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 | 1,799,433,645 | I_kwDOHa8MBc5rQS2t | 4,047 | https://github.com/huggingface/diffusers/issues/4047 | https://api.github.com/repos/huggingface/diffusers/issues/4047 | How to set lora scale when loading a LoRA model? | Hey there, first of all thanks for your fantastic work!
I am loading LoRA weights, and I would like to set the scale of them being applied. Checking the code, it appears to be possible as shown [here](https://github.com/huggingface/diffusers/blob/fc7aa64ea8f5979b67bd730777e8e1c32e3adb05/src/diffusers/loaders.py#L109... | closed | completed | false | 4 | [] | [] | 2023-07-11T17:38:05Z | 2023-08-29T05:30:44Z | 2023-07-11T22:51:36Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | pietrobolcato | 3,061,306 | MDQ6VXNlcjMwNjEzMDY= | User | false |
huggingface/diffusers | 1,365,057,407 | I_kwDOHa8MBc5RXR9_ | 405 | https://github.com/huggingface/diffusers/issues/405 | https://api.github.com/repos/huggingface/diffusers/issues/405 | single file or url is not supported for this tokenizer error | ### Describe the bug
Getting this error when running "setting up the model" cell: Calling CLIPTokenizer.from_pretrained() with the path to a single file or url is not supported for this tokenizer. Use a model identifier or the path to a directory instead
### Reproduction
#@title Load the tokenizer and add the placeh... | closed | completed | false | 5 | [
"bug"
] | [
"patil-suraj"
] | 2022-09-07T18:45:02Z | 2022-09-18T08:51:26Z | 2022-09-13T16:09:34Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | emmajane1313 | 72,898,465 | MDQ6VXNlcjcyODk4NDY1 | User | false |
huggingface/diffusers | 1,799,994,502 | I_kwDOHa8MBc5rSbyG | 4,052 | https://github.com/huggingface/diffusers/issues/4052 | https://api.github.com/repos/huggingface/diffusers/issues/4052 | has no attribute 'from_single_file' | ### Describe the bug
'StableDiffusionControlNetImg2ImgPipeline' has no attribute 'from_single_file'
ControlNetModel has no attribute 'from_single_file'
load single file is very important
pls
### Reproduction
controlnet = ControlNetModel.from_single_file
### Logs
_No response_
### System Info
0.18.2
### ... | closed | completed | false | 8 | [
"bug"
] | [] | 2023-07-12T01:13:05Z | 2025-08-19T20:22:26Z | 2023-07-21T05:18:12Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | zhaoyun0071 | 35,762,050 | MDQ6VXNlcjM1NzYyMDUw | User | false |
huggingface/diffusers | 1,800,105,776 | I_kwDOHa8MBc5rS28w | 4,053 | https://github.com/huggingface/diffusers/issues/4053 | https://api.github.com/repos/huggingface/diffusers/issues/4053 | Hardware auto-optimizing | **Is your feature request related to a problem? Please describe.**
The current Diffusers library does not provide an out-of-the-box functionality to _automatically_ apply system and model-level optimizations.
I often find myself manually tweaking system-level optimizations in PyTorch as well as model-specific ones ... | open | reopened | false | 4 | [
"wip"
] | [] | 2023-07-12T03:39:09Z | 2023-10-30T14:21:00Z | null | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | bghira | 59,658,056 | MDQ6VXNlcjU5NjU4MDU2 | User | false |
huggingface/diffusers | 1,800,120,586 | I_kwDOHa8MBc5rS6kK | 4,054 | https://github.com/huggingface/diffusers/issues/4054 | https://api.github.com/repos/huggingface/diffusers/issues/4054 | [SD-XL] How to apply invisible-watermark for latent output | ### Describe the bug
As a part of the license with SAI, we need to ensure the invisible watermark is applied across all images output by these models, including the Img2Img pipeline.
### Reproduction
```py
# if xformers or torch_2_0 is used attention block does not need
# to be in float32 which can... | closed | completed | false | 4 | [
"bug"
] | [] | 2023-07-12T03:58:04Z | 2023-07-12T10:21:29Z | 2023-07-12T10:21:29Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | bghira | 59,658,056 | MDQ6VXNlcjU5NjU4MDU2 | User | false |
huggingface/diffusers | 1,800,446,396 | I_kwDOHa8MBc5rUKG8 | 4,055 | https://github.com/huggingface/diffusers/issues/4055 | https://api.github.com/repos/huggingface/diffusers/issues/4055 | [Reproducibility] Calling a pipeline with the exact same inputs yields different results | If I do:
```python
from diffusers import DiffusionPipeline
import numpy as np
import torch
generator = torch.manual_seed(0)
dtype = torch.float16
pipeline = DiffusionPipeline.from_pretrained(
"runwayml/stable-diffusion-v1-5",
torch_dtype=dtype,
safety_checker=None
).to("cuda")
latents =... | closed | completed | false | 3 | [] | [] | 2023-07-12T08:22:41Z | 2023-07-12T16:06:58Z | 2023-07-12T09:39:32Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sayakpaul | 22,957,388 | MDQ6VXNlcjIyOTU3Mzg4 | User | false |
huggingface/diffusers | 1,800,855,367 | I_kwDOHa8MBc5rVt9H | 4,057 | https://github.com/huggingface/diffusers/issues/4057 | https://api.github.com/repos/huggingface/diffusers/issues/4057 | Blurred output from Finetuned Inpainting model | I am trying to finetune inpainting model on 5 images of dog. I have trained for 400, 1000, 1500 steps but at inference I get blurred output. What could have gone wrong?
!accelerate launch train_dreambooth_inpaint.py \
--pretrained_model_name_or_path="stabilityai/stable-diffusion-2-inpainting" \
--instance_dat... | closed | completed | false | 2 | [
"stale"
] | [] | 2023-07-12T12:16:01Z | 2023-08-20T15:03:28Z | 2023-08-20T15:03:28Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | shreyassks | 44,189,581 | MDQ6VXNlcjQ0MTg5NTgx | User | false |
huggingface/diffusers | 1,801,042,106 | I_kwDOHa8MBc5rWbi6 | 4,059 | https://github.com/huggingface/diffusers/issues/4059 | https://api.github.com/repos/huggingface/diffusers/issues/4059 | Found a bug in StableDiffusionControlNetReferencePipeline, and question about implementing StableDiffusionControlNetReferenceInpaintImg2ImgPipeline | ### Model/Pipeline/Scheduler description
@okotaku Hi, I'm trying to implement StableDiffusionControlNetReferenceInpaintImg2ImgPipeline following [StableDiffusionControlNetReferencePipeline](https://github.com/huggingface/diffusers/blob/main/examples/community/stable_diffusion_controlnet_reference.py) and [StableDiffus... | closed | completed | false | 3 | [
"stale"
] | [] | 2023-07-12T13:56:17Z | 2023-08-20T15:03:27Z | 2023-08-20T15:03:27Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | kunc01 | 18,576,899 | MDQ6VXNlcjE4NTc2ODk5 | User | false |
huggingface/diffusers | 1,801,101,546 | I_kwDOHa8MBc5rWqDq | 4,060 | https://github.com/huggingface/diffusers/issues/4060 | https://api.github.com/repos/huggingface/diffusers/issues/4060 | StableDiffusionXLEnsemblePipeline and StableDiffusionXLImg2ImgEnsemblePipeline | ### Model/Pipeline/Scheduler description
The current solution from PR #4015 does not result in an optimal solution for the use of the refinement model as an ensemble model because it works by a float value to determine the selection of timesteps to execute. This is further described in the commentary on the PR.
