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 | 1,499,476,177 | I_kwDOHa8MBc5ZYDDR | 1,721 | https://github.com/huggingface/diffusers/issues/1721 | https://api.github.com/repos/huggingface/diffusers/issues/1721 | clip skip for Stable diffusion pipelines | Hi,
Is the "clip skip" feature supported by the Stable diffusion pipelines? Just did a little bit of search on the documentation of StableDiffusionPipeline but found nothing related.
See: [Using Hidden States of CLIP’s Penultimate Layer](https://blog.novelai.net/novelai-improvements-on-stable-diffusion-e10d38db82a... | closed | completed | false | 18 | [
"stale"
] | [] | 2022-12-16T02:47:04Z | 2023-06-07T05:18:48Z | 2023-03-03T15:03:58Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | xunings | 43,399,422 | MDQ6VXNlcjQzMzk5NDIy | User | false |
huggingface/diffusers | 1,499,667,130 | I_kwDOHa8MBc5ZYxq6 | 1,722 | https://github.com/huggingface/diffusers/issues/1722 | https://api.github.com/repos/huggingface/diffusers/issues/1722 | Let the launched script handle KeyboardInterrupt | **Is your feature request related to a problem? Please describe.**
As part of my training script the model checkpoint is compressed via 7zip. In my code I handle KeyboardInterrupt so I can skip the compression if necessary (it takes a long time). When I do ctrl+c the signal isn't caught by my script, it's caught by Ac... | closed | completed | false | 1 | [] | [] | 2022-12-16T06:20:48Z | 2022-12-16T22:05:06Z | 2022-12-16T22:05:06Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Cyberes | 64,224,601 | MDQ6VXNlcjY0MjI0NjAx | User | false |
huggingface/diffusers | 1,500,148,333 | I_kwDOHa8MBc5ZanJt | 1,726 | https://github.com/huggingface/diffusers/issues/1726 | https://api.github.com/repos/huggingface/diffusers/issues/1726 | Errors: Converting inpainting diffusers to OG stable diffusion | Am I right in saying the diffusers to OG stable diffusion script (link below), shouldn't work for a non-standard architectures?
I've finetuned an inpainting model from Runway's SD 1.5 Inpainting checkpoint. It has 9 input channels instead of 5. I created a .ckpt file from this finetuned model with the script below,... | closed | completed | false | 3 | [
"stale"
] | [] | 2022-12-16T12:43:32Z | 2023-02-06T07:21:26Z | 2023-01-23T15:03:29Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | BenjaminIrwin | 59,652,119 | MDQ6VXNlcjU5NjUyMTE5 | User | false |
huggingface/diffusers | 1,500,258,123 | I_kwDOHa8MBc5ZbB9L | 1,728 | https://github.com/huggingface/diffusers/issues/1728 | https://api.github.com/repos/huggingface/diffusers/issues/1728 | Transformers library missing even tho its already installed? | 
| closed | completed | false | 3 | [
"stale"
] | [] | 2022-12-16T13:55:36Z | 2023-01-23T15:03:27Z | 2023-01-23T15:03:27Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Spiderfffun | 85,702,798 | MDQ6VXNlcjg1NzAyNzk4 | User | false |
huggingface/diffusers | 1,338,013,244 | I_kwDOHa8MBc5PwHY8 | 173 | https://github.com/huggingface/diffusers/issues/173 | https://api.github.com/repos/huggingface/diffusers/issues/173 | [Stable Diffusion] Longer prompt and rectangular image | Thanks for open-sourcing this great repo! I have been using this code with stable diffusion checkpoints. There are 2 key features that https://github.com/CompVis/stable-diffusion supports, but seem to be missing here:
* Generate rectangular images, which people showcase a lot on the internet.
* Currently the max to... | closed | completed | false | 13 | [
"enhancement"
] | [
"anton-l",
"patil-suraj"
] | 2022-08-13T17:20:46Z | 2022-08-23T12:37:06Z | 2022-08-23T12:37:06Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | DrJimFan | 3,018,029 | MDQ6VXNlcjMwMTgwMjk= | User | false |
huggingface/diffusers | 1,500,480,936 | I_kwDOHa8MBc5Zb4Wo | 1,730 | https://github.com/huggingface/diffusers/issues/1730 | https://api.github.com/repos/huggingface/diffusers/issues/1730 | Converted Model from Diffusers Dreambooth -> to CKPT dont work in stable-diffusion-webui ? | Hey there first thex for the amazing work :) i manged to get Dreambooth runnding on my trust 2080ti and the training works i have results in the 400 and 800 folder, but when i try to load them they dont apear in stable-diffusion-webui. im using his script for the conversion to ckpt https://gist.github.com/Christopher-H... | closed | completed | false | 2 | [
"stale"
] | [] | 2022-12-16T16:00:31Z | 2023-01-24T15:03:20Z | 2023-01-24T15:03:20Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | janoschsimon | 7,808,084 | MDQ6VXNlcjc4MDgwODQ= | User | false |
huggingface/diffusers | 1,500,520,221 | I_kwDOHa8MBc5ZcB8d | 1,731 | https://github.com/huggingface/diffusers/issues/1731 | https://api.github.com/repos/huggingface/diffusers/issues/1731 | Update xformers installation docs when xformers `0.0.16` is released | The wheels will have PyTorch pinned to the latest stable version:
https://github.com/facebookresearch/xformers/issues/533#issuecomment-1355162407
For older versions, we can recommend to install from source. | closed | completed | false | 9 | [] | [
"pcuenca"
] | 2022-12-16T16:31:31Z | 2023-02-01T18:56:34Z | 2023-02-01T18:56:34Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | pcuenca | 1,177,582 | MDQ6VXNlcjExNzc1ODI= | User | false |
huggingface/diffusers | 1,500,522,293 | I_kwDOHa8MBc5ZcCc1 | 1,732 | https://github.com/huggingface/diffusers/issues/1732 | https://api.github.com/repos/huggingface/diffusers/issues/1732 | [Community Pipeline] Superresolution with Transformers | :hugs: Transformers now also includes an image super resolution/restoration model: Swin2SR (based on Swinv2) :fire:
models on the hub: https://huggingface.co/caidas
docs: https://huggingface.co/docs/transformers/main/en/model_doc/swin2sr
We could combine this very nicely with `diffusers` to build some cool commun... | closed | completed | false | 4 | [
"community-examples",
"stale"
] | [] | 2022-12-16T16:32:50Z | 2023-04-20T15:03:54Z | 2023-04-20T15:03:54Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/diffusers | 1,500,568,131 | I_kwDOHa8MBc5ZcNpD | 1,733 | https://github.com/huggingface/diffusers/issues/1733 | https://api.github.com/repos/huggingface/diffusers/issues/1733 | [happy to fix this] Wrong a weighted mean squared error loss | https://github.com/huggingface/diffusers/blob/acd317810bc138b3a78fa30e1b3da1006c1d60ad/examples/unconditional_image_generation/train_unconditional.py#L421-L428
1. Big issue: `alpha_t / (1 - alpha_t)` is the **signal-to-noise ratio (SNR)**. The weighting scheme `snr_weights` is different from the paper and doesn't ma... | closed | completed | false | 2 | [
"stale"
] | [] | 2022-12-16T17:08:40Z | 2023-01-24T15:03:18Z | 2023-01-24T15:03:18Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | aengusng8 | 67,547,213 | MDQ6VXNlcjY3NTQ3MjEz | User | false |
huggingface/diffusers | 1,500,865,103 | I_kwDOHa8MBc5ZdWJP | 1,734 | https://github.com/huggingface/diffusers/issues/1734 | https://api.github.com/repos/huggingface/diffusers/issues/1734 | [Dreambooth] Multi-GPU training with accelerate is magnitudes slower than single GPU (non-flax) | ### Describe the bug
I have access to a machine with several NVIDIA A100s. Initially I was using multiple GPUs for Dreambooth training under the idea that it would speed up training. After using both multiple GPUs and a single GPU to train, I have found the opposite to be true. The speed difference is an order of ma... | closed | completed | false | 12 | [
"bug"
] | [
"williamberman"
] | 2022-12-16T21:07:11Z | 2023-01-25T13:48:48Z | 2022-12-28T16:01:34Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | jkcarney | 48,502,124 | MDQ6VXNlcjQ4NTAyMTI0 | User | false |
huggingface/diffusers | 1,500,874,345 | I_kwDOHa8MBc5ZdYZp | 1,735 | https://github.com/huggingface/diffusers/issues/1735 | https://api.github.com/repos/huggingface/diffusers/issues/1735 | [Question] Shouldn't the inputs for memory_efficient_attention in the CrossAttention have [B, sequence length, Heads, embeding size] format? | According to the xformer's documentation (https://facebookresearch.github.io/xformers/components/ops.html#xformers.ops.memory_efficient_attention ):
> Input tensors must be in format [B, M, H, K], where B is the batch size, M the sequence length, H the number of heads, and K the embeding size per head
> If inputs h... | closed | completed | false | 1 | [] | [] | 2022-12-16T21:16:15Z | 2022-12-21T11:28:35Z | 2022-12-21T11:28:35Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Warvito | 4,493,399 | MDQ6VXNlcjQ0OTMzOTk= | User | false |
huggingface/diffusers | 1,501,350,083 | I_kwDOHa8MBc5ZfMjD | 1,738 | https://github.com/huggingface/diffusers/issues/1738 | https://api.github.com/repos/huggingface/diffusers/issues/1738 | train_dreambooth_flax not found gpu error | No GPU/TPU found, falling back to CPU. (Set TF_CPP_MIN_LOG_LEVEL=0 and rerun for more info.) , i have the server with gpu its A100 80gb | closed | completed | false | 2 | [
"stale"
] | [] | 2022-12-17T11:36:33Z | 2023-01-25T15:03:37Z | 2023-01-25T15:03:37Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | rahilparsana | 52,197,155 | MDQ6VXNlcjUyMTk3MTU1 | User | false |
huggingface/diffusers | 1,338,101,387 | I_kwDOHa8MBc5Pwc6L | 174 | https://github.com/huggingface/diffusers/issues/174 | https://api.github.com/repos/huggingface/diffusers/issues/174 | Trying to train images from a local folder fails | ### Describe the bug
Following this section of the diffuser's training example:
