12094 | https://api.github.com/repos/huggingface/diffusers/issues/12094 | [Wan2.2] pipeline_wan miss the 'shift' parameter which used by Wan2.2-A14B-diffusers. | **Firstly, I found that the quality of output using diffusers is poor**
Later, I found that the pipeline_wan in diffusers[0.34.0] did not support two-stage processing. I noticed that the community had already updated it, so I installed diffusers[0.35.0-dev] by source code and it worked.
Then I found that the scheduler... | closed | completed | false | 7 | [] | [] | 2025-08-07T11:37:36Z | 2025-08-10T08:43:27Z | 2025-08-10T08:43:26Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | yvmilir | 42,015,376 | MDQ6VXNlcjQyMDE1Mzc2 | User | false |
huggingface/diffusers | 3,300,343,252 | I_kwDOHa8MBc7Etz3U | 12,096 | https://github.com/huggingface/diffusers/issues/12096 | https://api.github.com/repos/huggingface/diffusers/issues/12096 | WanVACEPipeline - doesn't work with apply_group_offloading | ### Describe the bug
When enable apply_group_offloading:
```
Traceback (most recent call last):
File "D:\Experiments\Video_Outpaint\2__Outpaint.py", line 165, in <module>
out = pipe(
^^^^^
File "C:\Users\BBCCA\AppData\Local\Programs\Python\Python312\Lib\site-packages\torch\utils\_contextlib.py", line... | closed | completed | false | 12 | [
"bug",
"contributions-welcome",
"group-offloading"
] | [] | 2025-08-07T12:37:50Z | 2025-12-06T00:28:02Z | 2025-12-06T00:28:02Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | SlimRG | 39,348,033 | MDQ6VXNlcjM5MzQ4MDMz | User | false |
huggingface/diffusers | 3,300,406,129 | I_kwDOHa8MBc7EuDNx | 12,097 | https://github.com/huggingface/diffusers/issues/12097 | https://api.github.com/repos/huggingface/diffusers/issues/12097 | Wan2.2 TI2V-5B VRAM OOM at the end | ### Describe the bug
After completing 50 steps of progress, the video memory usage skyrocketed from around 8GB to 26GB, resulting in very slow performance
### Reproduction
```
import torch
import numpy as np
from diffusers import WanImageToVideoPipeline, AutoencoderKLWan, ModularPipeline
from diffusers.utils import ... | open | null | false | 1 | [
"bug",
"stale"
] | [] | 2025-08-07T12:55:44Z | 2026-01-09T15:14:37Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | zhaoyun0071 | 35,762,050 | MDQ6VXNlcjM1NzYyMDUw | User | false |
huggingface/diffusers | 3,300,415,796 | I_kwDOHa8MBc7EuFk0 | 12,098 | https://github.com/huggingface/diffusers/issues/12098 | https://api.github.com/repos/huggingface/diffusers/issues/12098 | Qwen image transformers doesn't currently support from_single_file (i.e. GGUFs) | Currently it looks like qwen image doesn't support gguf since it's missing a conversion map/function from gguf to diffuser. I'd love to help creating it if you can point me in the right direction, I tried looking into it but without guidance it's above my paygrade :) | closed | completed | false | 4 | [] | [] | 2025-08-07T12:58:43Z | 2025-08-08T19:45:00Z | 2025-08-08T19:45:00Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | luke14free | 166,602 | MDQ6VXNlcjE2NjYwMg== | User | false |
huggingface/diffusers | 3,302,530,870 | I_kwDOHa8MBc7E2J82 | 12,102 | https://github.com/huggingface/diffusers/issues/12102 | https://api.github.com/repos/huggingface/diffusers/issues/12102 | Support training where width does not equal height for Qwen-Image | To support training where width does not equal height for Qwen-Image, the following code:
```
img_shapes = [
(1, args.resolution // vae_scale_factor // 2, args.resolution // vae_scale_factor // 2)
] * bsz
noisy_model_input = noisy_model_input.permute(0, 2, 1, 3, 4)
packed_noisy_model_input = QwenImagePipeline._pac... | open | null | false | 2 | [
"stale"
] | [] | 2025-08-08T03:57:10Z | 2026-02-03T15:19:33Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | yinguoweiOvO | 56,142,257 | MDQ6VXNlcjU2MTQyMjU3 | User | false |
huggingface/diffusers | 3,303,303,390 | I_kwDOHa8MBc7E5Gje | 12,104 | https://github.com/huggingface/diffusers/issues/12104 | https://api.github.com/repos/huggingface/diffusers/issues/12104 | IndexError: index 0 is out of bounds for dimension 0 with size 0 | ### Describe the bug
When I test the mit-han-lab/nunchaku-flux.1-kontext-dev model, it runs normally in a non-concurrent scenario, but throws an error when I try to run it with concurrent requests.
