repo stringclasses 1
value | github_id int64 1.27B 4.95B | github_node_id stringlengths 18 24 | number int64 8 14.3k | 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 2026-07-22 10:32:58 | updated_at stringdate 2022-06-12 22:18:01 2026-07-22 10:32:58 | closed_at stringdate 2022-06-12 22:18:01 2026-07-21 13:17:28 ⌀ | author_association stringclasses 3
values | milestone_title stringclasses 0
values | snapshot_id stringclasses 149
values | extracted_at stringdate 2026-04-07 13:34:13 2026-07-22 10:35:16 | author_login stringlengths 3 28 | author_id int64 1.54k 296M | author_node_id stringlengths 12 20 | author_type stringclasses 3
values | author_site_admin bool 1
class |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
huggingface/diffusers | 2,065,156,148 | I_kwDOHa8MBc57F8g0 | 6,452 | https://github.com/huggingface/diffusers/issues/6452 | https://api.github.com/repos/huggingface/diffusers/issues/6452 | add masked_image_latents input on StableDiffusionControlNetInpaintPipeline | **Is your feature request related to a problem? Please describe.**
* In `StableDiffusionInpaintPipeline`'s `__call__` method, there is a input argument called `masked_image_latents`, which can override encoded masked_image, but in `StableDiffusionControlNetInpaintPipeline` it seems not.
To align inputs I think it wil... | closed | completed | false | 2 | [
"stale"
] | [] | 2024-01-04T07:40:25Z | 2024-02-12T07:41:23Z | 2024-02-11T15:04:26Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | moono | 6,761,398 | MDQ6VXNlcjY3NjEzOTg= | User | false |
huggingface/diffusers | 2,065,186,794 | I_kwDOHa8MBc57GD_q | 6,454 | https://github.com/huggingface/diffusers/issues/6454 | https://api.github.com/repos/huggingface/diffusers/issues/6454 | ValueError: Attempting to unscale FP16 gradients. | ### Describe the bug
Traceback (most recent call last):
File "train_text_to_image_lora.py", line 950, in <module>
main()
File "train_text_to_image_lora.py", line 777, in main
accelerator.clip_grad_norm_(params_to_clip, args.max_grad_norm)
File "/home/qc/miniconda3/envs/lora-diffussion/lib/python3.8/... | closed | completed | false | 10 | [
"bug"
] | [] | 2024-01-04T08:04:03Z | 2025-03-03T10:07:56Z | 2024-01-07T08:09:11Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | qinchangchang | 95,670,613 | U_kgDOBbPRVQ | User | false |
huggingface/diffusers | 2,065,601,112 | I_kwDOHa8MBc57HpJY | 6,457 | https://github.com/huggingface/diffusers/issues/6457 | https://api.github.com/repos/huggingface/diffusers/issues/6457 | Gradient Accumulation - Multi-GPU training of SDXL-LoRA | ### Describe the bug
I'm not quite sure whether it's a bug or a discussion -
Looking at the example scripts for dreambooth training (SDXL-LoRA, or the more "traditional" training script), I see that the accumulation of gradients, via `accelerate`, in the beginning of the training loop, is set the accumulate gradient... | closed | completed | false | 10 | [
"bug",
"stale"
] | [] | 2024-01-04T12:46:20Z | 2024-03-24T15:13:24Z | 2024-03-24T15:13:24Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Dvirbeno | 15,541,443 | MDQ6VXNlcjE1NTQxNDQz | User | false |
huggingface/diffusers | 2,065,994,170 | I_kwDOHa8MBc57JJG6 | 6,458 | https://github.com/huggingface/diffusers/issues/6458 | https://api.github.com/repos/huggingface/diffusers/issues/6458 | **aMUSEd** - a new masked image model based on MUSE 🔥 | Today, we're excited to bring you aMUSEd - a lightweight masked-image-model (MIM) intended for fast generation 🚀
We're releasing aMUSEd as a research release because its quality is NOT SoTA.
Training code, paper, usage, fine-tuning code 👇
https://huggingface.co/blog/amused
 🔥
This is what you'd be looking forward to in the post:
... | closed | completed | false | 2 | [] | [] | 2024-01-04T16:47:03Z | 2024-02-13T15:04:48Z | 2024-02-13T15:04:48Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sayakpaul | 22,957,388 | MDQ6VXNlcjIyOTU3Mzg4 | User | false |
huggingface/diffusers | 1,387,824,022 | I_kwDOHa8MBc5SuIOW | 646 | https://github.com/huggingface/diffusers/issues/646 | https://api.github.com/repos/huggingface/diffusers/issues/646 | 'PNDMScheduler' object has no attribute 'set_format' | ### Describe the bug
Getting this error from pipeline_stable_diffusion.py
### Reproduction
_No response_
### Logs
_No response_
### System Info
On colab notebook: https://colab.research.google.com/github/WASasquatch/easydiffusion/blob/main/Stability_AI_Easy_Diffusion.ipynb | closed | completed | false | 10 | [
"bug",
"stale"
] | [] | 2022-09-27T14:01:59Z | 2022-11-08T15:03:41Z | 2022-11-08T15:03:41Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | jalexs82 | 69,216,807 | MDQ6VXNlcjY5MjE2ODA3 | User | false |
huggingface/diffusers | 2,066,048,311 | I_kwDOHa8MBc57JWU3 | 6,460 | https://github.com/huggingface/diffusers/issues/6460 | https://api.github.com/repos/huggingface/diffusers/issues/6460 | Endless loading for loading model from runwayml/stable-diffusion-inpainting | ### Model/Pipeline/Scheduler description
model_index.json: 100%548/548 [00:00<00:00, 36.9kB/s]
unet/diffusion_pytorch_model.safetensors not found
Fetching 24 files: 0% 0/24 [00:00<?, ?it/s]
this issue 'i'm dfacing while running in google colab the model is not is loading
code is this ->
devic... | closed | completed | false | 6 | [] | [] | 2024-01-04T17:17:45Z | 2024-01-05T15:44:31Z | 2024-01-05T15:44:30Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | 7iFinalBoSS | 87,640,946 | MDQ6VXNlcjg3NjQwOTQ2 | User | false |
huggingface/diffusers | 2,066,625,662 | I_kwDOHa8MBc57LjR- | 6,461 | https://github.com/huggingface/diffusers/issues/6461 | https://api.github.com/repos/huggingface/diffusers/issues/6461 | [Design Discussion] allowing `from_pretrained()` to also load single file checkpoints | Since we were considering adding an option like `single_file_format` to `save_pretrained()` of `DiffusionPipeline`, it makes sense to have something similar in `from_pretrained()` to have better feature parity.
We currently support loading single file checkpoints in `DiffusionPipeline` via `from_single_file()`. Som... | closed | completed | false | 20 | [
"stale"
] | [] | 2024-01-05T02:16:21Z | 2024-11-21T15:43:46Z | 2024-11-21T15:43:45Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sayakpaul | 22,957,388 | MDQ6VXNlcjIyOTU3Mzg4 | User | false |
huggingface/diffusers | 2,066,674,377 | I_kwDOHa8MBc57LvLJ | 6,463 | https://github.com/huggingface/diffusers/issues/6463 | https://api.github.com/repos/huggingface/diffusers/issues/6463 | Why is pipelines/audio_diffusion missing？ | ### Describe the bug
https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/audio_diffusion/pipeline_audio_diffusion.py
Why is pipelines/audio_diffusion missing? Where did it go?
### Reproduction
My code is throwing an error, and I'm not sure how to fix it.
```
# Load a pre-trained audio d... | closed | completed | false | 3 | [
"bug"
] | [] | 2024-01-05T03:35:07Z | 2024-01-08T01:06:44Z | 2024-01-08T01:06:44Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ZH-UCAS | 23,288,916 | MDQ6VXNlcjIzMjg4OTE2 | User | false |
huggingface/diffusers | 1,387,877,790 | I_kwDOHa8MBc5SuVWe | 647 | https://github.com/huggingface/diffusers/issues/647 | https://api.github.com/repos/huggingface/diffusers/issues/647 | tensor_format in new Dreambooth example | ### Describe the bug
Hi! After merging the Dreambooth example script, PR #534 was merged, but the `tensor_format` wasn't removed from the Dreambooth script. This gives an error when running the script.
@patil-suraj @patrickvonplaten
### Reproduction
_No response_
### Logs
_No response_
### System ... | closed | completed | false | 1 | [
"bug"
] | [
"kashif",
"patil-suraj"
] | 2022-09-27T14:36:50Z | 2022-09-27T20:05:59Z | 2022-09-27T20:05:59Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ghost | 10,137 | MDQ6VXNlcjEwMTM3 | User | false |
huggingface/diffusers | 2,068,338,569 | I_kwDOHa8MBc57SFeJ | 6,474 | https://github.com/huggingface/diffusers/issues/6474 | https://api.github.com/repos/huggingface/diffusers/issues/6474 | how to use xformers | Maybe this is a relatively low-level question, but what always bothers me is how does Xformer run when running SD? Or can it be accelerated by default after installing this library? Thank you all for answering your questions | closed | completed | false | 6 | [] | [] | 2024-01-06T03:34:16Z | 2024-01-11T03:38:19Z | 2024-01-11T03:28:26Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | babyta | 37,787,795 | MDQ6VXNlcjM3Nzg3Nzk1 | User | false |
huggingface/diffusers | 2,069,041,525 | I_kwDOHa8MBc57UxF1 | 6,476 | https://github.com/huggingface/diffusers/issues/6476 | https://api.github.com/repos/huggingface/diffusers/issues/6476 | export_to_video crashe after opencv version upgrade | ### Describe the bug
Due to some issues with OpenCV dependencies, even if you use the latest version of Gradio, Diffusers, and Transformers, you will still get an error when saving the video.
Maybe later, submit a PR
### Reproduction
Simply put, when loading the model and processing user input, using export_to_vi... | closed | completed | false | 4 | [
"bug",
"stale"
] | [] | 2024-01-07T08:34:23Z | 2024-02-19T15:04:50Z | 2024-02-19T15:04:50Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | soulteary | 1,500,781 | MDQ6VXNlcjE1MDA3ODE= | User | false |
huggingface/diffusers | 2,069,254,984 | I_kwDOHa8MBc57VlNI | 6,480 | https://github.com/huggingface/diffusers/issues/6480 | https://api.github.com/repos/huggingface/diffusers/issues/6480 | why diffusers use more ram comfyui? | why diffusers use more ram comfyui?