```
# Or just load images from a local folder!
config.dataset = "imagefolder"
dataset = load_dataset(config.dataset, data_dir="path/to/folder")
```
Fails with repeated errors while trying to load the dataset:
```
OSError: [WinErr... | closed | completed | false | 4 | [
"bug"
] | [] | 2022-08-14T01:04:55Z | 2022-09-13T15:43:35Z | 2022-09-13T15:43:35Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sam598 | 10,446,449 | MDQ6VXNlcjEwNDQ2NDQ5 | User | false |
huggingface/diffusers | 1,501,408,971 | I_kwDOHa8MBc5Zfa7L | 1,740 | https://github.com/huggingface/diffusers/issues/1740 | https://api.github.com/repos/huggingface/diffusers/issues/1740 | can't pickle local object in train_dreambooth.py | ### Describe the bug
When trying dream booth training I get this error
### Reproduction
accelerate launch train_dreambooth.py --mixed_precision="fp16" --pretrained_model_name_or_path="stabilityai/stable-diffusion-2-1-base" --train_text_encoder --instance_data_dir="C:\Users\VERTIGO\Documents\my_concept" --class_d... | closed | completed | false | 4 | [
"bug",
"stale"
] | [] | 2022-12-17T14:19:12Z | 2023-01-24T15:03:16Z | 2023-01-24T15:03:16Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | vibber | 4,029,429 | MDQ6VXNlcjQwMjk0Mjk= | User | false |
huggingface/diffusers | 1,501,475,765 | I_kwDOHa8MBc5ZfrO1 | 1,741 | https://github.com/huggingface/diffusers/issues/1741 | https://api.github.com/repos/huggingface/diffusers/issues/1741 | [Question][Solved] Can unet_traced.pt (TORCH.JIT.TRACE) make diffusers faster? | I saw it here
https://huggingface.co/riffusion/riffusion-model-v1/tree/main/unet_traced
and found here
https://huggingface.co/docs/diffusers/optimization/fp16
and here
https://pytorch.org/docs/stable/generated/torch.jit.trace.html
![Screenshot 2022-12-15 203423](https://user-images.githubusercontent.com/543702... | closed | completed | false | 9 | [] | [] | 2022-12-17T17:03:52Z | 2022-12-22T23:34:34Z | 2022-12-22T23:34:34Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | camenduru | 54,370,274 | MDQ6VXNlcjU0MzcwMjc0 | User | false |
huggingface/diffusers | 1,501,510,131 | I_kwDOHa8MBc5Zfznz | 1,742 | https://github.com/huggingface/diffusers/issues/1742 | https://api.github.com/repos/huggingface/diffusers/issues/1742 | list of prompts with same content leads to different inference speed | ### Describe the bug
There are two cases:
1.
```
prompt_1 = "green grass"
prompt_2 = "apple tree"
prompt = [prompt_1, prompt_2]
```
2.
```
prompt = "green grass, apple tree"
```
Parameter *prompt* of pipe object can receive string or list of string. However, if I pass list of string into pipe, inference spe... | closed | completed | false | 2 | [
"bug"
] | [] | 2022-12-17T18:22:08Z | 2022-12-18T10:41:52Z | 2022-12-18T10:41:52Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | chesha1 | 49,867,392 | MDQ6VXNlcjQ5ODY3Mzky | User | false |
huggingface/diffusers | 1,501,611,532 | I_kwDOHa8MBc5ZgMYM | 1,745 | https://github.com/huggingface/diffusers/issues/1745 | https://api.github.com/repos/huggingface/diffusers/issues/1745 | xformers assertion errors (5-10x greater error than allowed) | ### Describe the bug
When running the pytest suite, almost all xformers tests are all failing due to assertion errors,
### Reproduction
`pytest -k "xformers" tests
`
### Logs
```shell
`pytest -k "xformers" tests
`
tests/models/test_models_unet_2d.py . ... | closed | completed | false | 2 | [
"bug",
"stale"
] | [] | 2022-12-17T21:46:16Z | 2023-01-25T15:03:36Z | 2023-01-25T15:03:36Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Thomas-MMJ | 112,830,596 | U_kgDOBrmohA | User | false |
huggingface/diffusers | 1,501,635,315 | I_kwDOHa8MBc5ZgSLz | 1,746 | https://github.com/huggingface/diffusers/issues/1746 | https://api.github.com/repos/huggingface/diffusers/issues/1746 | error in pipeline_flax_stable_diffusion | ### Describe the bug
I run into this error when trying to run the flax example of dreambooth exactly as it is instructed https://github.com/huggingface/diffusers/tree/main/examples/dreambooth#running-with-flaxjax
the error message is:
```python
Traceback (most recent call last):
File "train_dreambooth_flax.p... | closed | completed | false | 4 | [
"bug"
] | [
"pcuenca"
] | 2022-12-17T23:21:27Z | 2022-12-19T19:42:26Z | 2022-12-19T19:42:26Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | yiyixuxu | 12,631,849 | MDQ6VXNlcjEyNjMxODQ5 | User | false |
huggingface/diffusers | 1,501,642,548 | I_kwDOHa8MBc5ZgT80 | 1,747 | https://github.com/huggingface/diffusers/issues/1747 | https://api.github.com/repos/huggingface/diffusers/issues/1747 | Getting a command not found error when trying to use script | Cloned the repo and tried using the convert_original_stable diffusion_to_diffusers.py to convert my .ckpt file and I'm getting a
`sudo: ./scripts/convert_original_stable_diffusion_to_diffusers.py: command not found` error. Any idea what I can do to utilize the script?
Command I'm running
`./scripts/convert_origi... | closed | completed | false | 2 | [
"stale"
] | [] | 2022-12-17T23:48:36Z | 2023-01-25T15:03:34Z | 2023-01-25T15:03:34Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | monis24 | 52,933,919 | MDQ6VXNlcjUyOTMzOTE5 | User | false |
huggingface/diffusers | 1,501,820,225 | I_kwDOHa8MBc5Zg_VB | 1,748 | https://github.com/huggingface/diffusers/issues/1748 | https://api.github.com/repos/huggingface/diffusers/issues/1748 | Does diffusers support Stable Diffusion Version 2 now？ | As mentioned in the question, Does diffusers support Stable Diffusion Version 2 now？
If supported，how to use it？Thanks. | closed | completed | false | 2 | [
"stale"
] | [] | 2022-12-18T08:53:43Z | 2023-01-25T15:03:33Z | 2023-01-25T15:03:33Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Youngboy12 | 34,464,054 | MDQ6VXNlcjM0NDY0MDU0 | User | false |
huggingface/diffusers | 1,338,107,126 | I_kwDOHa8MBc5PweT2 | 175 | https://github.com/huggingface/diffusers/issues/175 | https://api.github.com/repos/huggingface/diffusers/issues/175 | Is the use of `torch.manual_seed` in example training code correct? | ### Describe the bug
https://github.com/huggingface/diffusers/blob/92b6dbba1a25ed27b0bae38b089715132c7e6bcc/examples/train_unconditional.py#L168
This is a minor thing, but I think this should be `torch.Generator().manual_seed(0)`. In my understanding, if `torch.manual_seed` is called, it sets the seed globally an... | closed | completed | false | 10 | [
"bug"
] | [] | 2022-08-14T01:51:18Z | 2022-10-26T13:34:52Z | 2022-09-13T15:43:56Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | hysts | 25,161,192 | MDQ6VXNlcjI1MTYxMTky | User | false |
huggingface/diffusers | 1,502,014,227 | I_kwDOHa8MBc5ZhusT | 1,750 | https://github.com/huggingface/diffusers/issues/1750 | https://api.github.com/repos/huggingface/diffusers/issues/1750 | Symbol not defined on Debian VM | ### Describe the bug
I'm following the examples/dreambooth README, and after installing dependencies, I cannot configurate accelerate. When I run `accelerate config` or `accelerate config default`, it gives the following error:
`OSError: /opt/conda/lib/python3.7/site-packages/torch/lib/../../nvidia/cublas/lib/libcu... | closed | completed | false | 8 | [
"bug",
"stale"
] | [] | 2022-12-18T20:00:48Z | 2023-03-17T06:04:26Z | 2023-02-14T15:03:59Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | balintdecsi | 49,560,050 | MDQ6VXNlcjQ5NTYwMDUw | User | false |
huggingface/diffusers | 1,502,531,745 | I_kwDOHa8MBc5ZjtCh | 1,755 | https://github.com/huggingface/diffusers/issues/1755 | https://api.github.com/repos/huggingface/diffusers/issues/1755 | علاقات عامة | ### Model/Pipeline/Scheduler description
علاقات عامة
### Open source status
- [X] The model implementation is available
- [X] The model weights are available (Only relevant if addition is not a scheduler).
### Provide useful links for the implementation
_No response_ | closed | completed | false | 2 | [
"stale"
] | [] | 2022-12-19T08:37:49Z | 2023-01-27T15:03:12Z | 2023-01-27T15:03:12Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | naserafn | 62,495,734 | MDQ6VXNlcjYyNDk1NzM0 | User | false |
huggingface/diffusers | 1,338,142,150 | I_kwDOHa8MBc5Pwm3G | 176 | https://github.com/huggingface/diffusers/issues/176 | https://api.github.com/repos/huggingface/diffusers/issues/176 | Glide pipeline missing | ### Describe the bug
The readme file in pipelines describes glide but the link is broken as glide folder does not exist, was it removed?