My GPU is a single RTX 4090D.
How can I enable multi-concurrency support on a single GPU?
Thank you in advance for yo... | closed | completed | true | 1 | [
"bug"
] | [] | 2025-08-08T09:20:52Z | 2025-08-17T22:22:37Z | 2025-08-17T22:22:37Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | liushiton | 115,603,416 | U_kgDOBuP32A | User | false |
huggingface/diffusers | 3,306,188,208 | I_kwDOHa8MBc7FEG2w | 12,107 | https://github.com/huggingface/diffusers/issues/12107 | https://api.github.com/repos/huggingface/diffusers/issues/12107 | accelerator.init_trackers error when try with a custom object such as list | ### Describe the bug
I set multiple prompts with nargs for argument "--validation_prompt " in "train_dreambooth.py":
` parser.add_argument(
"--validation_prompt",
type=str,
default=["A photo of sks dog in a bucket", "A sks cat wearing a coat"],
nargs="*",
help="A prompt that... | open | null | false | 1 | [
"bug",
"stale"
] | [] | 2025-08-09T10:04:06Z | 2026-01-09T15:14:27Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | micklexqg | 13,776,012 | MDQ6VXNlcjEzNzc2MDEy | User | false |
huggingface/diffusers | 3,306,859,561 | I_kwDOHa8MBc7FGqwp | 12,108 | https://github.com/huggingface/diffusers/issues/12108 | https://api.github.com/repos/huggingface/diffusers/issues/12108 | Qwen Image and Chroma pipeline breaks using schedulers that enable flow matching by parameter. | ### Describe the bug
Several Schedulers support flow matching by using the prediction_type='flow_prediction" e.g.
```
pipe.scheduler = UniPCMultistepScheduler(prediction_type="flow_prediction", flow_shift=3.16, timestep_spacing='trailing', use_flow_sigmas=True)
```
However Chroma and Qwen Image will not work with th... | closed | completed | false | 1 | [
"bug"
] | [] | 2025-08-09T21:34:28Z | 2026-01-22T01:19:00Z | 2026-01-22T01:19:00Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Vargol | 62,868 | MDQ6VXNlcjYyODY4 | User | false |
huggingface/diffusers | 3,307,203,137 | I_kwDOHa8MBc7FH-pB | 12,110 | https://github.com/huggingface/diffusers/issues/12110 | https://api.github.com/repos/huggingface/diffusers/issues/12110 | Bug in initialization of UNet1DModel GaussianFourier time projection | ### Describe the bug
TLDR:
In the constructor for the UNet1dModel (line 104, seen [here](https://github.com/huggingface/diffusers/blob/v0.34.0/src/diffusers/models/unets/unet_1d.py#L104)), the embedding size is manually hardcoded to be 8 for no apparent good reason. Instead it should be block_out_channels[0].
Expla... | closed | completed | false | 1 | [
"bug"
] | [] | 2025-08-10T05:45:35Z | 2025-09-16T06:48:49Z | 2025-09-16T06:48:49Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | SammyAgrawal | 41,808,786 | MDQ6VXNlcjQxODA4Nzg2 | User | false |
huggingface/diffusers | 3,307,394,229 | I_kwDOHa8MBc7FItS1 | 12,112 | https://github.com/huggingface/diffusers/issues/12112 | https://api.github.com/repos/huggingface/diffusers/issues/12112 | Various Chroma pipeline issues | ### Describe the bug
Some minor issues that I have found while using the Chroma pipeline for the first time:
- if `sentencepiece` is not installed, you get a quite cryptic error message. Other pipelines such as Flux give you a nice error message that `sentencepiece` has to be installed
- `The config attributes {'guid... | open | null | false | 4 | [
"bug",
"stale"
] | [] | 2025-08-10T09:42:51Z | 2026-02-03T15:19:29Z | null | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | dxqb | 183,307,934 | U_kgDOCu0Ong | User | false |
huggingface/diffusers | 3,307,522,406 | I_kwDOHa8MBc7FJMlm | 12,113 | https://github.com/huggingface/diffusers/issues/12113 | https://api.github.com/repos/huggingface/diffusers/issues/12113 | WanImageToVideoPipeline: Given groups=1, weight of size [160, 12, 3, 3, 3], expected input[1, 3, 3, 258, 258] to have 12 channels, but got 3 channels instead | ### Describe the bug
Trying the example code with the following changes
pipe.vae.enable_tiling()
pipe.enable_sequential_cpu_offload()
Uninstalled and installed diffusers and peft from source before trying the example code.