When I load sdxl in comfyui it only uses 1-2 gb of ram but diffusers uses 8-9 gb of ram, why is this difference and how to make diffusers load sdxl and use 1-2 gb of ram? | closed | completed | false | 2 | [
"stale"
] | [] | 2024-01-07T19:19:17Z | 2024-02-15T15:05:17Z | 2024-02-15T15:05:17Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | loboere | 15,265,895 | MDQ6VXNlcjE1MjY1ODk1 | User | false |
huggingface/diffusers | 2,069,330,536 | I_kwDOHa8MBc57V3po | 6,481 | https://github.com/huggingface/diffusers/issues/6481 | https://api.github.com/repos/huggingface/diffusers/issues/6481 | Dreambooth training script doesn't work when train batch size is greater than one | ### Describe the bug
The script only works when train batch size is one. If more than one, this line got a raise https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/train_dreambooth.py#L1299
Possibly due to grad accumulation increase the batch size by N and Batch x N is not the same as the loss w... | closed | completed | false | 5 | [
"bug",
"stale",
"training"
] | [
"sayakpaul"
] | 2024-01-07T22:57:34Z | 2024-03-13T15:09:40Z | 2024-03-13T15:09:40Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ChenchaoZhao | 35,147,961 | MDQ6VXNlcjM1MTQ3OTYx | User | false |
huggingface/diffusers | 2,069,550,831 | I_kwDOHa8MBc57Wtbv | 6,482 | https://github.com/huggingface/diffusers/issues/6482 | https://api.github.com/repos/huggingface/diffusers/issues/6482 | train_dreambooth_lora_sdxl_advanced.py --resume_from_checkpoint fails with ValueError: Attempting to unscale FP16 gradients. | ### Describe the bug
After a system crash I attempted to resume from a prior checkpoint.
Expected Result:
Continues training from last checkpoint and finishes successfully
Actual Result:
```
File "/workspace/train_dreambooth_lora_sdxl_advanced.py", line 2104, in <module>
main(args)
File "/workspace/... | closed | completed | false | 3 | [
"bug"
] | [
"sayakpaul",
"linoytsaban"
] | 2024-01-08T03:24:26Z | 2024-01-16T09:00:50Z | 2024-01-16T09:00:50Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | steverhoades | 1,146,668 | MDQ6VXNlcjExNDY2Njg= | User | false |
huggingface/diffusers | 2,070,372,471 | I_kwDOHa8MBc57Z2B3 | 6,486 | https://github.com/huggingface/diffusers/issues/6486 | https://api.github.com/repos/huggingface/diffusers/issues/6486 | dreambooth advanced: embedding retraction happening twice | Describe the bug
https://github.com/huggingface/diffusers/blob/774f5c45817805546ae5eb914c175d4fe72dcfe9/examples/advanced_diffusion_training/train_dreambooth_lora_sdxl_advanced.py#L1868
The embedding handler already takes care of accounting for both text encoders, no need to loop over this.
(You instantiate it wit... | closed | completed | false | 4 | [] | [
"linoytsaban"
] | 2024-01-08T12:36:57Z | 2024-01-17T14:44:32Z | 2024-01-17T14:43:27Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | rvorias | 21,060,408 | MDQ6VXNlcjIxMDYwNDA4 | User | false |
huggingface/diffusers | 2,071,006,022 | I_kwDOHa8MBc57cQtG | 6,492 | https://github.com/huggingface/diffusers/issues/6492 | https://api.github.com/repos/huggingface/diffusers/issues/6492 | Image Generation resulting in black images | ### Describe the bug
I've trained a unconditional model with a set of images and when sampling I am getting a black image. I've also tried setting the torch_dtype to float32 when sampling, but it didn't work
### Reproduction
**Training script:**
accelerate launch train_unconditional.py \
--dataset_name="MatheusH... | closed | completed | false | 4 | [
"bug",
"training"
] | [] | 2024-01-08T18:29:13Z | 2024-02-26T15:04:48Z | 2024-02-26T15:04:48Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | MatheusHiro | 63,820,524 | MDQ6VXNlcjYzODIwNTI0 | User | false |
huggingface/diffusers | 2,071,108,606 | I_kwDOHa8MBc57cpv- | 6,493 | https://github.com/huggingface/diffusers/issues/6493 | https://api.github.com/repos/huggingface/diffusers/issues/6493 | [Pipeline] FreeNoise | ### Model/Pipeline/Scheduler description
Highlights from the paper:
- Can generate longer videos while maintaining content and temporal consistency.
- Current models support single text condition for entire duration of video. FreeNoise allows support for multiple prompt conditions, as video content changes over ti... | closed | completed | false | 2 | [] | [] | 2024-01-08T19:42:40Z | 2024-01-09T04:52:25Z | 2024-01-09T02:43:13Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | a-r-r-o-w | 72,266,394 | MDQ6VXNlcjcyMjY2Mzk0 | User | false |
huggingface/diffusers | 2,071,134,612 | I_kwDOHa8MBc57cwGU | 6,494 | https://github.com/huggingface/diffusers/issues/6494 | https://api.github.com/repos/huggingface/diffusers/issues/6494 | [Pipeline] PIA (Personalized Image Animator) | In #6328, me and @jon-chuang have been looking into adding support for Image-to-Video with AnimateDiff. What we have, at the moment, is a hacky idea that has been used in webui implementations. The results obtained aren't that great but somewhat decent. In the search for improving quality, I came across PIA which has f... | closed | completed | false | 5 | [
"stale"
] | [] | 2024-01-08T20:02:25Z | 2024-02-08T17:03:04Z | 2024-02-08T17:03:04Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | a-r-r-o-w | 72,266,394 | MDQ6VXNlcjcyMjY2Mzk0 | User | false |
huggingface/diffusers | 2,071,149,233 | I_kwDOHa8MBc57czqx | 6,495 | https://github.com/huggingface/diffusers/issues/6495 | https://api.github.com/repos/huggingface/diffusers/issues/6495 | Possible bug in pipelines that use denoising_start and denoising_end | Here are a few examples of what I think is a possible bug.
https://github.com/huggingface/diffusers/blob/6ae7e8112a49ac20ade10ac639a76edbd9817a51/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py#L1277
https://github.com/huggingface/diffusers/blob/6ae7e8112a49ac20ade10ac639a76ed... | closed | completed | false | 4 | [
"contributions-welcome"
] | [] | 2024-01-08T20:13:21Z | 2024-02-25T09:39:23Z | 2024-02-25T09:39:23Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | a-r-r-o-w | 72,266,394 | MDQ6VXNlcjcyMjY2Mzk0 | User | false |
huggingface/diffusers | 2,071,247,612 | I_kwDOHa8MBc57dLr8 | 6,496 | https://github.com/huggingface/diffusers/issues/6496 | https://api.github.com/repos/huggingface/diffusers/issues/6496 | Multiple People / Faces / Instances example in Dreambooth training | **Is your feature request related to a problem? Please describe.**
The dog example that you guys in under the dreambooth xl example is great, however, there needs a little more documentation on different scenarios, such as training an SDXL model to learn two faces and be able to generate images of couples.
**Des... | closed | completed | false | 1 | [
"stale"
] | [] | 2024-01-08T21:24:08Z | 2024-02-16T15:05:18Z | 2024-02-16T15:05:18Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | naseerfaheem | 38,170,316 | MDQ6VXNlcjM4MTcwMzE2 | User | false |
huggingface/diffusers | 2,071,283,056 | I_kwDOHa8MBc57dUVw | 6,497 | https://github.com/huggingface/diffusers/issues/6497 | https://api.github.com/repos/huggingface/diffusers/issues/6497 | [Model + Pipeline] DragNUWA | ### Model/Pipeline/Scheduler description
> DragNUWA enables users to manipulate backgrounds or objects within images directly, and the model seamlessly translates these actions into camera movements or object motions, generating the corresponding video.
![example](https://github.com/huggingface/diffusers/assets/7... | closed | completed | false | 9 | [
"community-examples",
"stale",
"contributions-welcome"
] | [] | 2024-01-08T21:52:56Z | 2024-08-30T11:23:05Z | 2024-08-30T11:23:05Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | a-r-r-o-w | 72,266,394 | MDQ6VXNlcjcyMjY2Mzk0 | User | false |
huggingface/diffusers | 2,072,015,269 | I_kwDOHa8MBc57gHGl | 6,500 | https://github.com/huggingface/diffusers/issues/6500 | https://api.github.com/repos/huggingface/diffusers/issues/6500 | 8-bit / 4-bit in `diffusers`? | ## TL;DR
This issue is about gathering interest for supporting 8-bit and 4-bit precision operations (via `bitsandbytes`) in `diffusers` like `transformers`. It does so by showing some preliminary code and discussing the gotchas we should be aware of. It's recommended to read the entirety of the description first, an... | closed | not_planned | false | 10 | [] | [] | 2024-01-09T09:52:33Z | 2024-05-14T08:11:23Z | 2024-01-11T03:41:09Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sayakpaul | 22,957,388 | MDQ6VXNlcjIyOTU3Mzg4 | User | false |
huggingface/diffusers | 2,072,073,776 | I_kwDOHa8MBc57gVYw | 6,501 | https://github.com/huggingface/diffusers/issues/6501 | https://api.github.com/repos/huggingface/diffusers/issues/6501 | RuntimeError: Expected a 'cuda' device type for generator but found 'cpu' | I am new to using the diffusers library and was following the tutorial [Understanding pipelines, models and schedulers](https://github.com/huggingface/diffusers/blob/aa1797e1096df707b8e5164b72dc1aa5c202d0d8/docs/source/en/using-diffusers/write_own_pipeline.md?plain=1#L13). I am getting an error [here](https://github.co... | closed | completed | false | 6 | [
"stale"
] | [] | 2024-01-09T10:24:22Z | 2024-02-09T05:35:20Z | 2024-02-09T05:35:20Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patient74 | 83,630,710 | MDQ6VXNlcjgzNjMwNzEw | User | false |
huggingface/diffusers | 2,072,125,297 | I_kwDOHa8MBc57gh9x | 6,502 | https://github.com/huggingface/diffusers/issues/6502 | https://api.github.com/repos/huggingface/diffusers/issues/6502 | question about training diffusion-inpainting model | Hi, everyone! I'm struggling with the inpainting/outpainting, and quite confused about the input of the model in the training stage, hope can get some help😢
In **diffusers/examples/research_projects/multi_subject_dreambooth_inpainting/train_multi_subject_dreambooth_inpainting.py** @gzguevara and **diffusers/examp... | open | null | false | 5 | [
"stale"
] | [] | 2024-01-09T10:53:27Z | 2026-05-22T07:10:53Z | null | NONE | null | 20260522T073512Z | 2026-05-22T07:35:12Z | D222097 | 50,061,868 | MDQ6VXNlcjUwMDYxODY4 | User | false |
huggingface/diffusers | 2,072,530,329 | I_kwDOHa8MBc57iE2Z | 6,503 | https://github.com/huggingface/diffusers/issues/6503 | https://api.github.com/repos/huggingface/diffusers/issues/6503 | [Training] Make officially maintained training scripts compatible with `torch.compile()` | We recently made the DreamBooth LoRA SDXL training script fully compatible with `torch.compile()` in the following PRs: https://github.com/huggingface/diffusers/pull/6483, https://github.com/huggingface/diffusers/pull/6511.