I a... | closed | completed | false | 6 | [
"stale"
] | [] | 2023-07-12T14:27:26Z | 2023-08-12T14:13:53Z | 2023-08-12T14:13:53Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | AmericanPresidentJimmyCarter | 110,263,573 | U_kgDOBpJ9FQ | User | false |
huggingface/diffusers | 1,801,112,866 | I_kwDOHa8MBc5rWs0i | 4,061 | https://github.com/huggingface/diffusers/issues/4061 | https://api.github.com/repos/huggingface/diffusers/issues/4061 | (SDXL) load lora error | ### Describe the bug
pipe.load_lora_weights(".", model_path)
File "/root/autodl-tmp/tools/demo/venv/lib/python3.10/site-packages/diffusers/configuration_utils.py", line 137, in __getattr__
raise AttributeError(f"'{type(self).__name__}' object has no attribute '{name}'")
AttributeError: 'StableDiffusionXLPip... | closed | completed | false | 13 | [
"bug",
"stale"
] | [] | 2023-07-12T14:33:27Z | 2023-08-21T15:03:45Z | 2023-08-21T15:03:45Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | zbulrush | 38,712,718 | MDQ6VXNlcjM4NzEyNzE4 | User | false |
huggingface/diffusers | 1,801,156,229 | I_kwDOHa8MBc5rW3aF | 4,063 | https://github.com/huggingface/diffusers/issues/4063 | https://api.github.com/repos/huggingface/diffusers/issues/4063 | Unloading a Textual Inversion Checkpoint | ### Model/Pipeline/Scheduler description
This is how I load a textual inversion checkpoint
```
pipeline.load_textual_inversion('myckpt.bin')
```
Is unloading implemented?
### Open source status
- [ ] The model implementation is available
- [ ] The model weights are available (Only relevant if addition is not... | closed | completed | false | 17 | [
"stale"
] | [] | 2023-07-12T14:51:05Z | 2024-02-23T21:31:19Z | 2024-02-23T21:31:19Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sarmientoj24 | 8,830,319 | MDQ6VXNlcjg4MzAzMTk= | User | false |
huggingface/diffusers | 1,801,399,562 | I_kwDOHa8MBc5rXy0K | 4,064 | https://github.com/huggingface/diffusers/issues/4064 | https://api.github.com/repos/huggingface/diffusers/issues/4064 | Add PSLD upscaler/deblur/inpainting Pipelines to Stable Diffusion and other LDMs | ### Model/Pipeline/Scheduler description
> We present the first framework to solve general inverse problems leveraging pre-trained latent diffusion models. Previously proposed algorithms (such as DPS and DDRM) only apply to pixel-space diffusion models. We theoretically analyze our algorithm showing provable sample ... | closed | completed | false | 2 | [
"community-examples",
"stale"
] | [] | 2023-07-12T17:06:02Z | 2023-08-22T15:04:11Z | 2023-08-22T15:04:11Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | AmericanPresidentJimmyCarter | 110,263,573 | U_kgDOBpJ9FQ | User | false |
huggingface/diffusers | 1,801,937,447 | I_kwDOHa8MBc5rZ2In | 4,067 | https://github.com/huggingface/diffusers/issues/4067 | https://api.github.com/repos/huggingface/diffusers/issues/4067 | Generation with lora_scale doesn't remain consistent when using prompt embeddings | ### Describe the bug
When using compel and prompt embeddings, and performing inference with LoRA weights loaded, `lora_scale` doesn't work as expected. Specifically, if I do the following actions in the following order:
1. Create a SD pipeline
2. Load a model
3. Load a LoRA
4. Generate an image with `lora_scal... | closed | completed | false | 5 | [
"bug"
] | [] | 2023-07-13T00:22:01Z | 2023-07-13T17:28:58Z | 2023-07-13T17:28:58Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | pietrobolcato | 3,061,306 | MDQ6VXNlcjMwNjEzMDY= | User | false |
huggingface/diffusers | 1,802,043,669 | I_kwDOHa8MBc5raQEV | 4,068 | https://github.com/huggingface/diffusers/issues/4068 | https://api.github.com/repos/huggingface/diffusers/issues/4068 | visualize the structure of the model |
What approach should I take to visualize the structure of the model? | closed | completed | false | 3 | [
"stale"
] | [] | 2023-07-13T02:42:28Z | 2023-08-22T15:04:09Z | 2023-08-22T15:04:09Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ruio248 | 93,460,643 | U_kgDOBZIYow | User | false |
huggingface/diffusers | 1,802,096,641 | I_kwDOHa8MBc5radAB | 4,069 | https://github.com/huggingface/diffusers/issues/4069 | https://api.github.com/repos/huggingface/diffusers/issues/4069 | Unloading a Textual Inversion Checkpoint from Pipeline | ### Describe the bug
This is how I load a textual inversion checkpoint
```
pipeline.load_textual_inversion('myckpt.bin')
```
Is unloading implemented?
### Reproduction
```
pipeline.load_textual_inversion('myckpt.bin')
```
### Logs
_No response_
### System Info
Latest diffusers
### Who can help?
_No r... | closed | completed | false | 2 | [
"bug",
"stale"
] | [] | 2023-07-13T03:41:37Z | 2023-08-20T15:03:26Z | 2023-08-20T15:03:26Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sarmientoj24 | 8,830,319 | MDQ6VXNlcjg4MzAzMTk= | User | false |
huggingface/diffusers | 1,802,406,106 | I_kwDOHa8MBc5rboja | 4,071 | https://github.com/huggingface/diffusers/issues/4071 | https://api.github.com/repos/huggingface/diffusers/issues/4071 | Inpaint only+lama support | **Is your feature request related to a problem? Please describe.**
It was comfy to use the inpaint only + lama controlnet on sd webui to do inpaint that match the firefly's inpaint quality.
But it's not support in diffusers.
**Describe the solution you'd like**
I hope diffusers can add an official controlnet inap... | closed | completed | false | 2 | [
"stale"
] | [] | 2023-07-13T07:44:00Z | 2023-08-31T08:51:07Z | 2023-08-20T15:03:25Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | adelacvg | 27,419,496 | MDQ6VXNlcjI3NDE5NDk2 | User | false |
huggingface/diffusers | 1,802,537,117 | I_kwDOHa8MBc5rcIid | 4,072 | https://github.com/huggingface/diffusers/issues/4072 | https://api.github.com/repos/huggingface/diffusers/issues/4072 | Can ControlNet support more than 3 channels input image? | Now the controlnet only take image(tensor) with 3 channels as input conditioning image.
If I want to use a tensor with **more than 3 channels** as the conditioning image, is there any solution? | closed | completed | false | 4 | [
"stale"
] | [] | 2023-07-13T08:58:03Z | 2023-08-20T15:03:24Z | 2023-08-20T15:03:24Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | irimsky | 42,032,614 | MDQ6VXNlcjQyMDMyNjE0 | User | false |
huggingface/diffusers | 1,802,589,374 | I_kwDOHa8MBc5rcVS- | 4,073 | https://github.com/huggingface/diffusers/issues/4073 | https://api.github.com/repos/huggingface/diffusers/issues/4073 | `'Tensor' object has no attribute 'astype'` when running DeepFloyd IF safety checker | ### Describe the bug
I guess someone has made a mistake on data types. When I run the example code of DeepFloyd IF, I got `'Tensor' object has no attribute 'astype' `
And I digged into the code a bit. The error comes from
https://github.com/huggingface/diffusers/blob/92a57a8e84a98a4da87b6a0621615eed4a23edc4/s... | closed | completed | false | 5 | [] | [
"yiyixuxu"
] | 2023-07-13T09:21:19Z | 2023-09-05T08:19:18Z | 2023-09-05T08:19:18Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ifsheldon | 39,153,080 | MDQ6VXNlcjM5MTUzMDgw | User | false |
huggingface/diffusers | 1,802,624,610 | I_kwDOHa8MBc5rcd5i | 4,074 | https://github.com/huggingface/diffusers/issues/4074 | https://api.github.com/repos/huggingface/diffusers/issues/4074 | SDXL weird blue and red points on image | 
I use diffusers to inference sdxl 0.9 checkpoints, the output image is ok, but I found blue and red points on image, it's weird.