### Reproduction
_No response_
### Logs
_No response_
### System Info
```shell
Missing glide folder in pipelines
```
| closed | completed | false | 7 | [
"bug",
"enhancement"
] | [] | 2022-08-14T05:46:15Z | 2026-01-09T17:47:45Z | 2026-01-09T17:47:45Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | jfdelgad | 36,968,968 | MDQ6VXNlcjM2OTY4OTY4 | User | false |
huggingface/diffusers | 1,502,936,391 | I_kwDOHa8MBc5ZlP1H | 1,760 | https://github.com/huggingface/diffusers/issues/1760 | https://api.github.com/repos/huggingface/diffusers/issues/1760 | bug: onnxruntime.capi.onnxruntime_pybind11_state.RuntimeException: [ONNXRuntimeError] : 6 : RUNTIME_EXCEPTION : Non-zero status code returned while running OpenVINO-EP-subgraph_4 node. | ### Describe the bug
I was trying to get inference for stable diffusion onnx {CompVis/stable-diffusion-v1-4} model, I got stuck at this error
image = pipe(prompt).images[0] File "onnx_openvpy39\lib\site-packages\diffusers\pipelines\stable_diffusion\pipeline_onnx_stable_diffusion.py", line 274, in __call__ ... | closed | completed | false | 24 | [
"bug"
] | [
"anton-l"
] | 2022-12-19T13:15:21Z | 2023-03-02T14:06:37Z | 2023-03-02T14:06:37Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | saikrishna2893 | 84,713,792 | MDQ6VXNlcjg0NzEzNzky | User | false |
huggingface/diffusers | 1,503,251,677 | I_kwDOHa8MBc5Zmczd | 1,768 | https://github.com/huggingface/diffusers/issues/1768 | https://api.github.com/repos/huggingface/diffusers/issues/1768 | [Tutorial] Transform Your Selfie into a Stunning AI Avatar with Stable Diffusion - Better than Lensa for Free | ERROR: type should be string, got "https://www.youtube.com/watch?v=mnCY8uM7E50\r\n\r\n[](https://www.youtube.com/watch?v=mnCY8uM7E50)\r\n\r\nIn this video, I will show you how to create your own Lensa app-like magic avatars without using any third-party apps. Our method ensures 100% privacy, unlike uploading your personal photos to external apps and platforms. I am sure you have seen many people sharing their magic avatars on social media. While these paid apps use free open-source AI models, they still require payment. Additionally, AI image generation apps may keep your private photos and use them as they please. Instead of paying and trusting the goodwill of these paid apps, we will use these open-source public AI models ourselves. I will provide step-by-step instructions, so even those with no technical expertise can follow along. We will be using Google's Colab, a free #AI platform that offers access to a powerful GPU at no cost. Therefore, this tutorial can even be completed on a mobile phone without a PC. All you need is a Gmail account. Once we train the open-source image generation model #StableDiffusion with #DreamBooth by using our own portrait images, the possibilities are endless. Unlike other applications, the method I will demonstrate does not impose any limitations on image generation. You will be able to freely compose any kind of image as many times as you desire. You can not only generate your own avatars, but also any other images you want, such as highly detailed car images or landscapes." | closed | completed | false | 1 | [] | [] | 2022-12-19T16:44:37Z | 2022-12-19T16:44:47Z | 2022-12-19T16:44:47Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | FurkanGozukara | 19,240,467 | MDQ6VXNlcjE5MjQwNDY3 | User | false |
huggingface/diffusers | 1,338,146,814 | I_kwDOHa8MBc5Pwn_- | 177 | https://github.com/huggingface/diffusers/issues/177 | https://api.github.com/repos/huggingface/diffusers/issues/177 | Add documentation for `subfolder` argument to `ModelMixin.from_pretrained` | **Is your feature request related to a problem? Please describe.**
The `ModelMixin.from_pretrained` has a keyword argument `subfolder`, but it's not documented in its docstring and it's hard for newcomers to know how to load models from HF Model Hub in case the model is saved with pipeline.
For example, when a DD... | closed | completed | false | 2 | [] | [] | 2022-08-14T06:16:47Z | 2022-09-13T17:18:03Z | 2022-09-13T17:18:03Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | hysts | 25,161,192 | MDQ6VXNlcjI1MTYxMTky | User | false |
huggingface/diffusers | 1,503,413,241 | I_kwDOHa8MBc5ZnEP5 | 1,772 | https://github.com/huggingface/diffusers/issues/1772 | https://api.github.com/repos/huggingface/diffusers/issues/1772 | Unconditional Image Generation generating noise | Hi, As mentioned in https://huggingface.co/docs/diffusers/training/unconditional_training, I am trying to fine-tune on the CartoonSet[https://google.github.io/cartoonset/] dataset. However, after training for 100 epochs, on inspecting via Tensorboard, I still see only noise being generated.
Dataset stats -
10k ima... | closed | completed | false | 8 | [
"stale"
] | [] | 2022-12-19T18:53:16Z | 2023-01-30T10:35:25Z | 2023-01-30T10:35:25Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | agneet42 | 22,055,826 | MDQ6VXNlcjIyMDU1ODI2 | User | false |
huggingface/diffusers | 1,503,562,815 | I_kwDOHa8MBc5Znow_ | 1,773 | https://github.com/huggingface/diffusers/issues/1773 | https://api.github.com/repos/huggingface/diffusers/issues/1773 | attention.py test can force you to uninstall xformer | ### Describe the bug
https://github.com/huggingface/diffusers/blob/9f8c915a756cf937a428186a2af8cc56f029f689/src/diffusers/models/attention.py#L305
https://github.com/huggingface/diffusers/blob/9f8c915a756cf937a428186a2af8cc56f029f689/src/diffusers/models/attention.py#L468
I have users that have xformer pre-compile... | closed | completed | false | 4 | [
"bug",
"stale"
] | [] | 2022-12-19T20:35:59Z | 2023-01-28T15:03:06Z | 2023-01-28T15:03:06Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | BurguerJohn | 32,472,975 | MDQ6VXNlcjMyNDcyOTc1 | User | false |
huggingface/diffusers | 1,503,564,596 | I_kwDOHa8MBc5ZnpM0 | 1,774 | https://github.com/huggingface/diffusers/issues/1774 | https://api.github.com/repos/huggingface/diffusers/issues/1774 | A6000 GPU: Getting torch.cuda.OutOfMemoryError: CUDA out of memory | I am trying to finetune a dreambooth model using diffusers repo.
GPU: **A6000 48GB VRAM**
but getting the following error:
```
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 1.50 GiB (GPU 0; 47.54 GiB total capacity; 43.42 GiB already allocated; 1.50 GiB free; 44.20 GiB reserved
in total b... | closed | completed | false | 4 | [
"stale"
] | [] | 2022-12-19T20:37:01Z | 2023-02-06T15:03:24Z | 2023-02-06T15:03:24Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | geekyayush | 25,874,080 | MDQ6VXNlcjI1ODc0MDgw | User | false |
huggingface/diffusers | 1,503,888,475 | I_kwDOHa8MBc5Zo4Rb | 1,779 | https://github.com/huggingface/diffusers/issues/1779 | https://api.github.com/repos/huggingface/diffusers/issues/1779 | Language support and contribution to docs | I would like to contribute to translating the documentation into Korean.
It seems that there is a version and language selection in the docs, but only EN is currently supported.
Please let me know if there is any possibility or way to contribute with translation. | closed | completed | false | 2 | [
"stale"
] | [] | 2022-12-20T02:20:50Z | 2023-02-03T15:03:31Z | 2023-02-03T15:03:31Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | seriousran | 4,317,641 | MDQ6VXNlcjQzMTc2NDE= | User | false |
huggingface/diffusers | 1,504,112,199 | I_kwDOHa8MBc5Zpu5H | 1,784 | https://github.com/huggingface/diffusers/issues/1784 | https://api.github.com/repos/huggingface/diffusers/issues/1784 | Add support for multiple timesteps in Noise Schedulers | **Is your feature request related to a problem? Please describe.**
When trying to call scheduler.step with multiple timesteps it throws an error: `Boolean value of Tensor with more than one value is ambiguous`. The origin of this error for DDIM as example is here: https://github.com/huggingface/diffusers/blob/261a448c... | closed | completed | false | 5 | [
"stale"
] | [] | 2022-12-20T07:32:53Z | 2023-09-04T21:12:16Z | 2023-01-30T05:12:13Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | bonlime | 14,337,581 | MDQ6VXNlcjE0MzM3NTgx | User | false |
huggingface/diffusers | 1,504,242,607 | I_kwDOHa8MBc5ZqOuv | 1,785 | https://github.com/huggingface/diffusers/issues/1785 | https://api.github.com/repos/huggingface/diffusers/issues/1785 | `num_images_per_prompt` > 1 still broken with `UnCLIPPipeline` | ### Describe the bug
If I use the setting `num_images_per_prompt` with `UnCLIPPipeline` I get the following error:
```
---------------------------------------------------------------------------
RuntimeError Traceback (most recent call last)
[<ipython-input-12-a06ca1ef5374>](https://lo... | closed | completed | false | 0 | [
"bug"
] | [
"patil-suraj"
] | 2022-12-20T09:23:20Z | 2022-12-20T15:03:40Z | 2022-12-20T15:03:40Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | apolinario | 788,417 | MDQ6VXNlcjc4ODQxNw== | User | false |
huggingface/diffusers | 1,504,479,579 | I_kwDOHa8MBc5ZrIlb | 1,786 | https://github.com/huggingface/diffusers/issues/1786 | https://api.github.com/repos/huggingface/diffusers/issues/1786 | Can the progress bar from pipe be suppressed ? | Hi,
When we pass a prompt to the pip (from for eg: pipe = StableDiffusionPipeline.from_pretrained("./stable-diffusion-v1-5")), it displays an output in this case, with a progress bar. Is it possible to get an output without the progress bar, or to somehow disable it?