Since there might be speedup benefits from `torch.compile()`, it makes sense to propagate th... | closed | completed | false | 9 | [
"Good second issue",
"contributions-welcome"
] | [] | 2024-01-09T14:51:00Z | 2024-01-15T14:14:55Z | 2024-01-15T14:13:27Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sayakpaul | 22,957,388 | MDQ6VXNlcjIyOTU3Mzg4 | User | false |
huggingface/diffusers | 2,072,707,637 | I_kwDOHa8MBc57iwI1 | 6,505 | https://github.com/huggingface/diffusers/issues/6505 | https://api.github.com/repos/huggingface/diffusers/issues/6505 | [consistency distillation] LoRA scripts omits EMA update | ### Describe the bug
Consistency distillation scripts omit EMA update of teacher unet
I think that this will limit the quality in the training limit, as the teacher will not be updated to produce better one-step denoising.
I think the idea of EMA is that it can mimic something like progressive distillation, wh... | closed | completed | false | 7 | [
"bug",
"stale"
] | [] | 2024-01-09T16:19:25Z | 2024-02-17T15:04:20Z | 2024-02-17T15:04:20Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | jon-chuang | 9,093,549 | MDQ6VXNlcjkwOTM1NDk= | User | false |
huggingface/diffusers | 2,072,719,186 | I_kwDOHa8MBc57iy9S | 6,506 | https://github.com/huggingface/diffusers/issues/6506 | https://api.github.com/repos/huggingface/diffusers/issues/6506 | StableVideoDiffusionPipeline returns a list instead of np.ndarray for output_type="np" | ### Describe the bug
The [comments](https://github.com/huggingface/diffusers/blob/4497b3ec982978eca99895ed1429addde4a84fff/src/diffusers/pipelines/stable_video_diffusion/pipeline_stable_video_diffusion.py#L64) for StableVideoDiffusionPipelineOutput (which is returned by the `__call__()` function of StableVideoDiffusio... | closed | completed | false | 4 | [
"bug"
] | [] | 2024-01-09T16:24:55Z | 2024-01-23T09:14:00Z | 2024-01-23T09:13:59Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | yondonfu | 5,933,273 | MDQ6VXNlcjU5MzMyNzM= | User | false |
huggingface/diffusers | 2,073,502,604 | I_kwDOHa8MBc57lyOM | 6,510 | https://github.com/huggingface/diffusers/issues/6510 | https://api.github.com/repos/huggingface/diffusers/issues/6510 | [PEFT] overriding an adapter | I was investigating a fix for https://github.com/huggingface/diffusers/issues/6442. All issues: https://github.com/huggingface/diffusers/issues?q=is%3Aissue+is%3Aopen+ValueError%3A+Attempting+to+unscale+FP16+gradients.+.
While resuming training from a checkpoint, we do: https://github.com/huggingface/diffusers/blob... | closed | completed | false | 4 | [] | [] | 2024-01-10T03:09:18Z | 2024-01-10T08:03:51Z | 2024-01-10T08:00:41Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sayakpaul | 22,957,388 | MDQ6VXNlcjIyOTU3Mzg4 | User | false |
huggingface/diffusers | 2,073,523,012 | I_kwDOHa8MBc57l3NE | 6,512 | https://github.com/huggingface/diffusers/issues/6512 | https://api.github.com/repos/huggingface/diffusers/issues/6512 | convert svd safetensors | How to convert svd safetensors into diffusers weights **locally**(not downloading from Hub)? Is convert_svd_to_diffusers.py a complete script ? It seems that this script only contain unet&vae convert, I cannot find how to change image_encoder in this script.
Thanks in advance for any answers! | closed | completed | false | 2 | [] | [] | 2024-01-10T03:38:08Z | 2024-01-11T13:46:47Z | 2024-01-11T13:46:47Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | shliu0 | 126,319,446 | U_kgDOB4d7Vg | User | false |
huggingface/diffusers | 2,073,652,358 | I_kwDOHa8MBc57mWyG | 6,513 | https://github.com/huggingface/diffusers/issues/6513 | https://api.github.com/repos/huggingface/diffusers/issues/6513 | Lacking support for t2iadapter_style_sd14v1 model in T2IAdapter | **Is your feature request related to a problem? Please describe.**
I was following [instructions for T2IAdapter](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/adapter) and trying to load the t2iadapter_style_sd14v1 adapter (Huggingface Hub [link](https://huggingface.co/TencentARC/T2I-Adapter/blo... | open | null | false | 6 | [
"stale"
] | [
"yiyixuxu"
] | 2024-01-10T06:02:11Z | 2024-09-14T15:20:23Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Firenze11 | 4,994,677 | MDQ6VXNlcjQ5OTQ2Nzc= | User | false |
huggingface/diffusers | 2,074,014,249 | I_kwDOHa8MBc57nvIp | 6,515 | https://github.com/huggingface/diffusers/issues/6515 | https://api.github.com/repos/huggingface/diffusers/issues/6515 | [Tests] fast tests single file checkpoints | Even though we support loading single file checkpoints via `from_single_file()`, we only run SLOW tests for them. It's better if we had a fast testing suite for them too.
@DN6 would you like to give this a try while you're at refactoring the single file loaders? | open | null | false | 11 | [
"stale",
"single_file"
] | [
"sayakpaul"
] | 2024-01-10T10:02:56Z | 2025-05-08T03:28:14Z | null | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sayakpaul | 22,957,388 | MDQ6VXNlcjIyOTU3Mzg4 | User | false |
huggingface/diffusers | 2,074,061,826 | I_kwDOHa8MBc57n6wC | 6,516 | https://github.com/huggingface/diffusers/issues/6516 | https://api.github.com/repos/huggingface/diffusers/issues/6516 | IP-adapter controlnet img2img mat1 and mat2 shapes cannot be multiplied (2x1024 and 1280x768) | Why isn't this line of code added here
output_hidden_state = False if isinstance(self.unet.encoder_hid_proj, ImageProjection) else True
https://github.com/huggingface/diffusers/blob/3e8b63216e502a0dd7424c0fef4134187802b93f/src/diffusers/pipelines/controlnet/pipeline_controlnet_img2img.py#L1089
| closed | completed | false | 4 | [
"stale"
] | [] | 2024-01-10T10:29:09Z | 2024-02-17T15:04:16Z | 2024-02-17T15:04:16Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Honey-666 | 76,274,942 | MDQ6VXNlcjc2Mjc0OTQy | User | false |
huggingface/diffusers | 2,074,321,927 | I_kwDOHa8MBc57o6QH | 6,517 | https://github.com/huggingface/diffusers/issues/6517 | https://api.github.com/repos/huggingface/diffusers/issues/6517 | Failed to import diffusers.pipelines.controlnet.pipeline_controlnet_sd_xl_img2img because of the following error (look up to see its traceback): | ### Describe the bug
Failed to import diffusers.pipelines.controlnet.pipeline_controlnet_sd_xl_img2img because of the following error (look up to see its traceback):
### Reproduction
from diffusers import AutoPipelineForText2Image
### Logs
```shell
'Traceback (most recent call last):
File "/usr/local/lib/python... | closed | completed | false | 2 | [
"bug"
] | [] | 2024-01-10T12:56:48Z | 2025-05-26T07:44:33Z | 2024-01-23T16:40:22Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | nihilisticneuralnet | 138,315,505 | U_kgDOCD6G8Q | User | false |
huggingface/diffusers | 2,074,511,056 | I_kwDOHa8MBc57pobQ | 6,518 | https://github.com/huggingface/diffusers/issues/6518 | https://api.github.com/repos/huggingface/diffusers/issues/6518 | unsupported operand type(s) for // : 'NoneType' and 'int' when running diffusers/examples/consistency_distillation /train_lcm_distill_sd_wds.py | ### Describe the bug
reason:
The class of .config of Unet is FrozenDict, teacher_unet.config["time_cond_proj_dim"] = args.unet_time_cond_proj_dim(line 925) may not change teacher_unet.config.time_cond_proj_dim,and make w_embedding = guidance_scale_embedding(w, embedding_dim=time_cond_proj_dim)(line 1128) get wrong ... | closed | completed | false | 1 | [
"bug"
] | [] | 2024-01-10T14:35:23Z | 2024-01-11T02:51:08Z | 2024-01-11T02:51:08Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | dreaming-huang | 68,060,593 | MDQ6VXNlcjY4MDYwNTkz | User | false |
huggingface/diffusers | 2,074,922,435 | I_kwDOHa8MBc57rM3D | 6,520 | https://github.com/huggingface/diffusers/issues/6520 | https://api.github.com/repos/huggingface/diffusers/issues/6520 | Issue with `log_validation` function. Giving blank results while training adapter network using `train_t2i_adapter_sdxl.py` | ### Describe the bug
Getting this warning in the `log_validation` function
```
{'adapter'} was not found in config. Values will be initialized to default values.
```
Results are blank as well.
But if I am loading the same model using the inference script the results are not blank.
### Reproduction
Following... | closed | completed | false | 7 | [
"bug"
] | [] | 2024-01-10T18:17:31Z | 2024-01-11T08:07:29Z | 2024-01-11T08:07:29Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | square-1111 | 31,164,802 | MDQ6VXNlcjMxMTY0ODAy | User | false |
huggingface/diffusers | 2,075,072,536 | I_kwDOHa8MBc57rxgY | 6,522 | https://github.com/huggingface/diffusers/issues/6522 | https://api.github.com/repos/huggingface/diffusers/issues/6522 | Question about UNet1DConditionModel | I am working on processing 1-D data with conditions using diffusion, and I am using Hugging Face models to accelerate my work. I am using UNet2DConditionModel, but the kernels are all in the shape of (3, 3), and I need to pad my 1-D data with padding=1, which means the upper (1, 3) and the lower (1, 3) in the kernels ... | closed | completed | false | 3 | [] | [] | 2024-01-10T19:45:26Z | 2024-01-11T20:12:16Z | 2024-01-11T20:12:16Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | boqian-li | 88,582,462 | MDQ6VXNlcjg4NTgyNDYy | User | false |
huggingface/diffusers | 2,075,732,801 | I_kwDOHa8MBc57uStB | 6,524 | https://github.com/huggingface/diffusers/issues/6524 | https://api.github.com/repos/huggingface/diffusers/issues/6524 | When lora is loaded a second time,lora does not work | ### Describe the bug
When loRA is loaded a second time,lora does not work
### Reproduction
```
import cv2
from PIL import Image
from insightface.app import FaceAnalysis
import torch
from diffusers import StableDiffusionPipeline, DDIMScheduler, AutoencoderKL
import os
def main():
source_dir = r''
... | closed | completed | false | 5 | [
"bug"
] | [] | 2024-01-11T04:41:24Z | 2024-01-18T01:45:20Z | 2024-01-18T01:45:20Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Daming-TF | 69,393,406 | MDQ6VXNlcjY5MzkzNDA2 | User | false |
huggingface/diffusers | 2,076,068,023 | I_kwDOHa8MBc57vki3 | 6,527 | https://github.com/huggingface/diffusers/issues/6527 | https://api.github.com/repos/huggingface/diffusers/issues/6527 | Bug in EulerDiscreteScheduler when calling "add_noise" function | ### Describe the bug
I am trying to use the EulerDiscreteScheduler to train a diffusion model (finetune the Stable Video Diffusion pipeline with the default scheduler), and I got the error message ```RuntimeError: a Tensor with 0 elements cannot be converted to Scalar```.