I have used fp16, enable_model_cpu_offload, enable_vae_slicing, enable_vae_tiling. An... | closed | completed | false | 2 | [] | [] | 2023-07-13T09:35:58Z | 2023-07-17T12:38:52Z | 2023-07-13T09:42:29Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ousinkou | 13,220,463 | MDQ6VXNlcjEzMjIwNDYz | User | false |
huggingface/diffusers | 1,802,697,631 | I_kwDOHa8MBc5rcvuf | 4,075 | https://github.com/huggingface/diffusers/issues/4075 | https://api.github.com/repos/huggingface/diffusers/issues/4075 | Minor error in the docstring of `ShapEImg2ImgPipeline` | The first argument of `__call__` is `image`, not `prompt` here.
https://github.com/huggingface/diffusers/blob/92a57a8e84a98a4da87b6a0621615eed4a23edc4/src/diffusers/pipelines/shap_e/pipeline_shap_e_img2img.py#L217-L218 | closed | completed | false | 2 | [
"stale"
] | [] | 2023-07-13T10:17:41Z | 2023-08-14T19:49:58Z | 2023-08-14T19:49:58Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | hysts | 25,161,192 | MDQ6VXNlcjI1MTYxMTky | User | false |
huggingface/diffusers | 1,802,861,934 | I_kwDOHa8MBc5rdX1u | 4,077 | https://github.com/huggingface/diffusers/issues/4077 | https://api.github.com/repos/huggingface/diffusers/issues/4077 | "NameError: name 'init_empty_weights' is not defined" when accelerate is not installed | https://github.com/huggingface/diffusers/blame/92a57a8e84a98a4da87b6a0621615eed4a23edc4/src/diffusers/pipelines/stable_diffusion/convert_from_ckpt.py#L1344
`init_empty_weights()` is imported from accelerate if is_accelerate_available() returns true. However if that's not the case there is no fallback.
We had to pin... | closed | completed | false | 4 | [
"stale"
] | [] | 2023-07-13T11:55:26Z | 2023-08-20T15:03:23Z | 2023-08-20T15:03:23Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | cstueckrath | 3,080,637 | MDQ6VXNlcjMwODA2Mzc= | User | false |
huggingface/diffusers | 1,803,562,253 | I_kwDOHa8MBc5rgC0N | 4,080 | https://github.com/huggingface/diffusers/issues/4080 | https://api.github.com/repos/huggingface/diffusers/issues/4080 | Add in-painting pipeline for Stable Diffusion XL | **Is your feature request related to a problem? Please describe.**
SDXL is awesome- is there a plan to add an in-painting pipeline?
**Describe the solution you'd like**
There are several pipes in the stable diffusion repo for SD 1.5 and 2.1:
- `StableDiffusionPipeline`
- `StableDiffusionImg2ImgPipeline`
- `Stab... | closed | completed | false | 3 | [] | [] | 2023-07-13T18:34:55Z | 2023-07-14T15:43:59Z | 2023-07-14T15:43:59Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | DaveFerBear | 12,205,022 | MDQ6VXNlcjEyMjA1MDIy | User | false |
huggingface/diffusers | 1,803,931,899 | I_kwDOHa8MBc5rhdD7 | 4,085 | https://github.com/huggingface/diffusers/issues/4085 | https://api.github.com/repos/huggingface/diffusers/issues/4085 | Example script for training a lora for the SDXL refiner | It would be neat to extend the SDXL dreambooth Lora script with an example of how to train the refiner. Just training the base model isn't feasible for accurately generating images of subjects such as people, animals, etc. | closed | completed | false | 3 | [
"stale"
] | [] | 2023-07-14T00:22:21Z | 2023-08-23T15:04:19Z | 2023-08-23T15:04:19Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | BitPhinix | 13,185,548 | MDQ6VXNlcjEzMTg1NTQ4 | User | false |
huggingface/diffusers | 1,803,935,215 | I_kwDOHa8MBc5rhd3v | 4,086 | https://github.com/huggingface/diffusers/issues/4086 | https://api.github.com/repos/huggingface/diffusers/issues/4086 | training controlnet with stabilityai/stable-diffusion-xl-base-0.9 does not seem to be working | ### Describe the bug
training controlnet with the latest stabilityai/stable-diffusion-xl-base-0.9 does not seem to be working
### Reproduction
Firstly install all needed packages, including whatever mentioned in the official model card https://huggingface.co/stabilityai/stable-diffusion-xl-base-0.9
Then follow th... | closed | completed | false | 1 | [
"bug"
] | [] | 2023-07-14T00:27:37Z | 2023-07-14T03:49:05Z | 2023-07-14T03:49:05Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | yutongli | 12,973,127 | MDQ6VXNlcjEyOTczMTI3 | User | false |
huggingface/diffusers | 1,804,001,547 | I_kwDOHa8MBc5rhuEL | 4,089 | https://github.com/huggingface/diffusers/issues/4089 | https://api.github.com/repos/huggingface/diffusers/issues/4089 | During multi-gpus training, each card will execute the cache text embedding operation once? | ```python
text_encoders = [text_encoder_one, text_encoder_two]
tokenizers = [tokenizer_one, tokenizer_two]
train_dataset = get_train_dataset(args, accelerator)
compute_embeddings_fn = functools.partial(
compute_embeddings,
text_encoders=text_encoders,
tokenizers=tokenizers,
proportion_empty_prom... | closed | completed | false | 8 | [
"stale"
] | [] | 2023-07-14T01:43:29Z | 2023-08-23T15:04:18Z | 2023-08-23T15:04:18Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | pixeli99 | 46,072,190 | MDQ6VXNlcjQ2MDcyMTkw | User | false |
huggingface/diffusers | 1,365,313,454 | I_kwDOHa8MBc5RYQeu | 409 | https://github.com/huggingface/diffusers/issues/409 | https://api.github.com/repos/huggingface/diffusers/issues/409 | AttributeError: 'StableDiffusionPipeline' object has no attribute 'enable_attention_slicing' | ### Describe the bug
I'm getting the error ```AttributeError: 'StableDiffusionPipeline' object has no attribute 'enable_attention_slicing'``` Without it I can use stable diffusion properly, but I'm trying to save space, as it uses all 10gb off my 3080
### Reproduction
```
# make sure you're logged in with `h... | closed | completed | false | 5 | [
"bug"
] | [] | 2022-09-07T23:43:38Z | 2023-09-19T09:59:35Z | 2022-09-13T07:57:47Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | kritonpc | 693,201 | MDQ6VXNlcjY5MzIwMQ== | User | false |
huggingface/diffusers | 1,804,108,907 | I_kwDOHa8MBc5riIRr | 4,091 | https://github.com/huggingface/diffusers/issues/4091 | https://api.github.com/repos/huggingface/diffusers/issues/4091 | training | Do the image resolutions in my custom dataset need to be the same for training | closed | completed | false | 2 | [
"stale"
] | [] | 2023-07-14T03:47:55Z | 2023-08-21T15:03:44Z | 2023-08-21T15:03:44Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | bistuxlf | 63,131,915 | MDQ6VXNlcjYzMTMxOTE1 | User | false |
huggingface/diffusers | 1,804,118,586 | I_kwDOHa8MBc5riKo6 | 4,092 | https://github.com/huggingface/diffusers/issues/4092 | https://api.github.com/repos/huggingface/diffusers/issues/4092 | UnboundLocalError: local variable 'images' referenced before assignment | I was fine-tuning stable-diffusion model using train_text_to_image_lora.py. Faced this error once the training is done and could not push model to hub.