Thanks in advance! | closed | completed | false | 3 | [] | [] | 2022-12-20T12:24:08Z | 2024-09-28T23:45:28Z | 2022-12-21T08:18:00Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Trisanu-007 | 56,588,589 | MDQ6VXNlcjU2NTg4NTg5 | User | false |
huggingface/diffusers | 1,505,525,570 | I_kwDOHa8MBc5ZvH9C | 1,790 | https://github.com/huggingface/diffusers/issues/1790 | https://api.github.com/repos/huggingface/diffusers/issues/1790 | How training works ? | Hi, I've tried different things on training, i tried with text encoder worked good, but the results on each prompt was the same, I'have tried first with text encoder then added more steps without text encoder it was the same, when i tried without text encoder in 1000 steps with 10 images every prompt was really good bu... | closed | completed | false | 8 | [
"stale"
] | [] | 2022-12-21T00:34:16Z | 2023-02-05T15:03:04Z | 2023-02-05T15:03:04Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ariand9 | 6,691,197 | MDQ6VXNlcjY2OTExOTc= | User | false |
huggingface/diffusers | 1,505,556,359 | I_kwDOHa8MBc5ZvPeH | 1,791 | https://github.com/huggingface/diffusers/issues/1791 | https://api.github.com/repos/huggingface/diffusers/issues/1791 | UnCLIPPipeline not supported on Mac/MPS | ### Describe the bug
_Tested on latest diffusers commit, [`7eee6c3`](https://github.com/huggingface/diffusers/tree/4125756e88e82370c197fecf28e9f0b4d7eee6c3)._
When I run UnCLIPPipeline on MPS backend: Python crashes with this error
```
Assertion failed: (vals.isSplat()), function getConstantValue, file Common... | closed | completed | false | 9 | [
"bug",
"stale"
] | [
"pcuenca"
] | 2022-12-21T01:09:57Z | 2023-03-13T15:04:18Z | 2023-03-13T15:04:18Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Birch-san | 6,141,784 | MDQ6VXNlcjYxNDE3ODQ= | User | false |
huggingface/diffusers | 1,505,650,341 | I_kwDOHa8MBc5Zvmal | 1,792 | https://github.com/huggingface/diffusers/issues/1792 | https://api.github.com/repos/huggingface/diffusers/issues/1792 | Img2Img pipeline initialization fails on SD variants with no feature extractor/safety checker | ### Describe the bug
The SD2.1 base repo didn't have the safety checker components. It still loaded fine in a normal StableDiffusionPipeline, but failed when using the StableDiffusionImg2ImgPipeline since that requires a safety checker. Initializing it with the components of the base StableDiffusionPipeline worked b... | closed | completed | false | 1 | [
"bug"
] | [
"pcuenca"
] | 2022-12-21T03:18:55Z | 2022-12-25T20:58:46Z | 2022-12-25T20:58:46Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | johnowhitaker | 6,575,163 | MDQ6VXNlcjY1NzUxNjM= | User | false |
huggingface/diffusers | 1,506,154,275 | I_kwDOHa8MBc5Zxhcj | 1,794 | https://github.com/huggingface/diffusers/issues/1794 | https://api.github.com/repos/huggingface/diffusers/issues/1794 | OOM when using PyTorch, if Flax/JAX are installed | Bisected to f106ab4 | closed | completed | false | 1 | [
"bug"
] | [
"pcuenca"
] | 2022-12-21T11:28:49Z | 2022-12-21T13:07:25Z | 2022-12-21T13:07:25Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | pcuenca | 1,177,582 | MDQ6VXNlcjExNzc1ODI= | User | false |
huggingface/diffusers | 1,506,428,032 | I_kwDOHa8MBc5ZykSA | 1,797 | https://github.com/huggingface/diffusers/issues/1797 | https://api.github.com/repos/huggingface/diffusers/issues/1797 | UNet1DModel BUG occuered | ### Describe the bug
UNet1DModel
the input tensor's shape is [bsz, len, dim_h] (16, 512, 768)
and it reports this error:
RuntimeError: Given groups=1, weight of size [32, 768, 1], expected input[16, 528, 768] to have 768 channels, but got 528 channels instead
### Reproduction
self.unet = UNet1DModel(sample_size=... | closed | completed | false | 4 | [
"bug",
"stale"
] | [] | 2022-12-21T14:54:28Z | 2023-03-06T13:05:01Z | 2023-02-05T15:03:02Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | hzy312 | 61,484,865 | MDQ6VXNlcjYxNDg0ODY1 | User | false |
huggingface/diffusers | 1,506,618,681 | I_kwDOHa8MBc5ZzS05 | 1,798 | https://github.com/huggingface/diffusers/issues/1798 | https://api.github.com/repos/huggingface/diffusers/issues/1798 | Flax dreambooth/training script currently does not support v_prediction and thus doesn't work on SD v2 | **Is your feature request related to a problem? Please describe.**
Flax version of the training script under example did not implement different loss target for v_prediction.
**Describe the solution you'd like**
Implement v_prediction
**Describe alternatives you've considered**
I could probably implement it ba... | closed | completed | false | 4 | [] | [] | 2022-12-21T16:52:08Z | 2023-01-07T14:13:07Z | 2023-01-07T14:13:07Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Lime-Cakes | 91,322,985 | MDQ6VXNlcjkxMzIyOTg1 | User | false |
huggingface/diffusers | 1,338,299,356 | I_kwDOHa8MBc5PxNPc | 180 | https://github.com/huggingface/diffusers/issues/180 | https://api.github.com/repos/huggingface/diffusers/issues/180 | `AutoencoderKL.encode` returns diagonal Gaussian instead of hidden states | This is mostly a question.
I find it a little weird that the output of `encode` in `AutoencoderKL` is the diagonal Gaussian distribution from which the hidden states are obtained, instead of the hidden states themselves: https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/vae.py#L436. Instead, th... | closed | completed | false | 1 | [] | [] | 2022-08-14T16:57:15Z | 2022-09-13T16:01:14Z | 2022-09-13T16:01:14Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | pcuenca | 1,177,582 | MDQ6VXNlcjExNzc1ODI= | User | false |
huggingface/diffusers | 1,506,920,676 | I_kwDOHa8MBc5Z0cjk | 1,801 | https://github.com/huggingface/diffusers/issues/1801 | https://api.github.com/repos/huggingface/diffusers/issues/1801 | getattr(): attribute name must be string | ### Describe the bug
started throwing this error:
getattr(): attribute name must be string
### Reproduction
code to reproduce error: ClipLoop.ipynb found in this repo: https://gith... | closed | completed | false | 10 | [
"bug"
] | [] | 2022-12-21T21:20:37Z | 2023-01-14T02:28:12Z | 2022-12-24T23:55:42Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | TheOneTrueGuy | 3,240,071 | MDQ6VXNlcjMyNDAwNzE= | User | false |
huggingface/diffusers | 1,507,096,282 | I_kwDOHa8MBc5Z1Hba | 1,802 | https://github.com/huggingface/diffusers/issues/1802 | https://api.github.com/repos/huggingface/diffusers/issues/1802 | encoder_hidden_states=None not working on unet_2d_condition | ### Describe the bug
Not sure if this is a bug or not.
In stable diffusion pipeline, I want to put None in encoder_hidden_states to get the true unconditional noise_pred, not with a text embedding of blank strings. However, this is giving me the error:
`RuntimeError: mat1 and mat2 shapes cannot be multiplied (4096... | closed | completed | false | 2 | [
"bug"
] | [] | 2022-12-22T00:48:32Z | 2023-01-04T12:50:28Z | 2023-01-04T12:50:28Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ghost | 10,137 | MDQ6VXNlcjEwMTM3 | User | false |
huggingface/diffusers | 1,507,525,546 | I_kwDOHa8MBc5Z2wOq | 1,805 | https://github.com/huggingface/diffusers/issues/1805 | https://api.github.com/repos/huggingface/diffusers/issues/1805 | Add Scalable Diffusion Pipline | ### Model/Pipeline/Scheduler description
Scalable Diffusion train latent diffusion models, replacing the commonly-used U-Net backbone with a transformer that operates on latent patches. We analyze the scalability of our Diffusion Transformers (DiTs) through the lens of forward pass complexity as measured by Gflops. We... | closed | completed | false | 2 | [
"stale"
] | [] | 2022-12-22T09:22:02Z | 2023-01-31T08:18:00Z | 2023-01-31T08:18:00Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | MichaelFan01 | 49,156,452 | MDQ6VXNlcjQ5MTU2NDUy | User | false |
huggingface/diffusers | 1,507,876,239 | I_kwDOHa8MBc5Z4F2P | 1,807 | https://github.com/huggingface/diffusers/issues/1807 | https://api.github.com/repos/huggingface/diffusers/issues/1807 | Stable diffusion inference gets slower with each generation | ### Describe the bug
I'm running a simple benchmark and the speed of SD drops with each generation. Tried with various settings and getting same speed decay. VRAM and RAM are not leaking.
### Reproduction
```
from diffusers import StableDiffusionPipeline
import torch
import time
use_xformers = False
u... | closed | completed | false | 5 | [
"bug",
"stale"
] | [] | 2022-12-22T13:20:22Z | 2023-05-07T21:16:33Z | 2023-02-05T15:03:01Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Daniel-Kelvich | 19,155,101 | MDQ6VXNlcjE5MTU1MTAx | User | false |
huggingface/diffusers | 1,508,216,913 | I_kwDOHa8MBc5Z5ZBR | 1,808 | https://github.com/huggingface/diffusers/issues/1808 | https://api.github.com/repos/huggingface/diffusers/issues/1808 | Combine `UnCLIPPipeline` and `StableDiffusionImageVariationPipeline` | ### Model/Pipeline/Scheduler description
`UnCLIPPipeline`("kakaobrain/karlo-v1-alpha") provide a prior model that can generate clip image embedding from text.
`StableDiffusionImageVariationPipeline`("lambdalabs/sd-image-variations-diffusers") provide a decoder model than can generate images from clip image embedding.... | closed | completed | false | 7 | [] | [] | 2022-12-22T16:54:10Z | 2023-01-23T01:05:34Z | 2023-01-23T01:05:34Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | budui | 16,448,529 | MDQ6VXNlcjE2NDQ4NTI5 | User | false |
huggingface/diffusers | 1,508,242,458 | I_kwDOHa8MBc5Z5fQa | 1,809 | https://github.com/huggingface/diffusers/issues/1809 | https://api.github.com/repos/huggingface/diffusers/issues/1809 | AutoEncoderKL isn't releasing memory | ### Describe the bug
Using the encode function to get the latent_dist product returns the proper object but I can't seem to be able to get rid of it once it's introduced, del the variable isn't working, accelerator.free_memory() not working, gc.collect etc.
Perhaps I'm doing something wrong or is it a bug?
###... | closed | completed | false | 9 | [
"bug"
] | [] | 2022-12-22T17:07:57Z | 2023-08-23T15:55:19Z | 2022-12-25T07:40:05Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | devilismyfriend | 87,043,616 | MDQ6VXNlcjg3MDQzNjE2 | User | false |
huggingface/diffusers | 1,339,049,836 | I_kwDOHa8MBc5P0Eds | 181 | https://github.com/huggingface/diffusers/issues/181 | https://api.github.com/repos/huggingface/diffusers/issues/181 | train_unconditional.py fails on writing image | ### Describe the bug
After the first epoch `train_unconditional.py` attempts to write an image and fails.
```
Epoch 0: 100%|███████████████████████████████████████████████████████| 5000/5000 [21:13<00:00, 3.93it/s, ema_decay=0.998, loss=0.027, lr=0.0001, step=4999]
100%|████████████████████████████████████████████... | closed | completed | false | 3 | [
"stale"
] | [] | 2022-08-15T14:07:15Z | 2022-10-10T13:01:16Z | 2022-10-10T13:01:15Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sam598 | 10,446,449 | MDQ6VXNlcjEwNDQ2NDQ5 | User | false |
huggingface/diffusers | 1,508,338,126 | I_kwDOHa8MBc5Z52nO | 1,810 | https://github.com/huggingface/diffusers/issues/1810 | https://api.github.com/repos/huggingface/diffusers/issues/1810 | Numpy Deprecated types | ### Describe the bug
diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py
Numpy latest has been deprecated float for cfloat I think?