### Reproduction
A simlified tr... | closed | completed | false | 8 | [
"bug",
"stale",
"scheduler"
] | [
"yiyixuxu"
] | 2024-01-11T08:47:29Z | 2024-02-11T07:25:48Z | 2024-02-11T07:25:48Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | SystemErrorWang | 26,077,659 | MDQ6VXNlcjI2MDc3NjU5 | User | false |
huggingface/diffusers | 2,076,208,868 | I_kwDOHa8MBc57wG7k | 6,529 | https://github.com/huggingface/diffusers/issues/6529 | https://api.github.com/repos/huggingface/diffusers/issues/6529 | SDXL Fooocus Inpaint | **Is your feature request related to a problem? Please describe.**
I have seen that diffusers StableDiffusionXLInpaintPipeline generates worse results than SD 1.5 pipeline.
**Describe the solution you'd like.**
Include Fooocus inpaint patch, you could specify with a new loader.
Weights are available right now in ... | closed | completed | true | 55 | [
"help wanted",
"stale",
"contributions-welcome",
"inpainting"
] | [] | 2024-01-11T10:06:01Z | 2024-11-06T00:34:18Z | 2024-11-06T00:34:18Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | WaterKnight1998 | 41,203,448 | MDQ6VXNlcjQxMjAzNDQ4 | User | false |
huggingface/diffusers | 2,076,459,582 | I_kwDOHa8MBc57xEI- | 6,530 | https://github.com/huggingface/diffusers/issues/6530 | https://api.github.com/repos/huggingface/diffusers/issues/6530 | SD XL K-diffusion with Controlnet | ### Model/Pipeline/Scheduler description
Hi everyone, been following #6295 and #6477 for some time now to generate images with mentioned schedulers to reproduce Automatic11 quality. Is there a way to use the new SD XL K-diffusion pipeline with controlnet or Ip adapter? or use another pipeline to generate high qualit... | closed | completed | false | 4 | [] | [] | 2024-01-11T12:09:46Z | 2024-03-22T04:16:55Z | 2024-01-15T06:32:46Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | aycaecemgul | 50,622,243 | MDQ6VXNlcjUwNjIyMjQz | User | false |
huggingface/diffusers | 2,077,092,765 | I_kwDOHa8MBc57zeud | 6,531 | https://github.com/huggingface/diffusers/issues/6531 | https://api.github.com/repos/huggingface/diffusers/issues/6531 | [Design] better design to switch pipelines for already loaded pipeline | having separate pipelines in diffusers for features is somewhat cumbersome, especially since pipeline inheritance is less than ideal (e.g. why doesn't StableDiffusionImage2ImagePipeline inherit from StableDiffusionPipeline so I cannot check current model type easily?) and AutoPipeline is does not have full coverage and... | closed | completed | false | 22 | [
"stale"
] | [
"yiyixuxu"
] | 2024-01-11T16:23:52Z | 2024-05-01T07:59:29Z | 2024-04-26T19:12:22Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sayakpaul | 22,957,388 | MDQ6VXNlcjIyOTU3Mzg4 | User | false |
huggingface/diffusers | 2,077,153,337 | I_kwDOHa8MBc57ztg5 | 6,533 | https://github.com/huggingface/diffusers/issues/6533 | https://api.github.com/repos/huggingface/diffusers/issues/6533 | Marigold - monocular depth estimation pipeline | ### Model/Pipeline/Scheduler description
Marigold depth is the current diffusion-based state-of-the-art monocular depth estimation pipeline (image-to-depth). It is derived from Stable Diffusion and fine-tuned with synthetic data. Marigold can zero-shot transfer to unseen data, offering usable results under the most ch... | closed | completed | false | 14 | [
"stale"
] | [] | 2024-01-11T16:50:41Z | 2024-06-13T19:58:32Z | 2024-06-13T19:58:32Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | toshas | 4,390,695 | MDQ6VXNlcjQzOTA2OTU= | User | false |
huggingface/diffusers | 2,077,224,971 | I_kwDOHa8MBc57z_AL | 6,535 | https://github.com/huggingface/diffusers/issues/6535 | https://api.github.com/repos/huggingface/diffusers/issues/6535 | [SVD] Is it possible to change the noise scheduler in StableVideoDiffusionPipeline? | **What API design would you like to have changed or added to the library? Why?**
I tried to replace the SVD's scheduler with DDIM (default is EULER) but found that it could not output normal results. Is this normal? Because the original paper stated that the results were obtained using DDIM.
<img width="1028" alt="... | closed | completed | false | 13 | [] | [
"yiyixuxu"
] | 2024-01-11T17:24:58Z | 2024-08-28T14:59:28Z | 2024-01-19T04:50:28Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | JulianJuaner | 37,148,692 | MDQ6VXNlcjM3MTQ4Njky | User | false |
huggingface/diffusers | 2,077,324,703 | I_kwDOHa8MBc570XWf | 6,536 | https://github.com/huggingface/diffusers/issues/6536 | https://api.github.com/repos/huggingface/diffusers/issues/6536 | Fine-tuning with stable diffusion with UNet3DConditionModel out-of-memory | ### Describe the bug
I tried to train SD based on the train_text_to_image.py, but I replaced UNet2D with UNet3D. During this training, I used 8 GPUs (i.e., 8 tesla v100) to train the model with 1 batch size. However, the job is failed with the error "out of memory". So, do you have any suggestions for the optimizati... | closed | completed | false | 1 | [
"bug"
] | [] | 2024-01-11T18:13:26Z | 2024-01-12T17:00:00Z | 2024-01-12T17:00:00Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | dahui-y | 93,740,259 | U_kgDOBZZc4w | User | false |
huggingface/diffusers | 2,077,935,962 | I_kwDOHa8MBc572sla | 6,540 | https://github.com/huggingface/diffusers/issues/6540 | https://api.github.com/repos/huggingface/diffusers/issues/6540 | "mat2 must be a matrix" error when finetuning controlnet with FSDP | ### Describe the bug
I tried to finetune controlnet with train_controlnet_sdxl.py and FSDP. But when started to validation, I got below exception.
File "train_controlnet_sdxl.py", line 1229, in main
image_logs = log_validation(
File "train_controlnet_sdxl.py", line 116, in log_validation
image = pipe... | closed | completed | false | 13 | [
"bug",
"training"
] | [] | 2024-01-12T02:08:45Z | 2025-08-18T07:08:14Z | 2024-02-26T15:04:44Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | JUN20240101 | 156,387,255 | U_kgDOCVJHtw | User | false |
huggingface/diffusers | 2,078,091,162 | I_kwDOHa8MBc573Sea | 6,545 | https://github.com/huggingface/diffusers/issues/6545 | https://api.github.com/repos/huggingface/diffusers/issues/6545 | [Tracker] modularize inferencing during and after training in the example scripts | We provide support for running validation inference during and after training in our officially maintained training examples. This is very helpful to keep track of the training progress.
We could modularize some bits in the example to reduce the LoC.
The `train_lcm_distill_lora_sdxl.py` [script](https://github.... | closed | completed | false | 9 | [
"Good second issue",
"contributions-welcome"
] | [] | 2024-01-12T05:19:48Z | 2025-03-12T15:50:32Z | 2025-03-12T15:50:32Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sayakpaul | 22,957,388 | MDQ6VXNlcjIyOTU3Mzg4 | User | false |
huggingface/diffusers | 2,078,377,882 | I_kwDOHa8MBc574Yea | 6,546 | https://github.com/huggingface/diffusers/issues/6546 | https://api.github.com/repos/huggingface/diffusers/issues/6546 | PALP: Prompt Aligned Personalization of Text-to-Image Models 🖌️🔥 | ### Model/Pipeline/Scheduler description
**PALP** (by Arar et al.) improves on existing personalization techniques, through better prompt alignment - allowing "rich and complex scene generation, including all elements of a condition prompt" - by using an additional score to align with the target prompt
![Screenshot 2... | closed | completed | false | 10 | [
"community-examples",
"contributions-welcome"
] | [] | 2024-01-12T09:00:41Z | 2025-07-25T02:37:57Z | 2024-02-20T15:04:32Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | linoytsaban | 57,615,435 | MDQ6VXNlcjU3NjE1NDM1 | User | false |
huggingface/diffusers | 2,078,381,686 | I_kwDOHa8MBc574ZZ2 | 6,548 | https://github.com/huggingface/diffusers/issues/6548 | https://api.github.com/repos/huggingface/diffusers/issues/6548 | Function `prepare_latents` mistreated image as latents in stable diffusion img2img pipeline (as well as StableDiffusionXLImg2ImgPipeline), when input PIL.Image.Image has 4 channels 'RGBA' | ### Describe the bug
In function
```
def prepare_latents(
self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None, add_noise=True
):
...
```
there is an if-else statement trying to determine whether input image is a 4-channel latent.
```
if image.shape[1] == 4:... | closed | completed | false | 4 | [
"bug"
] | [] | 2024-01-12T09:02:56Z | 2024-01-17T11:14:17Z | 2024-01-17T11:14:17Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | luocfprime | 79,003,314 | MDQ6VXNlcjc5MDAzMzE0 | User | false |
huggingface/diffusers | 2,078,403,427 | I_kwDOHa8MBc574etj | 6,549 | https://github.com/huggingface/diffusers/issues/6549 | https://api.github.com/repos/huggingface/diffusers/issues/6549 | IP-Adapter + Lora , lora_weight != 1.0 warning logs | https://github.com/huggingface/diffusers/blob/7d631825b0bfcc2ee3f80cb3f39a8eec6dfa1bc5/src/diffusers/models/attention_processor.py#L2238
hi
When I use lora + ip-adapter, lora_weight ! = 1.0 will print a lot of warning logs, can anyone tell me why?