Steps: 100% 500/500 [14:51<00:00, 1.76s/it, lr=0, step_loss=0.0177] Model weights saved in finetune-stable-diffusion-v1-4-planogram-lora/pytorch_lora_weights.b... | closed | completed | false | 2 | [
"stale"
] | [] | 2023-07-14T04:00:43Z | 2023-08-21T15:03:43Z | 2023-08-21T15:03:43Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | bharadwaj-kg | 71,129,599 | MDQ6VXNlcjcxMTI5NTk5 | User | false |
huggingface/diffusers | 1,804,364,714 | I_kwDOHa8MBc5rjGuq | 4,093 | https://github.com/huggingface/diffusers/issues/4093 | https://api.github.com/repos/huggingface/diffusers/issues/4093 | cannot import name 'is_wandb_available' from 'diffusers.utils' | when i run the script ‘train_text_to_image.py’， return a import Error.
`cannot import name 'is_wandb_available' from 'diffusers.utils'`
the version of diffusers is 0.18.0. As far as I know, this is the latest version.
--------
There are two versions of the library in my pc, pip and source code installation. Rem... | closed | completed | false | 0 | [] | [] | 2023-07-14T07:27:11Z | 2023-07-14T07:33:18Z | 2023-07-14T07:28:47Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | cqf7419 | 53,525,729 | MDQ6VXNlcjUzNTI1NzI5 | User | false |
huggingface/diffusers | 1,804,422,116 | I_kwDOHa8MBc5rjUvk | 4,094 | https://github.com/huggingface/diffusers/issues/4094 | https://api.github.com/repos/huggingface/diffusers/issues/4094 | [Examples] Add a training script for T2I adapters | @HimariO and @williamberman are adding support of [T2I adapters](https://github.com/TencentARC/T2I-Adapter) here: https://github.com/huggingface/diffusers/pull/3932.
Would be cool to have a training script for powerful T2I adapters.
Anyone's interested to pick this one up? | closed | completed | false | 1 | [
"stale"
] | [] | 2023-07-14T08:09:17Z | 2023-08-30T15:04:01Z | 2023-08-30T15:04:01Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sayakpaul | 22,957,388 | MDQ6VXNlcjIyOTU3Mzg4 | User | false |
huggingface/diffusers | 1,804,427,071 | I_kwDOHa8MBc5rjV8_ | 4,095 | https://github.com/huggingface/diffusers/issues/4095 | https://api.github.com/repos/huggingface/diffusers/issues/4095 | [Examples] Add a training script for HyperDreamBooth | Nathaniel Ruiz et al. dropped https://hyperdreambooth.github.io/.
It combines DreamBooth, LoRA, and hypernetworks for a 100kb residual, generated in 20 seconds, that rivals original DreamBooth in model integrity, editability, and subject fidelity.
Seems to be working exceptionally well on faces.
Is anyone in... | closed | not_planned | false | 41 | [
"good first issue",
"Good second issue"
] | [] | 2023-07-14T08:12:39Z | 2024-11-25T13:09:14Z | 2024-11-25T13:09:13Z | MEMBER | null | 20260407T133413Z | 2026-04-07T13:34:13Z | sayakpaul | 22,957,388 | MDQ6VXNlcjIyOTU3Mzg4 | User | false |
huggingface/diffusers | 1,804,664,627 | I_kwDOHa8MBc5rkP8z | 4,096 | https://github.com/huggingface/diffusers/issues/4096 | https://api.github.com/repos/huggingface/diffusers/issues/4096 | Adding Zero123 and other LDM enabled 3D reconstruction models | ### Model/Pipeline/Scheduler description
Recent advances in multiview synthesis with 2D LDMs has been made possible by training LDMs on camera extrinsics. This research has enabled a high-fidelity 3D reconstruction with a single input image using implicit models (NeRFs, SDFs).
While there are several papers that us... | closed | completed | false | 15 | [
"stale"
] | [] | 2023-07-14T10:52:43Z | 2023-09-03T15:03:52Z | 2023-09-03T15:03:52Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | pgn-dev | 3,833,633 | MDQ6VXNlcjM4MzM2MzM= | User | false |
huggingface/diffusers | 1,365,323,040 | I_kwDOHa8MBc5RYS0g | 410 | https://github.com/huggingface/diffusers/issues/410 | https://api.github.com/repos/huggingface/diffusers/issues/410 | Unet model only supports internal block channel sizes in multiples of 32 | **What API design would you like to have changed or added to the library? Why?**
I see that the current unet model breaks if `block_out_channels` is set to something other than a multiple of 32. I am just curious if this was on purpose? I would actually like to use a size smaller than 32. I tested this by changing t... | closed | completed | false | 2 | [] | [] | 2022-09-08T00:04:13Z | 2022-09-15T14:35:16Z | 2022-09-15T14:35:16Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | ezhang7423 | 54,998,055 | MDQ6VXNlcjU0OTk4MDU1 | User | false |
huggingface/diffusers | 1,804,831,481 | I_kwDOHa8MBc5rk4r5 | 4,100 | https://github.com/huggingface/diffusers/issues/4100 | https://api.github.com/repos/huggingface/diffusers/issues/4100 | [VAE upscast] Reminder issue | Add slow tests for broken VAEs for https://github.com/huggingface/diffusers/pull/4083 | closed | completed | false | 1 | [
"stale"
] | [
"patrickvonplaten"
] | 2023-07-14T12:50:59Z | 2023-08-21T15:03:40Z | 2023-08-21T15:03:40Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/diffusers | 1,804,850,999 | I_kwDOHa8MBc5rk9c3 | 4,101 | https://github.com/huggingface/diffusers/issues/4101 | https://api.github.com/repos/huggingface/diffusers/issues/4101 | bug:convert controlnet model(custom datasets train) to diffusers failure | I trained a controlnet model based on sd1.5 or sd2.1 using my own data. The trained controlnet model performs inference correctly. However, when I try to convert it using the scripts/convert_original_controlnet_to_diffusers.py script, I encounter the following error:
Traceback (most recent call last):
File "/home/j... | closed | completed | false | 7 | [
"stale"
] | [] | 2023-07-14T13:04:26Z | 2023-08-21T15:03:39Z | 2023-08-21T15:03:39Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | chenjingcheng | 25,758,005 | MDQ6VXNlcjI1NzU4MDA1 | User | false |
huggingface/diffusers | 1,804,959,769 | I_kwDOHa8MBc5rlYAZ | 4,102 | https://github.com/huggingface/diffusers/issues/4102 | https://api.github.com/repos/huggingface/diffusers/issues/4102 | VAE should have the ability to use bfloat16 if it is available to the user for LDMs that use a VAE, to save memory | **Is your feature request related to a problem? Please describe.**
Currently there are many casts to float32 so that the VAE can work without hitting zero values when in float16 mode. When in float16 mode or when dtype=torch.bfloat16 we should use the VAE in torch.bfloat16 if it is available in our environment. This w... | closed | completed | false | 5 | [
"stale"
] | [] | 2023-07-14T14:14:46Z | 2023-08-21T15:03:38Z | 2023-08-21T15:03:38Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | AmericanPresidentJimmyCarter | 110,263,573 | U_kgDOBpJ9FQ | User | false |
huggingface/diffusers | 1,804,982,478 | I_kwDOHa8MBc5rldjO | 4,103 | https://github.com/huggingface/diffusers/issues/4103 | https://api.github.com/repos/huggingface/diffusers/issues/4103 | AutoencoderKL receives `scaling_factor` but does not use it | Hello,
As I am trying to understand more the role of the scaling factor `0.18215` for VAE in stable diffusion, I realised that the class does not actually use the parameter. Where should it be used?
https://github.com/huggingface/diffusers/blob/ad8f985e81d0e0bac0375cd81b909987e8b2a7f9/src/diffusers/models/autoencod... | closed | completed | false | 2 | [] | [] | 2023-07-14T14:29:43Z | 2023-07-17T15:06:28Z | 2023-07-17T15:06:27Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | fdtomasi | 12,514,317 | MDQ6VXNlcjEyNTE0MzE3 | User | false |
huggingface/diffusers | 1,805,502,974 | I_kwDOHa8MBc5rncn- | 4,104 | https://github.com/huggingface/diffusers/issues/4104 | https://api.github.com/repos/huggingface/diffusers/issues/4104 | Black image output when running pipeline; "invalid value encountered in cast" | ### Describe the bug
Hey I'm having a very similar issue to #2153. I'm trying to run ControlNet, but weather I run StableDiffusionPipeline or StableDiffusionControlNetPipeline, I get a black output image.