### Reproduction
# OnnxStableDiffusionPipeline on CPU
### Logs
```shell
Traceback (most recent call last):
File "/usr/local/bin/docker-entrypoint.py", line 25... | closed | completed | false | 2 | [
"bug"
] | [] | 2022-12-22T18:06:23Z | 2023-01-02T12:24:36Z | 2023-01-02T12:24:35Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Danucas | 45,695,858 | MDQ6VXNlcjQ1Njk1ODU4 | User | false |
huggingface/diffusers | 1,508,482,711 | I_kwDOHa8MBc5Z6Z6X | 1,811 | https://github.com/huggingface/diffusers/issues/1811 | https://api.github.com/repos/huggingface/diffusers/issues/1811 | `from_pretrained` downloads `*.bin` and `*.safetensors` weights if both are present in the repo | ### Describe the bug
If both `*.bin` and `*.safetensors` weights exist in a repository both are downloaded even thought only the `*.safetensors` version are used, doubling the disk size & time to download models.
As we load `*.safetensors` by default, we should only download them.
### Reproduction
```py
import ... | closed | completed | false | 6 | [
"bug"
] | [] | 2022-12-22T20:08:36Z | 2022-12-28T09:48:21Z | 2022-12-28T09:48:21Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | apolinario | 788,417 | MDQ6VXNlcjc4ODQxNw== | User | false |
huggingface/diffusers | 1,508,606,349 | I_kwDOHa8MBc5Z64GN | 1,812 | https://github.com/huggingface/diffusers/issues/1812 | https://api.github.com/repos/huggingface/diffusers/issues/1812 | Paint By Example Fine-tuning | Is it possible to have a diffuser training implementation of paint by example pipeline? Should unlock, image conditioning for inpainting. Looking forward to contributing.
| closed | completed | false | 2 | [
"stale"
] | [] | 2022-12-22T22:06:34Z | 2023-02-04T15:02:59Z | 2023-02-04T15:02:59Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | kukeshajanth | 46,631,207 | MDQ6VXNlcjQ2NjMxMjA3 | User | false |
huggingface/diffusers | 1,508,868,086 | I_kwDOHa8MBc5Z73_2 | 1,813 | https://github.com/huggingface/diffusers/issues/1813 | https://api.github.com/repos/huggingface/diffusers/issues/1813 | train_text_to_image.py --resume_from_checkpoint logic is broken. | ### Describe the bug
Problem 1
---
An attempt to resume from a checkpoint *and* push the result to the hub fails:
```
accelerate train_text_to_image.py \
...
--resume_from_checkpoint="latest" \
--push_to_hub \
--hub_model_id="myusername/mymodel"
```
Error message:
```
OSError: Tried to clone a repository i... | closed | completed | false | 6 | [
"bug",
"stale"
] | [] | 2022-12-23T04:22:59Z | 2023-03-03T15:03:56Z | 2023-03-03T15:03:56Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | iuliaturc | 6,214,514 | MDQ6VXNlcjYyMTQ1MTQ= | User | false |
huggingface/diffusers | 1,509,071,911 | I_kwDOHa8MBc5Z8pwn | 1,814 | https://github.com/huggingface/diffusers/issues/1814 | https://api.github.com/repos/huggingface/diffusers/issues/1814 | support for change CLIP model. | I have tried changing the default CLIP model to chinese clip model, code like this:
```
import torch
from diffusers import StableDiffusionPipeline
from transformers import ChineseCLIPTextModel
pipe = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16)
pipe.text_e... | closed | completed | false | 2 | [
"stale"
] | [] | 2022-12-23T08:51:13Z | 2023-02-04T15:02:58Z | 2023-02-04T15:02:58Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | CrazyBoyM | 35,400,185 | MDQ6VXNlcjM1NDAwMTg1 | User | false |
huggingface/diffusers | 1,509,208,252 | I_kwDOHa8MBc5Z9LC8 | 1,815 | https://github.com/huggingface/diffusers/issues/1815 | https://api.github.com/repos/huggingface/diffusers/issues/1815 | --train-text-encoder not working with Deepspeed | ### Describe the bug
Hi, I try to implement the DreamBooth [tutorial](https://huggingface.co/docs/diffusers/v0.11.0/en/training/dreambooth#inference) and when I am adding parameter --train-text-encoder, pipeline run with error.
### Reproduction
accelerate launch diffusers/examples/dreambooth/train_dreambooth.py ... | closed | completed | false | 7 | [
"bug",
"stale"
] | [] | 2022-12-23T11:01:20Z | 2023-03-24T15:55:30Z | 2023-02-05T15:02:59Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | OleksandrKorovii | 107,404,835 | U_kgDOBmbeIw | User | false |
huggingface/diffusers | 1,509,542,275 | I_kwDOHa8MBc5Z-cmD | 1,817 | https://github.com/huggingface/diffusers/issues/1817 | https://api.github.com/repos/huggingface/diffusers/issues/1817 | [Dreambooth] TypeError: get_scheduler() got an unexpected keyword argument 'num_cycles' | ### Describe the bug
Hello,
When I run diffusers/examples/dreambooth/train_dreambooth.py, I get the following error:
```
│ 590 │ │ args.max_train_steps = args.num_train_epochs * num_update_step │
│ 591 │ │ overrode_max_train_steps = True │
│ 592 │ ... | closed | completed | false | 2 | [
"bug"
] | [] | 2022-12-23T16:41:36Z | 2022-12-29T09:43:32Z | 2022-12-29T09:43:31Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | claudiogreco | 7,571,915 | MDQ6VXNlcjc1NzE5MTU= | User | false |
huggingface/diffusers | 1,509,778,956 | I_kwDOHa8MBc5Z_WYM | 1,818 | https://github.com/huggingface/diffusers/issues/1818 | https://api.github.com/repos/huggingface/diffusers/issues/1818 | `--extract_ema` has no effect when converting .ckpt to diffusers | There's a bug in `scripts/convert_original_stable_diffusion_to_diffusers.py` that causes the `--extract_ema` flag to have no effect, so it always uses the non-EMA weights. I converted the sd-v1-4 ckpt to diffusers format both with and without the flag. `diff` reports both directories are identical.
```
$ diff -s sd... | closed | completed | false | 4 | [] | [] | 2022-12-23T21:07:27Z | 2023-01-10T22:54:40Z | 2023-01-10T22:54:40Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | feffy380 | 114,889,020 | U_kgDOBtkRPA | User | false |
huggingface/diffusers | 1,509,825,961 | I_kwDOHa8MBc5Z_h2p | 1,819 | https://github.com/huggingface/diffusers/issues/1819 | https://api.github.com/repos/huggingface/diffusers/issues/1819 | image sources for training examples for validation/comparison (data provided) | Below I've attached a list of image sources used for the Dreambooth and Textual Inversion paper, I've also included images provided by Joe Penna to be used for comparing training on a person (Joe Penna has given permission for his images to be used for this purpose).
I'd like to have this added to the examples fold... | closed | completed | false | 3 | [
"stale"
] | [] | 2022-12-23T22:39:30Z | 2023-02-04T15:02:57Z | 2023-02-04T15:02:57Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Thomas-MMJ | 112,830,596 | U_kgDOBrmohA | User | false |
huggingface/diffusers | 1,509,903,262 | I_kwDOHa8MBc5Z_0ue | 1,820 | https://github.com/huggingface/diffusers/issues/1820 | https://api.github.com/repos/huggingface/diffusers/issues/1820 | accelerator.save_state() get stuck When training with Multi-GPUs | ### Describe the bug
accelerator.save_state() get stuck When training with Multi-GPUs.
but it works well when training with a single GPU, how to solve this problem?
### Reproduction
if accelerator.is_main_process:
save_path = os.path.join(args.ckpt_path, f"checkpoint-{g... | closed | completed | false | 9 | [
"bug",
"stale"
] | [] | 2022-12-24T01:55:57Z | 2023-03-08T14:44:39Z | 2023-02-08T15:03:07Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | StrugglingForBetter | 20,771,340 | MDQ6VXNlcjIwNzcxMzQw | User | false |
huggingface/diffusers | 1,510,153,842 | I_kwDOHa8MBc5aAx5y | 1,821 | https://github.com/huggingface/diffusers/issues/1821 | https://api.github.com/repos/huggingface/diffusers/issues/1821 | The negative_prompt is having little to no effect with the inpaint pipeline. | ### Describe the bug
It seems like `negative_prompt` is having little to no effect on the inpainting pipeline. Demonstrated with the example below.
Input:
Mask:
![image](https://user-images.githu... | closed | completed | false | 3 | [
"bug",
"stale"
] | [] | 2022-12-24T17:15:43Z | 2023-02-07T15:03:32Z | 2023-02-07T15:03:32Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | radi-cho | 12,954,909 | MDQ6VXNlcjEyOTU0OTA5 | User | false |
huggingface/diffusers | 1,510,173,753 | I_kwDOHa8MBc5aA2w5 | 1,822 | https://github.com/huggingface/diffusers/issues/1822 | https://api.github.com/repos/huggingface/diffusers/issues/1822 | Documentation for Stable Diffusion Image Variations Pipeline | I think we should add it to the pipeline README and to the docs. There's a space here: https://huggingface.co/spaces/lambdalabs/stable-diffusion-image-variations | closed | completed | false | 1 | [] | [] | 2022-12-24T18:59:32Z | 2022-12-24T19:00:24Z | 2022-12-24T19:00:24Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | pcuenca | 1,177,582 | MDQ6VXNlcjExNzc1ODI= | User | false |
huggingface/diffusers | 1,510,177,594 | I_kwDOHa8MBc5aA3s6 | 1,823 | https://github.com/huggingface/diffusers/issues/1823 | https://api.github.com/repos/huggingface/diffusers/issues/1823 | Comments swapped in Stable Diffusion depth2img | ### Describe the bug
In [pipeline_stable_diffusion_depth2img.py](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_depth2img.py), lines 541 and 531 have swapped labels in the comments.
### Reproduction
-, mistake in the code
### Logs
_No response... | closed | completed | false | 2 | [
"bug"
] | [] | 2022-12-24T19:24:07Z | 2023-01-03T15:24:37Z | 2023-01-03T15:24:37Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | neverix | 46,641,404 | MDQ6VXNlcjQ2NjQxNDA0 | User | false |
huggingface/diffusers | 1,510,241,806 | I_kwDOHa8MBc5aBHYO | 1,825 | https://github.com/huggingface/diffusers/issues/1825 | https://api.github.com/repos/huggingface/diffusers/issues/1825 | Stable Diffusion Img2Img with a mask | The masking on inpainting mode works perfect. However, the issue is, that you always need a dedicated inpainting checkpoint for inpainting. This checkpoint must have a denoiser with (9, W, H) input shape instead of (5, W, H).