### os:
diffusers==0.26.0.dev0
### this is my code:
`... | closed | completed | false | 9 | [
"stale"
] | [] | 2024-01-12T09:15:51Z | 2024-02-12T07:45:39Z | 2024-02-12T07:45:39Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Honey-666 | 76,274,942 | MDQ6VXNlcjc2Mjc0OTQy | User | false |
huggingface/diffusers | 2,078,643,431 | I_kwDOHa8MBc575ZTn | 6,552 | https://github.com/huggingface/diffusers/issues/6552 | https://api.github.com/repos/huggingface/diffusers/issues/6552 | [Tracker] fix training resuming problem when using FP16 in the examples | We have been a lot of issues on this topic:
```bash
File "/content/train_text_to_image_lora_sdxl.py", line 1261, in <module>
main(args)
File "/content/train_text_to_image_lora_sdxl.py", line 1077, in main
accelerator.clip_grad_norm_(params_to_optimize, args.max_grad_norm)
File "/usr/local/lib/python3.10/... | closed | completed | false | 14 | [
"Good second issue",
"contributions-welcome"
] | [] | 2024-01-12T11:36:59Z | 2025-05-10T05:26:07Z | 2025-05-10T05:26:04Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sayakpaul | 22,957,388 | MDQ6VXNlcjIyOTU3Mzg4 | User | false |
huggingface/diffusers | 1,388,132,020 | I_kwDOHa8MBc5SvTa0 | 656 | https://github.com/huggingface/diffusers/issues/656 | https://api.github.com/repos/huggingface/diffusers/issues/656 | Error for the 16gb gpu dreambooth training | ### Describe the bug
I can train without the prior preservation fine if I try to train with it I get
“UNet2DConditionModel' object has no attribute 'enable_gradient_checkpointing'”
I also get an error that prior_loss_weight is a unrecognized argument
### Reproduction
_No response_
### Logs
_No response_
... | closed | completed | false | 6 | [
"stale"
] | [
"patil-suraj"
] | 2022-09-27T17:40:33Z | 2022-11-05T15:02:49Z | 2022-11-05T15:02:49Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | UglyStupidHonest | 68,284,461 | MDQ6VXNlcjY4Mjg0NDYx | User | false |
huggingface/diffusers | 2,080,089,009 | I_kwDOHa8MBc57-6Ox | 6,560 | https://github.com/huggingface/diffusers/issues/6560 | https://api.github.com/repos/huggingface/diffusers/issues/6560 | Using LoRA's results in a shape size error? regression? | ### Describe the bug
I tried
self.pipeline.load_lora_weights(lora_path,low_cpu_mem_usage=False, ignore_mismatched_sizes=True)
but it doesn't seem to take.
runtimeError: Error(s) in loading state_dict for LoRALinearLayer:
size mismatch for down.weight: copying a param with shape torch.Size([128, 3... | closed | completed | false | 4 | [
"bug",
"needs-code-example"
] | [] | 2024-01-13T05:13:05Z | 2024-02-26T15:19:35Z | 2024-02-26T15:04:39Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | CrossPr0duct | 151,576,163 | U_kgDOCQjeYw | User | false |
huggingface/diffusers | 2,080,163,521 | I_kwDOHa8MBc57_MbB | 6,561 | https://github.com/huggingface/diffusers/issues/6561 | https://api.github.com/repos/huggingface/diffusers/issues/6561 | OSError: h94/IP-Adapter does not appear to have a file named models\image_encoder\config.json | ### Describe the bug
I'm getting this error:
`OSError: h94/IP-Adapter does not appear to have a file named models\image_encoder\config.json. Checkout 'https://huggingface.co/h94/IP-Adapter/main' for available files.`
When running the IP-Adapter example from here: https://huggingface.co/docs/diffusers/main/en/u... | closed | completed | false | 1 | [
"bug"
] | [] | 2024-01-13T07:46:47Z | 2025-05-27T01:38:30Z | 2024-01-16T05:55:24Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | tin2tin | 1,322,593 | MDQ6VXNlcjEzMjI1OTM= | User | false |
huggingface/diffusers | 2,080,228,929 | I_kwDOHa8MBc57_cZB | 6,563 | https://github.com/huggingface/diffusers/issues/6563 | https://api.github.com/repos/huggingface/diffusers/issues/6563 | Making custom component <> custom file mapping really optional. | **What API design would you like to have changed or added to the library? Why?**
I have a custom pipeline using a custom component class defined locally, for example, a custom `UNet` in `mylib/unet.py`. The exported `model_index.json` will describe it as:
```json
"unet": [
"mylib.unet",
"myUNet"
],
```... | closed | completed | false | 8 | [
"stale"
] | [
"sayakpaul"
] | 2024-01-13T10:25:42Z | 2024-02-21T15:05:16Z | 2024-02-21T15:05:16Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ashawkey | 25,863,658 | MDQ6VXNlcjI1ODYzNjU4 | User | false |
huggingface/diffusers | 2,080,410,436 | I_kwDOHa8MBc58AItE | 6,565 | https://github.com/huggingface/diffusers/issues/6565 | https://api.github.com/repos/huggingface/diffusers/issues/6565 | Problems on DPO Diffusion Training. Loss spikes and noised samples generation. | ### Describe the bug
Using the script located in
diffusers/examples/research_projects/diffusion_dpo to try a DPO on my Stable DIffusion XL model I've seen a few bugs that vary a bit in the amount of steps between one training and another, but are always present. I ran more than 3 tests.
Generally, the training ru... | closed | completed | false | 11 | [
"bug",
"stale"
] | [
"sayakpaul"
] | 2024-01-13T17:44:10Z | 2024-02-21T15:05:14Z | 2024-02-21T15:05:14Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | artificialguybr | 83,289,567 | MDQ6VXNlcjgzMjg5NTY3 | User | false |
huggingface/diffusers | 2,080,488,923 | I_kwDOHa8MBc58Ab3b | 6,567 | https://github.com/huggingface/diffusers/issues/6567 | https://api.github.com/repos/huggingface/diffusers/issues/6567 | Importing SD 1.5: `text_config_dict` is provided which will be used to initialize `CLIPTextConfig`. | ### Describe the bug
I'm getting this error/debug info when I import the runway SD 1.5 checkpoint. Is it a bug or something I can do something about in my end?
```
`text_config_dict` is provided which will be used to initialize `CLIPTextConfig`. The value `text_config["id2label"]` will be overriden.
`text_config... | closed | completed | false | 5 | [
"bug",
"stale"
] | [] | 2024-01-13T21:56:12Z | 2024-02-21T15:05:12Z | 2024-02-21T15:05:12Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | tin2tin | 1,322,593 | MDQ6VXNlcjEzMjI1OTM= | User | false |
huggingface/diffusers | 1,388,194,235 | I_kwDOHa8MBc5Svim7 | 657 | https://github.com/huggingface/diffusers/issues/657 | https://api.github.com/repos/huggingface/diffusers/issues/657 | Handle `kwargs` and not serializable values as configuration attributes | Context: #655.
A `kwargs` argument was added to schedulers in https://github.com/huggingface/diffusers/commit/d7dcba4a130a2eeecd46fc1d2ed244f54fd2a8f6 just to display a deprecation warning. `kwargs` was interpreted as a configurable attribute, but it could not be serialized. This broke:
- Pipeline loading, via `__r... | closed | completed | false | 3 | [
"stale"
] | [] | 2022-09-27T18:31:43Z | 2022-10-30T15:33:43Z | 2022-10-30T15:33:42Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | pcuenca | 1,177,582 | MDQ6VXNlcjExNzc1ODI= | User | false |
huggingface/diffusers | 2,080,779,400 | I_kwDOHa8MBc58BiyI | 6,570 | https://github.com/huggingface/diffusers/issues/6570 | https://api.github.com/repos/huggingface/diffusers/issues/6570 | [StableDiffusionXLInstructPix2PixPipeline] RuntimeError: Sizes of tensors must match except in dimension 1 | ### Describe the bug
While working on #6569 related to #6545, I run the InstructPix2Pix SDXL training example [code](https://github.com/huggingface/diffusers/blob/main/examples/instruct_pix2pix/train_instruct_pix2pix_sdxl.py) and noticed this issue.
I think before merging #6569, we should fix this issue first.
... | closed | completed | false | 6 | [
"bug"
] | [] | 2024-01-14T15:43:38Z | 2024-01-16T02:20:35Z | 2024-01-16T02:20:35Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | kim-sangyeon | 55,663,336 | MDQ6VXNlcjU1NjYzMzM2 | User | false |
huggingface/diffusers | 2,081,182,379 | I_kwDOHa8MBc58DFKr | 6,572 | https://github.com/huggingface/diffusers/issues/6572 | https://api.github.com/repos/huggingface/diffusers/issues/6572 | ControlNetXS-SDXL-Inpaint-Pipeline | ### Model/Pipeline/Scheduler description
Will there be a pipeline for ControlNetXS-SDXL-Inpaint-Pipeline?
### Open source status
- [ ] The model implementation is available.
- [ ] The model weights are available (Only relevant if addition is not a scheduler).