The the only notable log is
`C:\Users\nwam\AppData\Local\Programs\Python\Python311\Lib\site-packages\diffu... | closed | completed | false | 4 | [
"bug"
] | [] | 2023-07-14T20:11:29Z | 2024-06-11T06:41:02Z | 2023-07-17T19:35:31Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | nwam | 9,338,772 | MDQ6VXNlcjkzMzg3NzI= | User | false |
huggingface/diffusers | 1,805,787,674 | I_kwDOHa8MBc5roiIa | 4,107 | https://github.com/huggingface/diffusers/issues/4107 | https://api.github.com/repos/huggingface/diffusers/issues/4107 | SDXL LoRA script adds strange noise/reduces quality of base model | ### Describe the bug
I spent the day training LoRAs for SDXL with the diffusers script and observed something odd. No matter how little or how much I train a LoRA, adding it to the base model adds significant noise and reduces its quality.
This is the output with a LoRA weight 1
<img width="942" alt="Screenshot... | closed | completed | false | 19 | [
"bug",
"stale"
] | [] | 2023-07-15T01:18:37Z | 2023-08-31T15:04:36Z | 2023-08-31T15:04:36Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | killah-t-cell | 26,853,713 | MDQ6VXNlcjI2ODUzNzEz | User | false |
huggingface/diffusers | 1,805,833,729 | I_kwDOHa8MBc5rotYB | 4,108 | https://github.com/huggingface/diffusers/issues/4108 | https://api.github.com/repos/huggingface/diffusers/issues/4108 | Loading a SDXL model using from_single_file forces the use of the StableDiffusionXLPipeline | When loading a SDXL model using `from_single_file()`, the returned pipeline is always an instance of `StableDiffusionXLPipeline`. This means that if you try to load a model into a `StableDiffusionXLImg2ImgPipeline` the returned pipeline is always `StableDiffusionXLPipeline` instead.
The culprint is this line - https... | closed | completed | false | 3 | [] | [] | 2023-07-15T02:48:21Z | 2023-07-19T12:40:15Z | 2023-07-19T12:40:15Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Mystfit | 795,851 | MDQ6VXNlcjc5NTg1MQ== | User | false |
huggingface/diffusers | 1,805,921,377 | I_kwDOHa8MBc5rpCxh | 4,109 | https://github.com/huggingface/diffusers/issues/4109 | https://api.github.com/repos/huggingface/diffusers/issues/4109 | StableDiffusionXLInpaintPipeline not working | ### Describe the bug
Getting this error in StableDiffusionXLInpaintPipeline,
File "/opt/conda/envs/ldm/lib/python3.11/site-packages/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py", line 1084, in __call__
latents_outputs = self.prepare_latents(
^^^^^^^^^^^^^^^^^^^^^
File "/opt/c... | closed | completed | false | 2 | [
"bug"
] | [] | 2023-07-15T06:07:37Z | 2023-07-15T14:10:41Z | 2023-07-15T14:10:41Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | adhikjoshi | 11,740,719 | MDQ6VXNlcjExNzQwNzE5 | User | false |
huggingface/diffusers | 1,365,330,402 | I_kwDOHa8MBc5RYUni | 411 | https://github.com/huggingface/diffusers/issues/411 | https://api.github.com/repos/huggingface/diffusers/issues/411 | Textual Inversion training notebook only takes in remote images | The [Textual Inversion training notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/sd_textual_inversion_training.ipynb) in Google Colab only takes in remote images, instead of also being able to take in local images either uploaded from the desktop or linked from Google Drive, ... | closed | completed | false | 9 | [
"stale"
] | [
"apolinario",
"patil-suraj"
] | 2022-09-08T00:15:25Z | 2023-01-16T13:14:09Z | 2023-01-16T13:14:09Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | minimaxir | 2,179,708 | MDQ6VXNlcjIxNzk3MDg= | User | false |
huggingface/diffusers | 1,805,967,244 | I_kwDOHa8MBc5rpN-M | 4,110 | https://github.com/huggingface/diffusers/issues/4110 | https://api.github.com/repos/huggingface/diffusers/issues/4110 | TypeError: For multiple controlnets: `image` must be type `list` | ### Describe the bug
`StableDiffusionControlNetInpaintPipeline` <br>
raises TypeError: For multiple controlnets: `image` must be type `list``
### Reproduction
```python
import itertools
import random
import cv2
# !pip install transformers accelerate
from diffusers import StableDiffusionControlNetInpaintPipel... | closed | completed | false | 5 | [
"bug"
] | [
"yiyixuxu"
] | 2023-07-15T08:19:10Z | 2023-07-26T05:07:40Z | 2023-07-18T10:40:20Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | elcolie | 18,206,728 | MDQ6VXNlcjE4MjA2NzI4 | User | false |
huggingface/diffusers | 1,805,978,729 | I_kwDOHa8MBc5rpQxp | 4,111 | https://github.com/huggingface/diffusers/issues/4111 | https://api.github.com/repos/huggingface/diffusers/issues/4111 | StableDiffusionXLImg2ImgPipeline: Unable to load with 16GB VRAM | ### Describe the bug
Loading StableDiffusionXLImg2ImgPipeline on 16GB VRAM A4000 hardware isn't possible anymore.
This runs out of memory with the latest master, it loaded in the past without low_cpu_mem_usage=True , which also doesn't work now.
pipe = StableDiffusionXLImg2ImgPipeline.from_pretrained(model_... | closed | completed | false | 5 | [
"bug",
"stale"
] | [] | 2023-07-15T08:56:57Z | 2023-08-22T15:04:03Z | 2023-08-22T15:04:03Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | PtrMan | 1,067,920 | MDQ6VXNlcjEwNjc5MjA= | User | false |
huggingface/diffusers | 1,806,264,502 | I_kwDOHa8MBc5rqWi2 | 4,116 | https://github.com/huggingface/diffusers/issues/4116 | https://api.github.com/repos/huggingface/diffusers/issues/4116 | StableDiffusionXLImg2ImgPipeline: passing embeddings did not work | ### Describe the bug
I retrieved the positive embeddings for a prompt with the method encode_prompt() to get the embeddings and passed it into the pipeline.
It did crash in some pipeline internals days ago.
Can't try it now because I only got access to 16GB machines and because of the other issue.
### Repro... | closed | completed | false | 1 | [
"bug"
] | [] | 2023-07-15T18:20:30Z | 2023-07-20T09:54:02Z | 2023-07-20T09:54:02Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | PtrMan | 1,067,920 | MDQ6VXNlcjEwNjc5MjA= | User | false |
huggingface/diffusers | 1,806,405,172 | I_kwDOHa8MBc5rq440 | 4,117 | https://github.com/huggingface/diffusers/issues/4117 | https://api.github.com/repos/huggingface/diffusers/issues/4117 | StableDiffusionXL not working for Dreambooth | ### Describe the bug
Hi,
Is this model available for dreambooth? I'm not sure wether to put this as an issue or a feature request, but I've tried to run the regular script for the XL model to see how good it is in comparison with the other sd versions but I'm getting an error :'(
I know that XL's architecture m... | closed | completed | false | 11 | [
"bug"
] | [] | 2023-07-16T02:40:46Z | 2023-07-31T02:02:21Z | 2023-07-16T05:33:31Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | mauricio-repetto | 11,300,196 | MDQ6VXNlcjExMzAwMTk2 | User | false |
huggingface/diffusers | 1,806,549,091 | I_kwDOHa8MBc5rrcBj | 4,118 | https://github.com/huggingface/diffusers/issues/4118 | https://api.github.com/repos/huggingface/diffusers/issues/4118 | 'AutoencoderKL' has no attribute 'from_single_file' | ### Describe the bug
now 0.18.2 many pipeline support from_single_file，very great work!
but I also found some modules not support.
for example ，I want to load a single file SD model SD-1-5.safetensors，this is ok。
but I want load a VAE single safetensor too,but now not support it.