### Reproduction
```python
import torch
import numpy as np
from diffusers import WanImageTo... | closed | completed | false | 0 | [
"bug"
] | [] | 2025-08-10T12:23:57Z | 2025-10-09T15:54:57Z | 2025-10-09T15:54:57Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | nitinmukesh | 2,102,186 | MDQ6VXNlcjIxMDIxODY= | User | false |
huggingface/diffusers | 3,307,810,586 | I_kwDOHa8MBc7FKS8a | 12,116 | https://github.com/huggingface/diffusers/issues/12116 | https://api.github.com/repos/huggingface/diffusers/issues/12116 | Attention masking in Chroma pipeline | ### Describe the bug
There is an issue with attention masking in the Chroma pipeline. With the prompt in your example here, https://huggingface.co/docs/diffusers/main/api/pipelines/chroma the difference is not very large, probably because there are enough meaningful tokens with some weight.
But short prompts fail bec... | closed | completed | false | 5 | [
"bug"
] | [] | 2025-08-10T17:44:11Z | 2025-09-29T08:50:07Z | 2025-09-29T08:50:07Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | dxqb | 183,307,934 | U_kgDOCu0Ong | User | false |
huggingface/diffusers | 1,441,824,910 | I_kwDOHa8MBc5V8ICO | 1,212 | https://github.com/huggingface/diffusers/issues/1212 | https://api.github.com/repos/huggingface/diffusers/issues/1212 | Community Integration: Making AIGC cheaper, faster, and more efficient. | **Is your feature request related to a problem? Please describe.**
AIGC has recently risen to be one of the hottest topics in AI. Unfortunately, large hardware requirements and training costs are still a severe impediment to the rapid growth of the AIGC industry. The Stable Diffusion v1 version of the model requires 1... | closed | completed | false | 18 | [
"stale"
] | [] | 2022-11-09T10:21:59Z | 2023-03-30T16:58:22Z | 2023-02-07T15:03:59Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | binmakeswell | 61,670,638 | MDQ6VXNlcjYxNjcwNjM4 | User | false |
huggingface/diffusers | 3,308,304,867 | I_kwDOHa8MBc7FMLnj | 12,120 | https://github.com/huggingface/diffusers/issues/12120 | https://api.github.com/repos/huggingface/diffusers/issues/12120 | How to train a lora with distilled flux model, such as flux-schnell??? | **Is your feature request related to a problem? Please describe.**
I can use flux as base model to train a lora, but it need 20 steps , it cost a lot of time , and I want to train a lora base on distill model to implement use fewer step make a better image, such as based on flux-schnell model train a lora it only nee... | closed | completed | true | 2 | [
"stale"
] | [] | 2025-08-11T03:07:42Z | 2026-01-09T23:20:30Z | 2026-01-09T23:20:30Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Johnson-yue | 10,268,274 | MDQ6VXNlcjEwMjY4Mjc0 | User | false |
huggingface/diffusers | 3,309,578,412 | I_kwDOHa8MBc7FRCis | 12,123 | https://github.com/huggingface/diffusers/issues/12123 | https://api.github.com/repos/huggingface/diffusers/issues/12123 | "Requested to load AutoencoderKL" | ### Describe the bug
I'm tryna generate images but it always crashes when it gets to this line. I'm no expert when it comes to stable diffusion so i don't really know what's going..
### Reproduction
got prompt
model weight dtype torch.float16, manual cast: None
model_type EPS
Using pytorch attention in VAE
Using pyt... | closed | completed | false | 1 | [
"bug"
] | [] | 2025-08-11T11:17:40Z | 2025-08-17T22:23:06Z | 2025-08-17T22:23:06Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ShadyRaion | 112,407,809 | U_kgDOBrM1AQ | User | false |
huggingface/diffusers | 3,310,011,998 | I_kwDOHa8MBc7FSsZe | 12,124 | https://github.com/huggingface/diffusers/issues/12124 | https://api.github.com/repos/huggingface/diffusers/issues/12124 | For qwen-image training file, Maybe "shuffle" of dataloader should be "False" when custom_instance_prompts is not None and cache_latents is False? | ### Describe the bug
I think "shuffle" of dataloader should be "False" when custom_instance_prompts is not None and cache_latents is False. Otherwise, it will lead to errors in the correspondence between prompt embedding and image during training, and prompt will not be followed when performing the task of T2I.
### R... | open | null | false | 4 | [
"bug",
"stale"
] | [] | 2025-08-11T13:15:21Z | 2026-02-03T15:19:24Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | yinguoweiOvO | 56,142,257 | MDQ6VXNlcjU2MTQyMjU3 | User | false |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.