### Provide useful links for the implementation
_No re... | closed | completed | false | 2 | [
"stale"
] | [] | 2024-01-15T03:32:18Z | 2024-02-23T15:04:35Z | 2024-02-23T15:04:35Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | xinli2008 | 71,311,087 | MDQ6VXNlcjcxMzExMDg3 | User | false |
huggingface/diffusers | 2,081,479,873 | I_kwDOHa8MBc58ENzB | 6,574 | https://github.com/huggingface/diffusers/issues/6574 | https://api.github.com/repos/huggingface/diffusers/issues/6574 | FutureWarning: expected range[0,1] tensor in SVD pipeline | ### Describe the bug
Why set this warning in [diffusers/image_processor.py:339:](https://github.com/huggingface/diffusers/blob/76c645d3a641c879384afcb43496f0b7db8cc5cb/src/diffusers/image_processor.py#L339C1-L343C14)
`FutureWarning: Passing `image` as torch tensor with value range in [-1,1] is deprecated. The expe... | closed | completed | false | 6 | [
"bug"
] | [
"DN6"
] | 2024-01-15T08:17:21Z | 2024-02-20T08:46:09Z | 2024-02-18T09:10:32Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | shliu0 | 126,319,446 | U_kgDOB4d7Vg | User | false |
huggingface/diffusers | 2,081,816,019 | I_kwDOHa8MBc58Ff3T | 6,575 | https://github.com/huggingface/diffusers/issues/6575 | https://api.github.com/repos/huggingface/diffusers/issues/6575 | Missing methods of `AutoencoderKL` in online docs | The [`decode`](https://github.com/huggingface/diffusers/blob/a080f0d3a21104c19effb4d9d95dcb710d99c19b/src/diffusers/models/autoencoders/autoencoder_kl.py#L281) method of `AutoencoderKL` is absent from the [online doc](https://huggingface.co/docs/diffusers/main/en/api/models/autoencoderkl#autoencoderkl); same for `encod... | open | reopened | false | 14 | [
"documentation",
"stale"
] | [] | 2024-01-15T11:15:46Z | 2024-12-15T15:07:34Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | thethomasboyer | 46,932,488 | MDQ6VXNlcjQ2OTMyNDg4 | User | false |
huggingface/diffusers | 2,082,524,953 | I_kwDOHa8MBc58IM8Z | 6,579 | https://github.com/huggingface/diffusers/issues/6579 | https://api.github.com/repos/huggingface/diffusers/issues/6579 | [Pipeline/Model] PhotoMaker | Highlights from the paper:
- Arbitrary number of identity images can be provided to create the "stacked ID embedding" - a representation of the input images that the model is trained to understand. Input images and prompt are encoded and combined to create the stacked embedding, which is fed into the cross attention l... | closed | completed | false | 16 | [
"stale"
] | [] | 2024-01-15T18:38:53Z | 2024-02-26T15:20:28Z | 2024-02-26T15:04:33Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | a-r-r-o-w | 72,266,394 | MDQ6VXNlcjcyMjY2Mzk0 | User | false |
huggingface/diffusers | 2,082,875,014 | I_kwDOHa8MBc58JiaG | 6,584 | https://github.com/huggingface/diffusers/issues/6584 | https://api.github.com/repos/huggingface/diffusers/issues/6584 | Unable to convert DreamShaper v8 Inpaint model → Diffusers | ### Describe the bug
Hi! I'm trying to convert dream shaper v8 inpaint model ([link](https://civitai.com/models/4384?modelVersionId=131004)) from SafeTensors to Diffusers using next script:
`python convert_original_stable_diffusion_to_diffusers.py --checkpoint_path dreamshaper_8Inpainting.safetensors --from_safeten... | closed | completed | false | 2 | [
"bug"
] | [] | 2024-01-16T01:30:52Z | 2024-01-20T10:46:01Z | 2024-01-20T10:46:00Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | nosferatu500 | 9,116,892 | MDQ6VXNlcjkxMTY4OTI= | User | false |
huggingface/diffusers | 2,082,883,131 | I_kwDOHa8MBc58JkY7 | 6,585 | https://github.com/huggingface/diffusers/issues/6585 | https://api.github.com/repos/huggingface/diffusers/issues/6585 | [Tracker] change to posix for better Windows support | In https://github.com/huggingface/diffusers/pull/6564, @fabiorigano introduced the use of Posix to better support Windows compatibility.
It'd be nice to change the instances of `os.path.join()` to `Path(...).as_posix()`.
Feel free to open PRs for this and tag me.
While opening PRs, please target only ONE sc... | closed | completed | false | 17 | [
"Good second issue",
"contributions-welcome"
] | [] | 2024-01-16T01:44:05Z | 2024-03-19T21:50:35Z | 2024-03-19T21:50:35Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sayakpaul | 22,957,388 | MDQ6VXNlcjIyOTU3Mzg4 | User | false |
huggingface/diffusers | 2,082,904,743 | I_kwDOHa8MBc58Jpqn | 6,586 | https://github.com/huggingface/diffusers/issues/6586 | https://api.github.com/repos/huggingface/diffusers/issues/6586 | [Tracker] use micro-conditioning for the SDXL trainers | SDXL makes use of micro-conditioning, and it does have quite a bit of an effect on the end results. For more details, refer to the paper [here](https://arxiv.org/abs/2307.01952).
Currently, not all of our SDXL trainers don't make use of micro-conditioning. So, it'd be nice to have micro-conditioning support as in h... | open | null | false | 12 | [
"stale",
"contributions-welcome",
"advanced"
] | [] | 2024-01-16T02:15:22Z | 2024-03-21T15:03:50Z | null | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sayakpaul | 22,957,388 | MDQ6VXNlcjIyOTU3Mzg4 | User | false |
huggingface/diffusers | 2,083,271,484 | I_kwDOHa8MBc58LDM8 | 6,593 | https://github.com/huggingface/diffusers/issues/6593 | https://api.github.com/repos/huggingface/diffusers/issues/6593 | Request: Let StableDiffusionXLImg2ImgPipeline use: load_lora_weights | ### Describe the bug
I don't know if this is a bug or a feature request, but I tried to load a LoRA into StableDiffusionXLImg2ImgPipeline, but it doesn't seem to work?
### Reproduction
```
import torch
from diffusers import StableDiffusionXLImg2ImgPipeline
from diffusers.utils import load_image
from diff... | closed | completed | false | 6 | [
"stale"
] | [] | 2024-01-16T08:17:14Z | 2024-03-14T02:07:52Z | 2024-03-13T15:17:41Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | tin2tin | 1,322,593 | MDQ6VXNlcjEzMjI1OTM= | User | false |
huggingface/diffusers | 2,083,503,131 | I_kwDOHa8MBc58L7wb | 6,594 | https://github.com/huggingface/diffusers/issues/6594 | https://api.github.com/repos/huggingface/diffusers/issues/6594 | requests.exceptions.ConnectionError: HTTPSConnectionPool(host='raw.githubusercontent.com', port=443): Max retries exceeded with url: /CompVis/stable-diffusion/main/configs/stable-diffusion/v1-inference.yaml | ### Describe the bug
Even I give all the weight and config_file, it still try to connect the internet
### Reproduction
pipe = diffusers.StableDiffusionPipeline.from_single_file(base_model,
torch_dtype=torch.float16,
... | closed | completed | false | 5 | [
"bug"
] | [] | 2024-01-16T09:57:33Z | 2024-02-18T09:09:58Z | 2024-02-18T09:09:58Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Louis24 | 18,719,360 | MDQ6VXNlcjE4NzE5MzYw | User | false |
huggingface/diffusers | 2,083,938,235 | I_kwDOHa8MBc58Nl-7 | 6,598 | https://github.com/huggingface/diffusers/issues/6598 | https://api.github.com/repos/huggingface/diffusers/issues/6598 | img2img, different results with same parameters, Diffusers vs A1111 | ### Describe the bug
In image to image feature, I found that different results with same parameters between Diffusers and A1111, and the results in Diffusers is not good.
**Key Parameters**
Use model: https://civitai.com/models/141122/raesanmix
Use Lora: https://civitai.com/models/110799/niji-19905
**The... | closed | completed | false | 5 | [
"bug",
"stale"
] | [] | 2024-01-16T13:11:51Z | 2024-03-06T07:29:32Z | 2024-02-23T15:04:24Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | zhndongyang | 6,086,729 | MDQ6VXNlcjYwODY3Mjk= | User | false |
huggingface/diffusers | 1,388,586,313 | I_kwDOHa8MBc5SxCVJ | 660 | https://github.com/huggingface/diffusers/issues/660 | https://api.github.com/repos/huggingface/diffusers/issues/660 | Running revision="fp16", torch_dtype=torch.float16 on mps M1 | ### Describe the bug
I'm using the following code:
```
!pip install diffusers
!pip install transformers scipy ftfy
pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", revision="fp16", torch_dtype=torch.float16, use_auth_token=True)
prompt = "Pineapple on a white table"
pipe.to("... | closed | completed | false | 0 | [
"bug"
] | [
"pcuenca"
] | 2022-09-28T01:24:14Z | 2022-10-29T19:09:34Z | 2022-10-29T19:09:34Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | MatthewWaller | 5,520,521 | MDQ6VXNlcjU1MjA1MjE= | User | false |
huggingface/diffusers | 2,084,490,320 | I_kwDOHa8MBc58PsxQ | 6,601 | https://github.com/huggingface/diffusers/issues/6601 | https://api.github.com/repos/huggingface/diffusers/issues/6601 | Better mechanism for getting values to attention processors | I think it would be nice to generally have some better infrastructure/mechanism for getting values to attention processors.
At the moment the main means of transportation seems to be cross_attention_kwargs but they are finally expanded as function arguments and have to match ALL function signatures that receive them... | closed | completed | false | 8 | [] | [
"yiyixuxu"
] | 2024-01-16T17:29:47Z | 2024-02-19T20:56:25Z | 2024-02-19T20:56:25Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | spezialspezial | 75,758,219 | MDQ6VXNlcjc1NzU4MjE5 | User | false |
huggingface/diffusers | 2,084,530,004 | I_kwDOHa8MBc58P2dU | 6,602 | https://github.com/huggingface/diffusers/issues/6602 | https://api.github.com/repos/huggingface/diffusers/issues/6602 | Loading pretrained model after `fuse_lora()` and `save_pretrained()` results in an error | ### Describe the bug
I am not able to cache a model to be re-loaded later after fusing a LoRA into it. My hope was that I could fuse the LoRA into the base model which would result in a new model that can be loaded as needed.
### Reproduction
```python
from diffusers import DiffusionPipeline
import torch
pipe_i... | closed | completed | false | 7 | [
"bug"
] | [] | 2024-01-16T17:58:07Z | 2024-01-31T07:15:42Z | 2024-01-31T07:15:41Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | bawgz | 19,415,566 | MDQ6VXNlcjE5NDE1NTY2 | User | false |
huggingface/diffusers | 2,084,531,404 | I_kwDOHa8MBc58P2zM | 6,603 | https://github.com/huggingface/diffusers/issues/6603 | https://api.github.com/repos/huggingface/diffusers/issues/6603 | Bug with AutoencoderKL | ### Describe the bug
Hello,
When attempting to generate images using Stable Video in ComfyUI, the process works correctly. However, when it transitions to "VAE Decode," it requests to load the "AutoencoderKL" model but encounters an issue and crashes.