For many custom models such a checkpoint does not exist. Other repositories like [stable-d... | closed | completed | false | 4 | [] | [] | 2022-12-25T03:57:42Z | 2024-03-06T05:05:25Z | 2023-01-03T15:57:42Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Mwni | 13,929,820 | MDQ6VXNlcjEzOTI5ODIw | User | false |
huggingface/diffusers | 1,510,452,358 | I_kwDOHa8MBc5aB6yG | 1,829 | https://github.com/huggingface/diffusers/issues/1829 | https://api.github.com/repos/huggingface/diffusers/issues/1829 | Best way to work around xformers backwards K limit on sm80/86/89 devices | XFormers currently can't run backwards on many Sm80/86/89 devices. It runs fine at any size on A100s, but on other devices in the class (including common ones like 3080, 3090, 4090) it only works backwards when dim_heads <= 64.
This is a pretty narrow use case (most people don't run backwards as standard inference d... | closed | completed | false | 7 | [
"stale"
] | [] | 2022-12-26T00:22:43Z | 2023-03-08T13:49:53Z | 2023-03-07T15:04:07Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | hafriedlander | 114,769 | MDQ6VXNlcjExNDc2OQ== | User | false |
huggingface/diffusers | 1,340,050,674 | I_kwDOHa8MBc5P34zy | 183 | https://github.com/huggingface/diffusers/issues/183 | https://api.github.com/repos/huggingface/diffusers/issues/183 | Allow non-default schedulers to be easily swapped into `DiffusionPipeline` classes | **Is your feature request related to a problem? Please describe.**
By default the stable diffusion pipeline uses the PNDM scheduler, but one could easily use other schedulers (we only need to overwrite the `self.scheduler`) attribute.
This can be done with the following code-snippet:
```python
pipe = StableDi... | closed | completed | false | 19 | [
"good first issue"
] | [
"apolinario",
"pcuenca",
"natolambert",
"anton-l",
"patil-suraj"
] | 2022-08-16T09:13:51Z | 2022-12-31T08:06:29Z | 2022-08-31T08:35:18Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/diffusers | 1,510,456,652 | I_kwDOHa8MBc5aB71M | 1,830 | https://github.com/huggingface/diffusers/issues/1830 | https://api.github.com/repos/huggingface/diffusers/issues/1830 | Better Cross Attention Support For Graphcore IPUs | Hello. In Graphcore's [Stable Diffusion demo](https://ipu.dev/3FzCfsz), it seems they replace [your cross attention implementation](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/cross_attention.py
) with their own, to work around some issues:
https://github.com/gradient-ai/Graphcore-Huggi... | closed | completed | false | 9 | [
"stale"
] | [] | 2022-12-26T00:35:49Z | 2023-02-11T15:03:04Z | 2023-02-11T15:03:04Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | brucethemoose | 8,422,224 | MDQ6VXNlcjg0MjIyMjQ= | User | false |
huggingface/diffusers | 1,510,806,626 | I_kwDOHa8MBc5aDRRi | 1,831 | https://github.com/huggingface/diffusers/issues/1831 | https://api.github.com/repos/huggingface/diffusers/issues/1831 | [Dreambooth] Precision mismatch when generating class images | ### Describe the bug
Hello!
I was trying to finetune Anything 3.0 and when I was generating the class images I got precision mismatch error (see log below). I already had float16 errors on my GPU, so I set `mixed_precision` to false and it worked with *pre-generated* class images, but not when the pipeline generate... | closed | completed | false | 0 | [
"bug"
] | [] | 2022-12-26T10:01:49Z | 2022-12-27T18:51:33Z | 2022-12-27T18:51:33Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | kabachuha | 14,872,007 | MDQ6VXNlcjE0ODcyMDA3 | User | false |
huggingface/diffusers | 1,511,177,268 | I_kwDOHa8MBc5aErw0 | 1,837 | https://github.com/huggingface/diffusers/issues/1837 | https://api.github.com/repos/huggingface/diffusers/issues/1837 | Adding image logging to textual_inversion script | **Is your feature request related to a problem? Please describe.**
Currently, the textual inversion script just runs the full course without any logging of images and no way for the users to see what kind of images it will produce.
**Describe the solution you'd like**
At every --log_frequency steps, make images lo... | closed | completed | false | 4 | [
"stale"
] | [] | 2022-12-26T19:20:46Z | 2023-02-19T15:17:45Z | 2023-02-19T15:17:45Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | isamu-isozaki | 23,430,101 | MDQ6VXNlcjIzNDMwMTAx | User | false |
huggingface/diffusers | 1,340,054,131 | I_kwDOHa8MBc5P35pz | 184 | https://github.com/huggingface/diffusers/issues/184 | https://api.github.com/repos/huggingface/diffusers/issues/184 | Split `tests_modeling_utils` | The function https://github.com/huggingface/diffusers/blob/main/tests/test_modeling_utils.py becomes too long which makes it a bit cumbersome to only test part of the code.
We should at least move all the pipeline tests into a new class and probably also also move the individual model testing classes, such as https... | closed | completed | false | 0 | [
"good first issue"
] | [] | 2022-08-16T09:16:40Z | 2022-08-24T11:27:17Z | 2022-08-24T11:27:17Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/diffusers | 1,511,427,646 | I_kwDOHa8MBc5aFo4- | 1,841 | https://github.com/huggingface/diffusers/issues/1841 | https://api.github.com/repos/huggingface/diffusers/issues/1841 | Generation works correctly without ckpt files | If I do not place ckeckpoint files(e.g. v2-1_768-ema-pruned.ckpt) in the model folder, generation works as usual.
I used to think ckpt files save the model weights after training on LAION-5B. If there is no ckpt file, the model needs to inference from scratch and produce images with low quality.
However, the quality... | closed | completed | false | 3 | [] | [] | 2022-12-27T04:39:01Z | 2022-12-30T18:14:13Z | 2022-12-30T18:14:13Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | chesha1 | 49,867,392 | MDQ6VXNlcjQ5ODY3Mzky | User | false |
huggingface/diffusers | 1,511,534,741 | I_kwDOHa8MBc5aGDCV | 1,842 | https://github.com/huggingface/diffusers/issues/1842 | https://api.github.com/repos/huggingface/diffusers/issues/1842 | [Error] got an unexpected keyword argument `eta` | ### Describe the bug
When I try to make a sampling using DDIM pipeline, an error occurs
```bash
python utils/ddim.py
0%| | 0/50 [00:00<?, ?it/s]
Tra... | closed | completed | false | 3 | [
"bug"
] | [] | 2022-12-27T07:51:20Z | 2023-01-10T13:52:34Z | 2023-01-10T13:52:34Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | mireiffe | 47,289,460 | MDQ6VXNlcjQ3Mjg5NDYw | User | false |
huggingface/diffusers | 1,511,541,893 | I_kwDOHa8MBc5aGEyF | 1,843 | https://github.com/huggingface/diffusers/issues/1843 | https://api.github.com/repos/huggingface/diffusers/issues/1843 | Training multiple faces concepts with single model | Hi I tried to train diffusion model on multiple concepts with code from this repo. My idea was to create a script that takes in a directory full of labeled images like actors' names or character names. Then I run this script on a video to create directories full of frames where the actor/ character appears. Now I can ... | closed | completed | false | 2 | [
"stale"
] | [] | 2022-12-27T08:01:49Z | 2023-02-04T15:02:53Z | 2023-02-04T15:02:53Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Abhi001vj | 9,322,388 | MDQ6VXNlcjkzMjIzODg= | User | false |
huggingface/diffusers | 1,511,944,420 | I_kwDOHa8MBc5aHnDk | 1,845 | https://github.com/huggingface/diffusers/issues/1845 | https://api.github.com/repos/huggingface/diffusers/issues/1845 | Providing custom depth maps for depth2img | **What API design would you like to have changed or added to the library? Why?**
Currently there's an inconsistency with image handling around depth maps. While you can provide either a PIL image or tensor data for the main image in img2img and depth2img pipelines, currently only tensor data is accepted in depth2img.
... | closed | completed | false | 4 | [
"stale"
] | [
"patil-suraj"
] | 2022-12-27T16:07:13Z | 2023-02-27T15:04:00Z | 2023-02-27T15:04:00Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | subpanic | 120,813,334 | U_kgDOBzN3Fg | User | false |
huggingface/diffusers | 1,512,287,266 | I_kwDOHa8MBc5aI6wi | 1,847 | https://github.com/huggingface/diffusers/issues/1847 | https://api.github.com/repos/huggingface/diffusers/issues/1847 | Post choosing accerlate config-- Which type of machine are you using-- What do I choose? | > Please select a choice using the arrow or number keys, and selecting with enter
No distributed trainingg
multi-CPU
* multi-GPU
TPU
MPS
I have a normal computer, an rtx 2060 and a normal cpu, core i-5 11400F, and I have no idea what TPU and MPS stands for, but from google searches both seem lik... | closed | completed | false | 5 | [
"stale"
] | [] | 2022-12-28T00:38:09Z | 2023-02-04T15:02:51Z | 2023-02-04T15:02:51Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Jcdude105 | 118,931,007 | U_kgDOBxa-Pw | User | false |
huggingface/diffusers | 1,513,192,168 | I_kwDOHa8MBc5aMXro | 1,851 | https://github.com/huggingface/diffusers/issues/1851 | https://api.github.com/repos/huggingface/diffusers/issues/1851 | Dreambooth example with multi-gpu significantly slower than single GPU | ### Describe the bug
If the `accelerate config` is setup for multi-GPU (default config works) then training speed appears to dramatically slow. With a 2x 3090 system I'm seeing training go from ~2 it/s with single 3090 configured to ~6 s/it with multi-GPU configured.