It now feels like a mixture o... | closed | completed | false | 3 | [
"bug"
] | [] | 2023-07-16T11:38:46Z | 2023-07-21T05:18:18Z | 2023-07-21T05:18:18Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | zhaoyun0071 | 35,762,050 | MDQ6VXNlcjM1NzYyMDUw | User | false |
huggingface/diffusers | 1,806,594,242 | I_kwDOHa8MBc5rrnDC | 4,120 | https://github.com/huggingface/diffusers/issues/4120 | https://api.github.com/repos/huggingface/diffusers/issues/4120 | Inverse a simple DDIMPipeline | I want to do inversion of a simple `DDIMPipeline` using the `DDIMInverseScheduler`. Why does the `DDIMPipeline` not contain an `invert` function similarly to, for example the `StableDiffusionPix2PixZeroPipeline` class? How can I do inversion of it? | closed | completed | false | 4 | [
"stale"
] | [] | 2023-07-16T14:03:43Z | 2023-08-23T15:04:12Z | 2023-08-23T15:04:12Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | GiilDe | 38,991,777 | MDQ6VXNlcjM4OTkxNzc3 | User | false |
huggingface/diffusers | 1,806,599,140 | I_kwDOHa8MBc5rroPk | 4,122 | https://github.com/huggingface/diffusers/issues/4122 | https://api.github.com/repos/huggingface/diffusers/issues/4122 | DPMSolverMultistepScheduler has wrong number of timesteps given num_inference_steps with StableDiffusionXL pipelines | ### Describe the bug
For example when 100 steps are given for num_inference_steps, only 95 are performed.
### Reproduction
```py
from diffusers import StableDiffusionXLPipeline
from diffusers.schedulers import DPMSolverMultistepScheduler
import torch
scheduler = DPMSolverMultistepScheduler(
beta... | closed | completed | false | 8 | [
"bug"
] | [
"yiyixuxu"
] | 2023-07-16T14:18:16Z | 2023-09-20T04:58:07Z | 2023-08-24T15:04:36Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | AmericanPresidentJimmyCarter | 110,263,573 | U_kgDOBpJ9FQ | User | false |
huggingface/diffusers | 1,806,603,065 | I_kwDOHa8MBc5rrpM5 | 4,123 | https://github.com/huggingface/diffusers/issues/4123 | https://api.github.com/repos/huggingface/diffusers/issues/4123 | where's scheduler.step? Help implementing SD sonar | Hi, in the code for SD webui sonar you have this:
```
def sample_euler_ancestral_ex(model:CFGDenoiser, x:Tensor, sigmas:List, extra_args={}, callback=None, eta=1.):
momentum = settings['momentum']
momentum_sign = settings['momentum_sign']
momentum_hist = settings['momentum_hist']
... | closed | completed | false | 2 | [] | [] | 2023-07-16T14:28:50Z | 2023-07-18T12:50:22Z | 2023-07-18T12:50:22Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | alexblattner | 15,870,094 | MDQ6VXNlcjE1ODcwMDk0 | User | false |
huggingface/diffusers | 1,806,717,164 | I_kwDOHa8MBc5rsFDs | 4,125 | https://github.com/huggingface/diffusers/issues/4125 | https://api.github.com/repos/huggingface/diffusers/issues/4125 | [SD 1.5 LoRA Fine-Tuning] Is this the expected loss curve when fine-tuning SD with LoRA? | ### Describe the bug
Hi there,
I'm currently fine-tuning SD1.5 on my machine using the diffusers LoRA script (detailed [here](https://huggingface.co/docs/diffusers/training/lora)) on the Pokemon BLIP dataset.
However, my (unsmoothed/smoothed) loss curve looks like this:
![unsmoothed poke curve](https://github.c... | closed | completed | false | 2 | [
"bug",
"stale"
] | [] | 2023-07-16T20:56:14Z | 2023-08-24T15:04:35Z | 2023-08-24T15:04:35Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | codingperks | 56,696,610 | MDQ6VXNlcjU2Njk2NjEw | User | false |
huggingface/diffusers | 1,806,738,129 | I_kwDOHa8MBc5rsKLR | 4,126 | https://github.com/huggingface/diffusers/issues/4126 | https://api.github.com/repos/huggingface/diffusers/issues/4126 | nvm | nevermind | closed | completed | false | 0 | [
"bug"
] | [] | 2023-07-16T22:20:14Z | 2023-07-16T22:24:01Z | 2023-07-16T22:23:48Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | fractaldna22 | 37,323,518 | MDQ6VXNlcjM3MzIzNTE4 | User | false |
huggingface/diffusers | 1,806,826,169 | I_kwDOHa8MBc5rsfq5 | 4,127 | https://github.com/huggingface/diffusers/issues/4127 | https://api.github.com/repos/huggingface/diffusers/issues/4127 | Does SD 2.1 config need update? | Hi, I am testing SD2.1 and find the generation has some problems.
I initialize each component, unet/vae/text encoder from config and model checkpoint(not fp16, not safe tensor) from each folder in SD2.1 model hub. I need to do a little name change in vae state_checkpoint (key to to_k, value to to_v, etc). Then the g... | closed | completed | false | 3 | [
"stale"
] | [] | 2023-07-17T02:02:06Z | 2023-08-24T15:04:34Z | 2023-08-24T15:04:34Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Alexander-Jin | 27,693,552 | MDQ6VXNlcjI3NjkzNTUy | User | false |
huggingface/diffusers | 1,807,171,427 | I_kwDOHa8MBc5rtz9j | 4,129 | https://github.com/huggingface/diffusers/issues/4129 | https://api.github.com/repos/huggingface/diffusers/issues/4129 | About SDXL ControlNet | Hello, it seems that wandb is recording all black images. Can you confirm if it's working properly on your end during development? I have also printed the pixel values of the val image, and they are all 0.
Comm... | closed | completed | false | 8 | [] | [] | 2023-07-17T07:37:04Z | 2023-07-17T08:47:24Z | 2023-07-17T08:47:14Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | pixeli99 | 46,072,190 | MDQ6VXNlcjQ2MDcyMTkw | User | false |
huggingface/diffusers | 1,807,309,501 | I_kwDOHa8MBc5ruVq9 | 4,130 | https://github.com/huggingface/diffusers/issues/4130 | https://api.github.com/repos/huggingface/diffusers/issues/4130 | I want a loading solution based on this library that is compatible with most xl versions of Lora on the C site, rather than splitting it into its own algorithms like WebUI | Now someone can provide a solution
| closed | completed | false | 2 | [
"stale"
] | [] | 2023-07-17T09:00:16Z | 2023-08-24T15:04:33Z | 2023-08-24T15:04:32Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | skyheros001 | 32,533,832 | MDQ6VXNlcjMyNTMzODMy | User | false |
huggingface/diffusers | 1,808,128,609 | I_kwDOHa8MBc5rxdph | 4,133 | https://github.com/huggingface/diffusers/issues/4133 | https://api.github.com/repos/huggingface/diffusers/issues/4133 | Kohya LORA and variant Loading for SDXL | Currently, LoraLoaderMixin supports Koha format for older SD models. However, with the change in architecture and the two text encoders, the process is now different for SDXL.