imgtovid model : svd_xt
### Reproduction
Using directml with ... | closed | completed | false | 2 | [
"bug",
"stale"
] | [] | 2024-01-16T17:59:07Z | 2024-02-25T15:04:06Z | 2024-02-25T15:04:06Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | LeoGuibert1 | 90,767,469 | MDQ6VXNlcjkwNzY3NDY5 | User | false |
huggingface/diffusers | 2,085,171,904 | I_kwDOHa8MBc58STLA | 6,606 | https://github.com/huggingface/diffusers/issues/6606 | https://api.github.com/repos/huggingface/diffusers/issues/6606 | cross_attention_kwargs={"scale": 1} not work | cross_attention_kwargs value only goes down it doesn't go up
first I generate an image with cross_attention_kwargs={"scale": 1}
```
image = pipe(prompt=prompt,
negative_prompt="",
cross_attention_kwargs={"scale": 1} ,
num_inference_steps=25,
... | closed | completed | false | 5 | [
"stale"
] | [] | 2024-01-17T00:18:28Z | 2024-03-27T22:36:22Z | 2024-03-27T22:36:22Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | loboere | 15,265,895 | MDQ6VXNlcjE1MjY1ODk1 | User | false |
huggingface/diffusers | 2,085,171,994 | I_kwDOHa8MBc58STMa | 6,607 | https://github.com/huggingface/diffusers/issues/6607 | https://api.github.com/repos/huggingface/diffusers/issues/6607 | Clarification on noise scheduler mismatch in example train_controlnet.py | Hello,
In examples/controlnet/train_controlnet.py, the noise scheduler class is DDPMScheduler but however the scheduler config loaded based on the one in the model's scheduler folder i.e PNDM Scheduler per https://huggingface.co/runwayml/stable-diffusion-v1-5/blob/main/scheduler/scheduler_config.json . Why don't the... | closed | completed | false | 3 | [] | [] | 2024-01-17T00:18:35Z | 2024-01-17T18:04:32Z | 2024-01-17T18:04:32Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | vidyasiv | 81,709,031 | MDQ6VXNlcjgxNzA5MDMx | User | false |
huggingface/diffusers | 2,085,530,519 | I_kwDOHa8MBc58TquX | 6,610 | https://github.com/huggingface/diffusers/issues/6610 | https://api.github.com/repos/huggingface/diffusers/issues/6610 | num_in_channels has no effect when call StableDiffusionXLInpaintPipeline.from_single_file | ### Describe the bug
num_in_channels has no effect when call `StableDiffusionXLInpaintPipeline.from_single_file` and `StableDiffusionXLPipeline.from_single_file`
### Reproduction
- download normal sdxl model from here: https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/sd_xl_base_1.0.safetens... | closed | completed | false | 3 | [
"bug"
] | [
"DN6"
] | 2024-01-17T06:59:14Z | 2024-01-23T09:30:14Z | 2024-01-23T09:30:13Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Sanster | 3,998,421 | MDQ6VXNlcjM5OTg0MjE= | User | false |
huggingface/diffusers | 2,086,239,793 | I_kwDOHa8MBc58WX4x | 6,614 | https://github.com/huggingface/diffusers/issues/6614 | https://api.github.com/repos/huggingface/diffusers/issues/6614 | How to train text_to_image with images which is resolution of 512x768 ? | I want to finetune the sd1.5 with 50k images, all the image is resolution of 512x768. But I got error like this:
`train_text_to_image.py:` error: argument --resolution: invalid int value: '[512,768]'`
so, how to train text_to_image with images which is resolution of 512x768? | closed | completed | false | 2 | [] | [] | 2024-01-17T13:51:16Z | 2024-01-25T14:28:01Z | 2024-01-25T14:28:01Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | lingxuan630 | 617,207 | MDQ6VXNlcjYxNzIwNw== | User | false |
huggingface/diffusers | 2,086,255,402 | I_kwDOHa8MBc58Wbsq | 6,615 | https://github.com/huggingface/diffusers/issues/6615 | https://api.github.com/repos/huggingface/diffusers/issues/6615 | Pipeline.from_single_file/download_from_original_stable_diffusion_ckpt broken for local config yaml files | ### Describe the bug
When passing a local config yaml file path to Pipeline.from_single_file the file is not read in but handed over as a file path to yaml.safe_load(original_config_file). yaml expects the config contents and errors out. Something like this should be added:
```
diff --git a/src/diffusers/pipelines... | closed | completed | false | 5 | [
"bug"
] | [] | 2024-01-17T13:59:31Z | 2024-01-19T02:55:04Z | 2024-01-19T02:55:04Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | spezialspezial | 75,758,219 | MDQ6VXNlcjc1NzU4MjE5 | User | false |
huggingface/diffusers | 2,086,277,168 | I_kwDOHa8MBc58WhAw | 6,616 | https://github.com/huggingface/diffusers/issues/6616 | https://api.github.com/repos/huggingface/diffusers/issues/6616 | Implementation of Classifier-guidance in diffusers | ### Model/Pipeline/Scheduler description
Hello the community,
I was wondering if there's an implementation of classifier-guidance (from the paper "Diffusion Models Beat GANs on Image Synthesis") with stable diffusion in diffusers (or in any github repo, please let we know). Although I know it's not the most effi... | closed | completed | false | 4 | [
"stale"
] | [] | 2024-01-17T14:10:50Z | 2024-02-25T15:51:17Z | 2024-02-25T15:04:01Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | shengyenlin | 61,533,030 | MDQ6VXNlcjYxNTMzMDMw | User | false |
huggingface/diffusers | 1,388,719,233 | I_kwDOHa8MBc5SxiyB | 663 | https://github.com/huggingface/diffusers/issues/663 | https://api.github.com/repos/huggingface/diffusers/issues/663 | accelerate config | removed | closed | completed | false | 0 | [] | [] | 2022-09-28T04:44:19Z | 2022-09-28T04:44:39Z | 2022-09-28T04:44:39Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | JustAnOkapi | 39,418,801 | MDQ6VXNlcjM5NDE4ODAx | User | false |
huggingface/diffusers | 2,088,544,229 | I_kwDOHa8MBc58fKfl | 6,631 | https://github.com/huggingface/diffusers/issues/6631 | https://api.github.com/repos/huggingface/diffusers/issues/6631 | DPMMultiStepSolver does not reset model_outputs for generating new samples | Hi! First of all, thanks for your effort in implementing lots of papers and making them available for use!
I have been trying to use `DPMSolverMultistepScheduler` with point clouds, and I have noticed one potential bug. One thing about point clouds is that the number of points can change quite a lot, which is differ... | closed | completed | false | 4 | [
"stale"
] | [] | 2024-01-18T15:37:09Z | 2024-03-13T15:09:18Z | 2024-03-13T15:09:17Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | nuneslu | 6,524,770 | MDQ6VXNlcjY1MjQ3NzA= | User | false |
huggingface/diffusers | 2,088,584,716 | I_kwDOHa8MBc58fUYM | 6,632 | https://github.com/huggingface/diffusers/issues/6632 | https://api.github.com/repos/huggingface/diffusers/issues/6632 | is save_state necessary for lora training scripts? | ### Describe the bug
For examples/text_to_image/train_text_to_image_lora.py and examples/text_to_image/train_text_to_image_lora_sdxl.py:
During training checkpointing accelerator.save_state(save_path), is used to save a full pytorch_model.bin (could be a couple GBs),
Whereas after training is finished, just pipeline... | closed | completed | false | 2 | [
"bug"
] | [] | 2024-01-18T15:56:09Z | 2024-01-18T18:10:52Z | 2024-01-18T18:10:51Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | philhatgames | 150,477,807 | U_kgDOCPgb7w | User | false |
huggingface/diffusers | 2,088,614,544 | I_kwDOHa8MBc58fbqQ | 6,634 | https://github.com/huggingface/diffusers/issues/6634 | https://api.github.com/repos/huggingface/diffusers/issues/6634 | `use_clipped_model_output` in `DDIMInverseScheduler` | ### Describe the bug
The docstring of [`DDIMInverseScheduler.step`](https://github.com/huggingface/diffusers/blob/a65ca8a05974d079a72c736ea60721d01375b13b/src/diffusers/schedulers/scheduling_ddim_inverse.py#L291) does not fit its function api.
There are some redundant args in its docstring:
- eta, variance_noise
... | closed | completed | false | 2 | [
"bug",
"stale"
] | [] | 2024-01-18T16:12:12Z | 2024-02-20T13:47:33Z | 2024-02-20T13:47:33Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | david20571015 | 51,911,434 | MDQ6VXNlcjUxOTExNDM0 | User | false |
huggingface/diffusers | 2,089,230,083 | I_kwDOHa8MBc58hx8D | 6,635 | https://github.com/huggingface/diffusers/issues/6635 | https://api.github.com/repos/huggingface/diffusers/issues/6635 | [Model] Scalable Interpolant Transformers | We have another diffusion model without a UNet backbone - SiT (Scalable Interpolant Transformers), with a great demonstration and a really lovely paper!
Metrics for class-conditional generation on ImageNet 256... | closed | completed | false | 7 | [
"stale",
"contributions-welcome"
] | [] | 2024-01-18T23:08:05Z | 2024-03-05T17:19:58Z | 2024-02-27T15:04:12Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | a-r-r-o-w | 72,266,394 | MDQ6VXNlcjcyMjY2Mzk0 | User | false |
huggingface/diffusers | 2,089,301,504 | I_kwDOHa8MBc58iDYA | 6,636 | https://github.com/huggingface/diffusers/issues/6636 | https://api.github.com/repos/huggingface/diffusers/issues/6636 | Cannot import name 'DIFFUSERS_SLOW_IMPORT' from 'diffusers.utils' | ### Describe the bug
I got errors like below while I was using lpw_stable_diffusion
ImportError: cannot import name 'DIFFUSERS_SLOW_IMPORT' from 'diffusers.utils'
(/usr/local/lib/python3.10/dist-packages/diffusers/utils/__init__.py)
### Reproduction
1 from typing import TYPE_CHECKING ... | closed | completed | false | 7 | [
"bug",
"stale"
] | [] | 2024-01-18T23:50:15Z | 2024-02-26T15:20:48Z | 2024-02-26T15:04:22Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Eneskoser | 25,254,239 | MDQ6VXNlcjI1MjU0MjM5 | User | false |
huggingface/diffusers | 2,090,260,736 | I_kwDOHa8MBc58ltkA | 6,639 | https://github.com/huggingface/diffusers/issues/6639 | https://api.github.com/repos/huggingface/diffusers/issues/6639 | SDXL "load_lora_weights" results in different ouputs than loading in unet and text encoder in two steps | ### Describe the bug
I seem to get very different results depending on how I load the weights of my trained SDXL LoRA.
* In the first experience, I load the weights with the `load_lora_weights` method.