I'm guessing I'm either misunderstanding what sh... | closed | completed | false | 8 | [
"bug",
"stale"
] | [
"williamberman",
"patil-suraj"
] | 2022-12-28T21:49:35Z | 2023-02-25T15:04:09Z | 2023-02-25T15:04:09Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | subpanic | 120,813,334 | U_kgDOBzN3Fg | User | false |
huggingface/diffusers | 1,513,232,573 | I_kwDOHa8MBc5aMhi9 | 1,852 | https://github.com/huggingface/diffusers/issues/1852 | https://api.github.com/repos/huggingface/diffusers/issues/1852 | please delete | Please close this, I meant to report this to a forked project but it went here instead. | closed | completed | false | 0 | [] | [] | 2022-12-28T23:15:47Z | 2022-12-29T12:43:10Z | 2022-12-29T12:43:10Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | David-Hari | 7,530,738 | MDQ6VXNlcjc1MzA3Mzg= | User | false |
huggingface/diffusers | 1,513,235,904 | I_kwDOHa8MBc5aMiXA | 1,853 | https://github.com/huggingface/diffusers/issues/1853 | https://api.github.com/repos/huggingface/diffusers/issues/1853 | Can't paste or even type my token for logging into huggingface from the commandline? | Hi, for context, I'm trying to follow this [guide](https://github.com/ShivamShrirao/diffusers/tree/main/examples/dreambooth) in order to make dreambooth require less vram since I only have an rtx 2060 6gb and I got stuck following the guide. I used to use gitbash for my commands before I switched to windows command bec... | closed | completed | false | 3 | [
"stale"
] | [] | 2022-12-28T23:25:12Z | 2023-02-07T15:03:26Z | 2023-02-07T15:03:26Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Jcdude105 | 118,931,007 | U_kgDOBxa-Pw | User | false |
huggingface/diffusers | 1,513,395,754 | I_kwDOHa8MBc5aNJYq | 1,854 | https://github.com/huggingface/diffusers/issues/1854 | https://api.github.com/repos/huggingface/diffusers/issues/1854 | Custom pipeline as attribute | Right now we need to load custom pipelines as parameter in DiffusionPipeline
`pipe = DiffusionPipeline.from_pretrained(model_id, custom_pipeline="lpw_stable_diffusion", revision="fp16", torch_dtype=torch.float16)`
But, can we pass custom_pipeline as attribute? this will allow changing pipelines easily without lo... | closed | completed | false | 6 | [] | [] | 2022-12-29T05:40:15Z | 2023-01-26T15:45:58Z | 2023-01-26T15:45:19Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | adhikjoshi | 11,740,719 | MDQ6VXNlcjExNzQwNzE5 | User | false |
huggingface/diffusers | 1,513,480,424 | I_kwDOHa8MBc5aNeDo | 1,855 | https://github.com/huggingface/diffusers/issues/1855 | https://api.github.com/repos/huggingface/diffusers/issues/1855 | Should I do 'accelerator.prepare()' again after 'accelerator.unwrap_model()' ? | For multi-gpu training:
`model = accelerator.prepare(model)`
I want to save ckpt during training. Then I do this:
`model = accelerator.unwrap_model(model)`
`pipeline = Pipeline(model=model)`
`pipeline.save_pretrained(...)`
And I want to continue training. Should I do `model = accelerator.prepare(model)` again aft... | closed | completed | false | 3 | [
"stale"
] | [] | 2022-12-29T07:55:07Z | 2023-02-24T15:03:58Z | 2023-02-24T15:03:58Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | DtYXs | 59,554,522 | MDQ6VXNlcjU5NTU0NTIy | User | false |
huggingface/diffusers | 1,513,703,173 | I_kwDOHa8MBc5aOUcF | 1,857 | https://github.com/huggingface/diffusers/issues/1857 | https://api.github.com/repos/huggingface/diffusers/issues/1857 | UnCLIP Pipeline is currently poorly tested | ### Describe the bug
The UnCLIP pipeline, like any new pipeline in `src/pipelines` should depend on the tester mixin: https://github.com/huggingface/diffusers/blob/debc74f442dc74210528eb6d8a4d1f7f27fa18c3/tests/test_pipelines_common.py#L35 just like Stable Diffusion does: https://github.com/huggingface/diffusers/blob/... | closed | completed | false | 2 | [
"bug"
] | [
"williamberman"
] | 2022-12-29T12:46:57Z | 2023-01-23T19:36:41Z | 2023-01-23T19:36:41Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/diffusers | 1,514,023,211 | I_kwDOHa8MBc5aPikr | 1,863 | https://github.com/huggingface/diffusers/issues/1863 | https://api.github.com/repos/huggingface/diffusers/issues/1863 | Stable Diffusion 2.x -- Wrong CLIP layer is being used for conditioning? | ### Describe the bug
Looking at the ["official" SD implementation](https://github.com/Stability-AI/stablediffusion), we can see in the end of the inference config file the following lines:
https://github.com/Stability-AI/stablediffusion/blob/d55bcd4d31d0316fcbdf552f2fd2628fdc812500/configs/stable-diffusion/v2-inf... | closed | completed | false | 5 | [
"bug"
] | [] | 2022-12-29T20:11:24Z | 2023-01-05T20:58:20Z | 2022-12-30T11:47:14Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | jorgemcgomes | 3,987,574 | MDQ6VXNlcjM5ODc1NzQ= | User | false |
huggingface/diffusers | 1,514,094,081 | I_kwDOHa8MBc5aPz4B | 1,864 | https://github.com/huggingface/diffusers/issues/1864 | https://api.github.com/repos/huggingface/diffusers/issues/1864 | EMA model in Stable Diffusion training example is wrong | ### Describe the bug
The step call is updating self.decay, which will then be used to compute the step-dependent-decay in the next steps, resulting in a bit of a mess that makes EMA not really work as it should.
https://github.com/huggingface/diffusers/blob/ab0e92fdc8b42e1739840d10bb0b1dc7edaf2ff3/examples/text_t... | closed | completed | false | 1 | [
"bug"
] | [] | 2022-12-29T22:31:55Z | 2022-12-30T11:42:43Z | 2022-12-30T11:42:43Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | jorgemcgomes | 3,987,574 | MDQ6VXNlcjM5ODc1NzQ= | User | false |
huggingface/diffusers | 1,514,193,844 | I_kwDOHa8MBc5aQMO0 | 1,865 | https://github.com/huggingface/diffusers/issues/1865 | https://api.github.com/repos/huggingface/diffusers/issues/1865 | using zero stage3 finetune sd2, dimension error occurs | ### Describe the bug
An error is reported when using deepspeed's zero stage3 finetune diffusers/examples/text_to_image/train_text_to_image.py script. My machine's GPU is 4*2080ti, and because a single GPU cannot accommodate all SD2 parameters, the deepspeed zero stage3 strategy must be used.
### Reproduction
a... | closed | completed | false | 5 | [
"bug",
"stale"
] | [] | 2022-12-30T02:35:58Z | 2023-05-11T15:59:22Z | 2023-05-11T15:59:22Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Nipi64310 | 31,469,418 | MDQ6VXNlcjMxNDY5NDE4 | User | false |
huggingface/diffusers | 1,514,316,584 | I_kwDOHa8MBc5aQqMo | 1,866 | https://github.com/huggingface/diffusers/issues/1866 | https://api.github.com/repos/huggingface/diffusers/issues/1866 | Stable Diffusion img2img: DPMSolverSinglestepScheduler does not work with certain strength values | ### Describe the bug
When using `StableDiffusionImg2ImgPipeline`, the `DPMSolverSinglestepScheduler` schedule, with certain img2img strength values, produces an error resulting from previous model outputs not being available.
### Settings:
Inference steps: 15
Strength: 0.1, 0.3, 0.4, 0.45, 0.55, 0.7, 0.8, 0.95... | closed | completed | false | 9 | [
"bug",
"wip"
] | [
"patrickvonplaten"
] | 2022-12-30T07:42:31Z | 2024-11-21T15:30:15Z | 2024-11-21T15:30:15Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | cadaeix | 26,790,650 | MDQ6VXNlcjI2NzkwNjUw | User | false |
huggingface/diffusers | 1,514,349,637 | I_kwDOHa8MBc5aQyRF | 1,867 | https://github.com/huggingface/diffusers/issues/1867 | https://api.github.com/repos/huggingface/diffusers/issues/1867 | Stable Diffusion img2img: img2img pipeline does not appear to work at below 0.1 strength with any scheduler | ### Describe the bug
With `StableDiffusionImg2ImgPipeline`, all the schedulers I have tested in the course of testing #1866 error out when the img2img strength is 0.0 or 0.05 during the NSFW safety checker
While investigating this issue, a local copy of a community pipeline (`StableDiffusionLongPromptWeightingPip... | closed | completed | false | 2 | [
"bug",
"stale"
] | [] | 2022-12-30T08:45:53Z | 2023-02-07T15:03:23Z | 2023-02-07T15:03:23Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | cadaeix | 26,790,650 | MDQ6VXNlcjI2NzkwNjUw | User | false |
huggingface/diffusers | 1,514,456,167 | I_kwDOHa8MBc5aRMRn | 1,869 | https://github.com/huggingface/diffusers/issues/1869 | https://api.github.com/repos/huggingface/diffusers/issues/1869 | [Community Pipeline] UnCLIP image / text interpolations | ### Model/Pipeline/Scheduler description
Copied from https://github.com/huggingface/diffusers/pull/1858:
> UnCLIP / Karlo: https://huggingface.co/spaces/kakaobrain/karlo gives some very nice and precise results when doing image generation and can strongly outperform Stable Diffusion in some - see:
> https://www.re... | closed | completed | false | 49 | [
"stale"
] | [
"williamberman",
"patrickvonplaten"
] | 2022-12-30T11:28:32Z | 2023-07-24T15:04:17Z | 2023-07-24T15:04:17Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/diffusers | 1,514,903,174 | I_kwDOHa8MBc5aS5aG | 1,872 | https://github.com/huggingface/diffusers/issues/1872 | https://api.github.com/repos/huggingface/diffusers/issues/1872 | Stable Diffusion `img2img` Pipeline does not check image type | It would be better if `img2img` pipeline checks if the type of `image` is valid. Currently, this pipeline only supports `tensor`, `PIL.Image`, or a list of `PIL.Image`, but this type check is missing in https://github.com/huggingface/diffusers/blob/62608a9102e423ad0fe79f12a8ceb1710d2027b2/src/diffusers/pipelines/stable... | closed | completed | false | 1 | [] | [] | 2022-12-30T23:13:00Z | 2023-01-05T21:32:41Z | 2023-01-05T21:32:30Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | tsunrise | 22,125,551 | MDQ6VXNlcjIyMTI1NTUx | User | false |
huggingface/diffusers | 1,515,078,646 | I_kwDOHa8MBc5aTkP2 | 1,873 | https://github.com/huggingface/diffusers/issues/1873 | https://api.github.com/repos/huggingface/diffusers/issues/1873 | Is there a way when calling .from_pretrained() to download either the safetensor or original instead of both? | Not a massive issue, but I'm pretty low on storage space on my device. I'm fine with either version, just both eats into my storage and internet plan quickly | closed | completed | false | 5 | [
"stale"
] | [] | 2022-12-31T09:36:12Z | 2023-03-04T15:04:01Z | 2023-03-04T15:04:01Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ethansmith2000 | 98,723,285 | U_kgDOBeJl1Q | User | false |
huggingface/diffusers | 1,515,257,508 | I_kwDOHa8MBc5aUP6k | 1,874 | https://github.com/huggingface/diffusers/issues/1874 | https://api.github.com/repos/huggingface/diffusers/issues/1874 | FlaxUNet2DConditionModel very slow to JIT due to huge jaxpr | Even this trivial example of `FlaxUNet2DConditionModel` with a single block produces 3,319 lines of jaxpr, leading to compile times in the 10s of seconds.