I don't feel overly confident implementing this myself, but I have made an SDXL lora (Lycoris) that loads up fine in comfyui to test with:
... | closed | completed | false | 13 | [
"stale"
] | [] | 2023-07-17T16:26:29Z | 2024-12-01T18:34:54Z | 2023-08-27T15:03:34Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | CoffeeVampir3 | 48,565,901 | MDQ6VXNlcjQ4NTY1OTAx | User | false |
huggingface/diffusers | 1,808,432,782 | I_kwDOHa8MBc5ryn6O | 4,134 | https://github.com/huggingface/diffusers/issues/4134 | https://api.github.com/repos/huggingface/diffusers/issues/4134 | Apple M1 Max (MPS) Tensor Size Not Matching For ControlNet From Pretrained Model | ### Describe the bug
I am unable to run the above despite ControlNet showing support for MPS in the python code.
### Reproduction
controlnet = ControlNetModel.from_pretrained(
"DionTimmer/controlnet_qrcode-control_v11p_sd21",
torch_dtype=torch.float32,
)
pipe = StableDiffusionControlNetImg2ImgPip... | closed | completed | false | 5 | [
"bug"
] | [
"pcuenca"
] | 2023-07-17T19:12:09Z | 2023-10-24T04:47:39Z | 2023-07-24T21:44:32Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | b2jant | 29,758,865 | MDQ6VXNlcjI5NzU4ODY1 | User | false |
huggingface/diffusers | 1,809,099,497 | I_kwDOHa8MBc5r1Krp | 4,137 | https://github.com/huggingface/diffusers/issues/4137 | https://api.github.com/repos/huggingface/diffusers/issues/4137 | Diffusers pipeline LoRA results different with WebUI and sd-scripts | ### Describe the bug
I want to use LoRA in the diffusers pipeline, however, the result is not the same as if I'm using webUI or sd-scripts
![https://raw.githubusercontent.com/lifeisboringsoprogramming/youtube/master/0019-introducing-stable-diffusion-xl-0-9-txt2img-automati-c1111-webui-extension/images/diffusers-l... | closed | completed | false | 4 | [
"bug"
] | [] | 2023-07-18T04:50:25Z | 2023-07-18T07:49:22Z | 2023-07-18T07:49:22Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | lifeisboringsoprogramming | 119,839,123 | U_kgDOBySZkw | User | false |
huggingface/diffusers | 1,809,499,884 | I_kwDOHa8MBc5r2sbs | 4,139 | https://github.com/huggingface/diffusers/issues/4139 | https://api.github.com/repos/huggingface/diffusers/issues/4139 | Cast precision for custom diffusion attention processor. | ### Describe the bug
I see in lora, the dtype are explicitly upcast:
https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py#L529
However, this is not done in custom diffusion. The weights need to be FP32 for mixed precision training, and during validation custom diffusion wi... | closed | completed | false | 4 | [
"bug",
"stale"
] | [] | 2023-07-18T09:05:42Z | 2023-08-17T23:36:02Z | 2023-08-17T23:36:02Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | eliphatfs | 23,738,781 | MDQ6VXNlcjIzNzM4Nzgx | User | false |
huggingface/diffusers | 1,809,505,735 | I_kwDOHa8MBc5r2t3H | 4,140 | https://github.com/huggingface/diffusers/issues/4140 | https://api.github.com/repos/huggingface/diffusers/issues/4140 | `_encode_prompt` function | At the moment, we have a bit of an inconsistent design with the `_encode_prompt` / `encode_prompt` functions.
We essentially have three different designs:
- 1.) The original stable diffusion 1.5 design, which means we have a private `_encode_prompt` function that does **not** use by default use `torch.no_grad()` ... | closed | completed | false | 17 | [
"wip",
"Good Example PR"
] | [
"williamberman",
"yiyixuxu",
"sayakpaul"
] | 2023-07-18T09:09:14Z | 2023-10-09T14:33:42Z | 2023-08-30T03:27:28Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/diffusers | 1,809,738,192 | I_kwDOHa8MBc5r3mnQ | 4,141 | https://github.com/huggingface/diffusers/issues/4141 | https://api.github.com/repos/huggingface/diffusers/issues/4141 | StableDiffusionControlNetInpaintPipeline does not respect 2nd controlnet | ### Describe the bug
`StableDiffusionControlNetInpaintPipeline` does not follow the 2nd `controlNet`
### Reproduction
```python
import copy
import itertools
import random
import cv2
# !pip install transformers accelerate
from diffusers import StableDiffusionControlNetInpaintPipeline, ControlNetModel, DDIMSch... | closed | completed | false | 2 | [
"bug"
] | [] | 2023-07-18T11:28:43Z | 2023-07-18T13:40:34Z | 2023-07-18T13:40:33Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | elcolie | 18,206,728 | MDQ6VXNlcjE4MjA2NzI4 | User | false |
huggingface/diffusers | 1,809,951,901 | I_kwDOHa8MBc5r4ayd | 4,144 | https://github.com/huggingface/diffusers/issues/4144 | https://api.github.com/repos/huggingface/diffusers/issues/4144 | No operator found for `memory_efficient_attention_backward` | ### Describe the bug
Training a controlnet in `diffusers/examples/controlnet/train_controlnet.py`, `xformers=0.0.16` is installed
```
No operator found for `memory_efficient_attention_backward` with inputs:
query : shape=(16, 1824, 1, 80) (torch.float32)
key : shape=(16, 77, 1, 80) (tor... | closed | completed | false | 6 | [
"bug"
] | [] | 2023-07-18T13:35:01Z | 2023-07-18T14:41:18Z | 2023-07-18T14:41:18Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | KyleYueye | 34,768,573 | MDQ6VXNlcjM0NzY4NTcz | User | false |
huggingface/diffusers | 1,810,509,405 | I_kwDOHa8MBc5r6i5d | 4,146 | https://github.com/huggingface/diffusers/issues/4146 | https://api.github.com/repos/huggingface/diffusers/issues/4146 | What's the difference between DreamBooth Lora and Dreambooth? | I see there's a train_dreambooth.py and train_dreambooth_lora.py. What's the difference between them? i also see there's a train_dreambooth_lora_sdxl.py, when will there be a pure dreambooth version of sdxl? i.e train_dreambooth_sdxl.py. Another question is, is it possible to pass negative prompt into SDXL? | closed | completed | false | 4 | [] | [] | 2023-07-18T18:51:42Z | 2023-07-20T16:01:18Z | 2023-07-20T06:39:58Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | MaxTran96 | 15,274,946 | MDQ6VXNlcjE1Mjc0OTQ2 | User | false |
huggingface/diffusers | 1,810,973,596 | I_kwDOHa8MBc5r8UOc | 4,148 | https://github.com/huggingface/diffusers/issues/4148 | https://api.github.com/repos/huggingface/diffusers/issues/4148 | SDXL model config error for inpainting | https://github.com/huggingface/diffusers/blob/ed2a3584ab2ee59676eb95c884ba2d7bb831f41d/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py#L643
hi, thank you for inpainting task.
I got error that "AttributeError: 'FrozenDict' object has no attribute 'force_upcast'".
I check sdxl mod... | closed | completed | false | 3 | [
"stale"
] | [] | 2023-07-19T01:26:24Z | 2023-08-26T15:03:10Z | 2023-08-26T15:03:10Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | junhyoung-ew | 129,705,636 | U_kgDOB7smpA | User | false |
huggingface/diffusers | 1,811,032,435 | I_kwDOHa8MBc5r8ilz | 4,149 | https://github.com/huggingface/diffusers/issues/4149 | https://api.github.com/repos/huggingface/diffusers/issues/4149 | Running a batch of prompts and negative prompts in SemanticStableDiffusionPipeline causes tensor dimension mismatch | ### Describe the bug
Hi,
I noticed that running a batch of prompts and negative prompts with the SemanticStableDiffusionPipeline causes tensor dimension mismatch between the negative prompt embeddings and prompt embeddings.