Here is the output for basic text to image pipeline inference:
![first_method_resized_256](https://github.com/... | closed | completed | false | 12 | [
"bug"
] | [] | 2024-01-19T10:54:49Z | 2024-02-19T09:58:05Z | 2024-02-19T09:58:05Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | To-jak | 33,063,955 | MDQ6VXNlcjMzMDYzOTU1 | User | false |
huggingface/diffusers | 2,090,440,666 | I_kwDOHa8MBc58mZfa | 6,640 | https://github.com/huggingface/diffusers/issues/6640 | https://api.github.com/repos/huggingface/diffusers/issues/6640 | Adding LoRA to Unet3DConditional | ### Model/Pipeline/Scheduler description
If i compare apple to apple with LoRA for Unet2DConditional, what should be the `UNET_TARGET_REPLACE_MODULE_CONV` , `DEFAULT_TARGET_REPLACE` parameters and also the schema for the inherited class to modify the LoRA network. In short it will be nice to see aimplementation of LoR... | closed | completed | false | 1 | [
"stale"
] | [] | 2024-01-19T12:19:29Z | 2024-02-26T15:04:18Z | 2024-02-26T15:04:18Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sauradip | 23,197,350 | MDQ6VXNlcjIzMTk3MzUw | User | false |
huggingface/diffusers | 2,090,775,516 | I_kwDOHa8MBc58nrPc | 6,641 | https://github.com/huggingface/diffusers/issues/6641 | https://api.github.com/repos/huggingface/diffusers/issues/6641 | Implement StreamDiffusion | Would it be possible to implement the pipeline optimisations from Stream Diffusion ?
https://github.com/cumulo-autumn/StreamDiffusion
https://huggingface.co/papers/2312.12491
| closed | completed | false | 2 | [
"stale"
] | [] | 2024-01-19T15:23:31Z | 2024-02-28T15:04:24Z | 2024-02-28T15:04:24Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | gabgren | 1,724,721 | MDQ6VXNlcjE3MjQ3MjE= | User | false |
huggingface/diffusers | 2,092,043,315 | I_kwDOHa8MBc58sgwz | 6,648 | https://github.com/huggingface/diffusers/issues/6648 | https://api.github.com/repos/huggingface/diffusers/issues/6648 | Which version of the image encoder is used in stable video diffusion? | Hi,
I noticed that in the SVD, it's mentioned that CLIP's image encoder is used. I would like to know which specific version of the parameters is being referred to. | closed | completed | false | 3 | [
"stale"
] | [] | 2024-01-20T12:52:20Z | 2024-02-27T15:04:06Z | 2024-02-27T15:04:06Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | pixeli99 | 46,072,190 | MDQ6VXNlcjQ2MDcyMTkw | User | false |
huggingface/diffusers | 2,092,592,951 | I_kwDOHa8MBc58um83 | 6,652 | https://github.com/huggingface/diffusers/issues/6652 | https://api.github.com/repos/huggingface/diffusers/issues/6652 | stabilityai/stable-diffusion-2-1 is missing diffusion_pytorch_model.fp16.safetensors | null | closed | completed | false | 2 | [
"stale"
] | [] | 2024-01-21T13:45:56Z | 2024-02-28T15:04:22Z | 2024-02-28T15:04:22Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Gr6800 | 143,057,736 | U_kgDOCIbjSA | User | false |
huggingface/diffusers | 2,092,761,965 | I_kwDOHa8MBc58vQNt | 6,657 | https://github.com/huggingface/diffusers/issues/6657 | https://api.github.com/repos/huggingface/diffusers/issues/6657 | Frames and channels are swapped in UNet3DConditionModel | ### Describe the bug
In docstring for the forward of this model it's stated that sample should have the shape: `batch, num_frames, channel, height, width`, but later before any permutations the model has a string `num_frames = sample.shape[2]`. It seems that these statements contradict each other. The model works when... | closed | completed | false | 1 | [
"bug"
] | [] | 2024-01-21T20:41:19Z | 2024-02-18T07:56:17Z | 2024-02-18T07:56:17Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | MK-2012 | 70,899,588 | MDQ6VXNlcjcwODk5NTg4 | User | false |
huggingface/diffusers | 2,093,154,918 | I_kwDOHa8MBc58wwJm | 6,659 | https://github.com/huggingface/diffusers/issues/6659 | https://api.github.com/repos/huggingface/diffusers/issues/6659 | Adopting different image aspect ratio in training? | **Is your feature request related to a problem? Please describe.**
Hello there,
I have a question regarding the UNet and Lora Training in image AR other than 1:1. Right now I saw the data is resized to 1:1 using the linear interpolation provided from pytorch. But my question is how to handle 16:9 images? Thanks.
... | closed | completed | false | 5 | [
"stale"
] | [] | 2024-01-22T06:04:29Z | 2024-11-23T00:10:13Z | 2024-11-23T00:10:12Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | linnanwang | 4,675,611 | MDQ6VXNlcjQ2NzU2MTE= | User | false |
huggingface/diffusers | 1,389,544,339 | I_kwDOHa8MBc5S0sOT | 666 | https://github.com/huggingface/diffusers/issues/666 | https://api.github.com/repos/huggingface/diffusers/issues/666 | How to resume from checkpoint. | I followed Colab for training. May I know how to resume the training.
Thanks and regards | closed | completed | false | 4 | [
"stale"
] | [
"patil-suraj"
] | 2022-09-28T15:17:34Z | 2022-12-15T18:49:41Z | 2022-12-15T18:49:41Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | BakingBrains | 51,019,420 | MDQ6VXNlcjUxMDE5NDIw | User | false |
huggingface/diffusers | 2,093,294,315 | I_kwDOHa8MBc58xSLr | 6,661 | https://github.com/huggingface/diffusers/issues/6661 | https://api.github.com/repos/huggingface/diffusers/issues/6661 | OOM with MPS on MBP M3 Max 128GB when training specific size of images. | ### Describe the bug
1. When training diffusers, a bug occurs causing out-of-memory (OOM) errors or kernel crashes when the input image size is set to 128x128.
2. Sizes like 64x64, 256x256, and 512x512 do not trigger any errors, and memory usage remains stable, allowing for normal usage.
3. I experimented by only ch... | open | null | false | 3 | [
"bug",
"stale"
] | [] | 2024-01-22T07:49:37Z | 2024-03-17T15:03:40Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | alien2327 | 44,251,382 | MDQ6VXNlcjQ0MjUxMzgy | User | false |
huggingface/diffusers | 2,093,362,895 | I_kwDOHa8MBc58xi7P | 6,662 | https://github.com/huggingface/diffusers/issues/6662 | https://api.github.com/repos/huggingface/diffusers/issues/6662 | DDIB: Dual Diffusion Implicit Bridges for Image to Image Translation | **Is your feature request related to a problem? Please describe.**
Common image translation methods typically depend on joint training using data from both the source and target domains. However, [DDIB](https://suxuann.github.io/ddib/) addresses this challenge by independently training diffusion models on each domain,... | open | null | false | 8 | [
"stale",
"contributions-welcome"
] | [] | 2024-01-22T08:34:04Z | 2025-02-28T15:09:28Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | nityanandmathur | 77,379,835 | MDQ6VXNlcjc3Mzc5ODM1 | User | false |
huggingface/diffusers | 2,093,692,090 | I_kwDOHa8MBc58yzS6 | 6,666 | https://github.com/huggingface/diffusers/issues/6666 | https://api.github.com/repos/huggingface/diffusers/issues/6666 | Black validation images. | ### Describe the bug
Hello, @yiyixuxu , @sayakpaul
I am trying to train controlnet for sdxl. I have followed every step as instructed in Readme_sdxl in examples/controlnet. The pipeline runs fine, the data and model is loaded normally and the training runs smooth, but the validation images I get over on wandb are b... | closed | completed | false | 3 | [
"bug"
] | [] | 2024-01-22T11:23:44Z | 2024-01-23T10:45:43Z | 2024-01-23T10:45:42Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Tauseefahmed1451 | 145,332,470 | U_kgDOCKmY9g | User | false |
huggingface/diffusers | 2,093,854,955 | I_kwDOHa8MBc58zbDr | 6,667 | https://github.com/huggingface/diffusers/issues/6667 | https://api.github.com/repos/huggingface/diffusers/issues/6667 | controlnet-xs-inpaint | ### Describe the bug
Hi, thank you for your nice implement in conbiming controlnet-xs with diffusers. I try the code [here](https://github.com/UmerHA/diffusers/tree/cnxs-training/examples/controlnet_xs) to try the image outpaint (similiar to inpaint task), but i found a Strange thing, I don't know why, can you give me... | open | null | false | 5 | [
"bug",
"stale"
] | [] | 2024-01-22T12:52:50Z | 2024-02-21T15:04:25Z | null | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | xinli2008 | 71,311,087 | MDQ6VXNlcjcxMzExMDg3 | User | false |
huggingface/diffusers | 2,095,166,824 | I_kwDOHa8MBc584bVo | 6,679 | https://github.com/huggingface/diffusers/issues/6679 | https://api.github.com/repos/huggingface/diffusers/issues/6679 | multi device inference of stable diffusion | stable diffusion pipeline is typically composed of multiple Torch models. Many users employing this pipeline may not have high-end GPUs, leading to out-of-memory (OOM) issues when generating high-resolution results or using more models. So, why not load different models onto different GPUs, allowing the entire pipeline... | closed | completed | false | 12 | [
"stale"
] | [] | 2024-01-23T03:24:29Z | 2024-11-22T01:44:05Z | 2024-11-22T01:44:04Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | zhangvia | 38,352,569 | MDQ6VXNlcjM4MzUyNTY5 | User | false |
huggingface/diffusers | 1,389,599,333 | I_kwDOHa8MBc5S05pl | 668 | https://github.com/huggingface/diffusers/issues/668 | https://api.github.com/repos/huggingface/diffusers/issues/668 | [dreambooth] RuntimeError when trying to Train on a Tesla T4 on Colab. | ### Describe the bug
just right when it starts running the training it crashes with out of memory error. When i set it to fp16, i get a unscale_() was already called for error.. I'm not sure how to run this on a 16gb GPU?
### Reproduction
Tesla T4, using 12 training images and i set the step count 1200 and max 100 ... | closed | completed | false | 3 | [
"bug"
] | [] | 2022-09-28T15:55:52Z | 2022-10-03T08:54:03Z | 2022-10-03T08:54:02Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Maki9009 | 62,298,209 | MDQ6VXNlcjYyMjk4MjA5 | User | false |
huggingface/diffusers | 2,095,296,754 | I_kwDOHa8MBc5847Dy | 6,681 | https://github.com/huggingface/diffusers/issues/6681 | https://api.github.com/repos/huggingface/diffusers/issues/6681 | export_to_video issue | #### This issue is coming, it was not there yesterday:
ValueError Traceback (most recent call last)
[<ipython-input-2-0a17f43bdbef>](https://localhost:8080/#) in <cell line: 10>()
8 prompt = "Spiderman is surfing"
9 video_frames = pipe(prompt).frames
---> 10 video_path ... | closed | completed | false | 5 | [] | [] | 2024-01-23T05:57:54Z | 2024-02-18T08:41:11Z | 2024-02-18T08:41:11Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | lucken99 | 33,358,596 | MDQ6VXNlcjMzMzU4NTk2 | User | false |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.