```
model = diffusers.FlaxUNet2DConditionModel(
sample_size = 32,
in_channels = 4,
out_channels = 4,
down_block_types = ("DownBlock2D"),
... | closed | completed | false | 8 | [
"stale"
] | [] | 2023-01-01T01:13:24Z | 2023-02-20T15:03:11Z | 2023-02-20T15:03:11Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | SamPruden | 19,550,501 | MDQ6VXNlcjE5NTUwNTAx | User | false |
huggingface/diffusers | 1,515,507,215 | I_kwDOHa8MBc5aVM4P | 1,875 | https://github.com/huggingface/diffusers/issues/1875 | https://api.github.com/repos/huggingface/diffusers/issues/1875 | Imagic pipeline training on fp16 fails with: Input type (float) and bias type (c10::Half) should be the same | ### Describe the bug
I get an out of memory when trying to run the Imagic stable diffusion pipeline on Google Colab. So I tried to use half precision by passing `revision='fp16'` and `torch_dtype=torch.float16` when creating the pipeline, but now the training fails with
```
RuntimeError: Input type (float) and bias ... | closed | completed | false | 3 | [
"bug",
"stale"
] | [] | 2023-01-01T11:52:06Z | 2023-02-19T15:02:50Z | 2023-02-19T15:02:50Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | dzlab | 1,645,304 | MDQ6VXNlcjE2NDUzMDQ= | User | false |
huggingface/diffusers | 1,515,523,931 | I_kwDOHa8MBc5aVQ9b | 1,876 | https://github.com/huggingface/diffusers/issues/1876 | https://api.github.com/repos/huggingface/diffusers/issues/1876 | Some checkpoints have no state_dict and cannot be converted to diffusers. | ### Describe the bug
I downloaded two popular model from civitai and both of them has no `state_dict`
I put them here:
https://huggingface.co/ckpt/Protogen_V2.2/blob/main/Protogen_V2.2.ckpt
https://huggingface.co/ckpt/elldrethSLucidMix_v10/blob/main/elldrethSLucidMix_v10.ckpt
inside elldrethSLucidMix_v10.... | closed | completed | false | 2 | [
"bug"
] | [] | 2023-01-01T12:15:17Z | 2023-01-01T16:22:32Z | 2023-01-01T16:22:32Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | camenduru | 54,370,274 | MDQ6VXNlcjU0MzcwMjc0 | User | false |
huggingface/diffusers | 1,515,579,801 | I_kwDOHa8MBc5aVemZ | 1,877 | https://github.com/huggingface/diffusers/issues/1877 | https://api.github.com/repos/huggingface/diffusers/issues/1877 | How to generate ckpt and yaml file from diffuser model? - Error(s) in loading state_dict for LatentDiffusion: size mismatch for model.diffusion_model.input_blocks. | I have generated ckpt file from this repo : https://huggingface.co/lambdalabs/dreambooth-avatar
The file is generated
I have used this command
`python convert_diffusers_to_original_stable_diffusion.py --model_path "C:\dreambooth-avatar" --checkpoint_path "C:\Users\King\Documents\PythonApplication1\dreambooth_a... | closed | completed | false | 13 | [
"stale"
] | [] | 2023-01-01T13:49:15Z | 2023-02-08T15:02:58Z | 2023-02-08T15:02:58Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | FurkanGozukara | 19,240,467 | MDQ6VXNlcjE5MjQwNDY3 | User | false |
huggingface/diffusers | 1,515,745,647 | I_kwDOHa8MBc5aWHFv | 1,880 | https://github.com/huggingface/diffusers/issues/1880 | https://api.github.com/repos/huggingface/diffusers/issues/1880 | Refactor `attention.py` | `attention.py` has at the moment two concurrent attention implementations which essentially do the exact same thing:
- https://github.com/huggingface/diffusers/blob/62608a9102e423ad0fe79f12a8ceb1710d2027b2/src/diffusers/models/attention.py#L256
and
- https://github.com/huggingface/diffusers/blob/62608a9102e423ad0... | closed | completed | false | 16 | [
"wip"
] | [
"williamberman"
] | 2023-01-01T18:16:15Z | 2023-05-31T17:31:04Z | 2023-05-31T17:31:03Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/diffusers | 1,515,826,608 | I_kwDOHa8MBc5aWa2w | 1,881 | https://github.com/huggingface/diffusers/issues/1881 | https://api.github.com/repos/huggingface/diffusers/issues/1881 | Tried converting a diffuser to onnx, AMD GPU? | 
| closed | completed | false | 2 | [
"stale"
] | [] | 2023-01-01T20:54:52Z | 2023-02-09T15:03:33Z | 2023-02-09T15:03:33Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | DOUBLEXX666 | 66,040,365 | MDQ6VXNlcjY2MDQwMzY1 | User | false |
huggingface/diffusers | 1,515,940,439 | I_kwDOHa8MBc5aW2pX | 1,882 | https://github.com/huggingface/diffusers/issues/1882 | https://api.github.com/repos/huggingface/diffusers/issues/1882 | Stable Diffusion Flax Img2Img pipeline expects different image format input than output | ### Describe the bug
when running img2img pipeline, the outputs are all too light. i.e. nothing darker than middle gray. it's not very pronounced when using a higher/default "strength" value, but for use cases akin to style transfer where only minor noise is added to original image, it's very apparent. in other word... | closed | completed | false | 18 | [
"wip"
] | [
"yiyixuxu"
] | 2023-01-02T01:32:17Z | 2024-11-21T15:33:16Z | 2024-11-21T15:33:15Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | krahnikblis | 84,637,076 | MDQ6VXNlcjg0NjM3MDc2 | User | false |
huggingface/diffusers | 1,516,083,310 | I_kwDOHa8MBc5aXZhu | 1,883 | https://github.com/huggingface/diffusers/issues/1883 | https://api.github.com/repos/huggingface/diffusers/issues/1883 | Enabling gradient checkpointing for VAE in Stable Diffusion Pipeline/Model | **Is your feature request related to a problem? Please describe.**
For an experiment, I am interested in finetuning VAE present in stable diffusion pipeline. I dont have access to a GPU with large VRAM. Enabling gradient checkpointing for VAE model will help me finetune the model even with low GPU memory.
**Desc... | closed | completed | false | 4 | [
"good first issue"
] | [] | 2023-01-02T06:46:46Z | 2023-03-17T21:59:59Z | 2023-03-17T21:59:58Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | iamsimha | 3,105,044 | MDQ6VXNlcjMxMDUwNDQ= | User | false |
huggingface/diffusers | 1,516,386,259 | I_kwDOHa8MBc5aYjfT | 1,887 | https://github.com/huggingface/diffusers/issues/1887 | https://api.github.com/repos/huggingface/diffusers/issues/1887 | How can we use DPM++ 2M Karras sampler in Diffusers? | I'm trying to use the latest samplers which provide better performance with stable diffusion in the diffusers library, but I couldn't find DPM++ 2M Karras sampler in the library. These samplers are present in Automatic111. Can anyone guide me on how to use these samplers in the Diffusers library? | closed | completed | false | 10 | [
"stale"
] | [
"patrickvonplaten"
] | 2023-01-02T13:14:20Z | 2023-10-25T19:36:11Z | 2023-02-24T15:03:48Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | danishboy000 | 35,116,832 | MDQ6VXNlcjM1MTE2ODMy | User | false |
huggingface/diffusers | 1,516,649,644 | I_kwDOHa8MBc5aZjys | 1,889 | https://github.com/huggingface/diffusers/issues/1889 | https://api.github.com/repos/huggingface/diffusers/issues/1889 | DPM++ SDE variant | ### Model/Pipeline/Scheduler description
It looks like there is a variant of the very nice DPM++ scheduler on k-diffusion, see here: https://github.com/crowsonkb/k-diffusion/blob/5b3af030dd83e0297272d861c19477735d0317ec/k_diffusion/sampling.py#L543
It would be nice to add this one to `diffusers` as well.
Also g... | closed | completed | false | 4 | [
"New scheduler",
"wip"
] | [] | 2023-01-02T19:05:52Z | 2024-11-23T00:04:07Z | 2024-11-23T00:04:07Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/diffusers | 1,340,471,634 | I_kwDOHa8MBc5P5flS | 189 | https://github.com/huggingface/diffusers/issues/189 | https://api.github.com/repos/huggingface/diffusers/issues/189 | Question about PR #185 | Is it possible to use the K-LMS scheduler using the stable diffusion snapshot loaded with `from_pretrained`?
Thanks~ | closed | completed | false | 3 | [] | [] | 2022-08-16T14:46:49Z | 2022-08-16T15:30:15Z | 2022-08-16T15:26:10Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | atarashansky | 16,548,075 | MDQ6VXNlcjE2NTQ4MDc1 | User | false |
huggingface/diffusers | 1,516,822,094 | I_kwDOHa8MBc5aaN5O | 1,890 | https://github.com/huggingface/diffusers/issues/1890 | https://api.github.com/repos/huggingface/diffusers/issues/1890 | Does attention masking actually work? | I tried passing in an `attention_mask`, for use in a stable-diffusion Unet but it doesn't actually get passed down as deep as `CrossAttention#forward`.
I tried fixing it to pass the param down, but it blows up on tensor size mismatch, because self-attention and cross-attention have different masking requirements.
... | closed | completed | false | 17 | [
"stale"
] | [] | 2023-01-03T00:39:53Z | 2025-03-30T17:56:59Z | 2023-03-13T15:04:08Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Birch-san | 6,141,784 | MDQ6VXNlcjYxNDE3ODQ= | User | false |
huggingface/diffusers | 1,516,823,018 | I_kwDOHa8MBc5aaOHq | 1,891 | https://github.com/huggingface/diffusers/issues/1891 | https://api.github.com/repos/huggingface/diffusers/issues/1891 | Fuse the application of attention mask | When attention mask is applied: it does so by addition. But there's actually an addition we can get for free during the `baddbmm` above that we're not currently using.
I tried fusing the attention masking into the `baddbmm`:
https://github.com/Birch-san/diffusers/commit/f55dbf312f22fc514f7983599be88319d9b3e928
... | closed | completed | false | 6 | [] | [] | 2023-01-03T00:42:44Z | 2023-01-26T16:36:34Z | 2023-01-26T16:36:34Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Birch-san | 6,141,784 | MDQ6VXNlcjYxNDE3ODQ= | User | false |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.