I believe a fix to this would be in Line 485 of diffusers/pipelines/semantic_stable_diffus... | closed | completed | false | 3 | [
"bug"
] | [] | 2023-07-19T02:43:37Z | 2023-07-28T16:14:24Z | 2023-07-28T16:14:24Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Guruprasad68 | 51,136,597 | MDQ6VXNlcjUxMTM2NTk3 | User | false |
huggingface/diffusers | 1,811,044,814 | I_kwDOHa8MBc5r8lnO | 4,150 | https://github.com/huggingface/diffusers/issues/4150 | https://api.github.com/repos/huggingface/diffusers/issues/4150 | How to train text-to-image model based on SDXL? | Can I use the train_text_to_image.py code directly? | closed | completed | false | 3 | [] | [] | 2023-07-19T02:59:00Z | 2023-07-21T15:23:30Z | 2023-07-20T04:54:47Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | EnzoWuu | 45,420,316 | MDQ6VXNlcjQ1NDIwMzE2 | User | false |
huggingface/diffusers | 1,811,212,263 | I_kwDOHa8MBc5r9Ofn | 4,151 | https://github.com/huggingface/diffusers/issues/4151 | https://api.github.com/repos/huggingface/diffusers/issues/4151 | How does --num_vectors in textual inversion actually work? | Hi all,
I'm currently using `examples/textual_inversion/textual_inversion.py` to learn some specific subjects, and I encountered two difficulties and would like to ask for help or ideas:
1. When using different values for the argument `--num_vectors` (all others stay the same), I observed that the loss curve of t... | closed | completed | false | 5 | [
"stale"
] | [] | 2023-07-19T05:58:50Z | 2023-09-04T15:03:54Z | 2023-09-04T15:03:54Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | hkazami | 36,339,155 | MDQ6VXNlcjM2MzM5MTU1 | User | false |
huggingface/diffusers | 1,811,407,294 | I_kwDOHa8MBc5r9-G- | 4,152 | https://github.com/huggingface/diffusers/issues/4152 | https://api.github.com/repos/huggingface/diffusers/issues/4152 | Support using different attention kwargs for different types of processors in one model. | **Is your feature request related to a problem? Please describe.**
Say you want to use CustomDiffusion for some layers, and LoRA for some others.
You want to pass a {'scale': 0.5} to LoRA.
Then the code goes:
```python
TypeError: __call__() got an unexpected keyword argument 'scale'
```
Because CustomDiffusion... | closed | completed | false | 11 | [
"stale"
] | [] | 2023-07-19T08:08:54Z | 2023-08-26T15:03:09Z | 2023-08-26T15:03:09Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | eliphatfs | 23,738,781 | MDQ6VXNlcjIzNzM4Nzgx | User | false |
huggingface/diffusers | 1,811,494,227 | I_kwDOHa8MBc5r-TVT | 4,153 | https://github.com/huggingface/diffusers/issues/4153 | https://api.github.com/repos/huggingface/diffusers/issues/4153 | The handling of the "!" token in SDv2/XL is different from the StabilityAI version. | In SD v2/XL tokenizer, ```!``` token is replaced by pad token, which makes a difference in tokenization.
```
from transformers import CLIPTokenizer
import open_clip
tokenizer = CLIPTokenizer.from_pretrained("stabilityai/stable-diffusion-xl-base-0.9", subfolder="tokenizer_2")
prompt = "! !! !!!"
input_ids = to... | closed | completed | false | 2 | [
"stale"
] | [] | 2023-07-19T08:58:41Z | 2023-08-26T15:03:08Z | 2023-08-26T15:03:08Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | laksjdjf | 22,386,664 | MDQ6VXNlcjIyMzg2NjY0 | User | false |
huggingface/diffusers | 1,812,155,443 | I_kwDOHa8MBc5sA0wz | 4,157 | https://github.com/huggingface/diffusers/issues/4157 | https://api.github.com/repos/huggingface/diffusers/issues/4157 | Replacing loaded LoRa weights | Quick question for you, @sayakpaul. If I want to reuse the same pipeline and load in new LoRa weights, can I call `load_lora_weights` again or do I need to first call `unload_lora_weights`? | closed | completed | false | 1 | [] | [] | 2023-07-19T15:05:00Z | 2023-07-20T06:37:12Z | 2023-07-20T06:37:11Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | levi | 8,493 | MDQ6VXNlcjg0OTM= | User | false |
huggingface/diffusers | 1,812,253,910 | I_kwDOHa8MBc5sBMzW | 4,158 | https://github.com/huggingface/diffusers/issues/4158 | https://api.github.com/repos/huggingface/diffusers/issues/4158 | the program gets stuck for a long tim when use from_single_file | ### Describe the bug
In my project, users can choose their preferred model, so I need to switch frequently between StableDiffusionPipeline and StableDiffusionXlPipeline. Sometimes, when the program switches to StableDiffusionPipeline and then back to StableDiffusionXlPipeline, it gets stuck at the from_single_file s... | closed | completed | false | 6 | [
"bug",
"stale"
] | [] | 2023-07-19T16:00:23Z | 2023-08-27T15:03:29Z | 2023-08-27T15:03:29Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | hckj588ku | 132,969,506 | U_kgDOB-z0Ig | User | false |
huggingface/diffusers | 1,812,698,980 | I_kwDOHa8MBc5sC5dk | 4,159 | https://github.com/huggingface/diffusers/issues/4159 | https://api.github.com/repos/huggingface/diffusers/issues/4159 | DiffEdit - Huge VRAM requirements (>63GB) | ### Describe the bug
I have tried running the DiffEdit pipeline with the given example and it requires a huge amount of VRAM.
If I am to use the `raw_image = load_image(img_url).convert("RGB").resize((768, 768))` as per example I get an OOM saying it requires ~63GB of VRAM.
### Reproduction
To reproduce i... | closed | completed | false | 1 | [
"bug"
] | [] | 2023-07-19T20:42:02Z | 2023-07-19T20:59:16Z | 2023-07-19T20:59:16Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | danieltudosiu | 10,502,740 | MDQ6VXNlcjEwNTAyNzQw | User | false |
huggingface/diffusers | 1,812,832,173 | I_kwDOHa8MBc5sDZ-t | 4,160 | https://github.com/huggingface/diffusers/issues/4160 | https://api.github.com/repos/huggingface/diffusers/issues/4160 | How to use diffusers force zeros? | it seems that it only has effect if its used on instance of diffusers class before model is loaded,
but i only get instance when i call from_pretrained or from_single_file
| closed | completed | false | 6 | [
"stale",
"SD.Next"
] | [
"patrickvonplaten"
] | 2023-07-19T22:36:38Z | 2023-09-01T13:09:28Z | 2023-09-01T13:09:27Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/diffusers | 1,812,833,044 | I_kwDOHa8MBc5sDaMU | 4,161 | https://github.com/huggingface/diffusers/issues/4161 | https://api.github.com/repos/huggingface/diffusers/issues/4161 | Kandinsky Prior doesn't work with SD.Next | How to configure prior to work with kandinsky 2.2?
Seems to not work in SD.Next | closed | completed | false | 11 | [
"SD.Next"
] | [
"patrickvonplaten"
] | 2023-07-19T22:37:33Z | 2023-07-26T15:14:12Z | 2023-07-26T15:14:11Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/diffusers | 1,812,833,833 | I_kwDOHa8MBc5sDaYp | 4,162 | https://github.com/huggingface/diffusers/issues/4162 | https://api.github.com/repos/huggingface/diffusers/issues/4162 | Problem with Img2Img in SD.Next | Img2img and init latents don't work with SD.Next | closed | completed | false | 3 | [
"SD.Next"
] | [
"patrickvonplaten"
] | 2023-07-19T22:38:31Z | 2023-11-09T15:06:54Z | 2023-11-09T15:06:54Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/diffusers | 1,812,834,194 | I_kwDOHa8MBc5sDaeS | 4,163 | https://github.com/huggingface/diffusers/issues/4163 | https://api.github.com/repos/huggingface/diffusers/issues/4163 | Callback Kandinsky | Implement callback for kandinsky | closed | completed | false | 1 | [
"SD.Next"
] | [
"patrickvonplaten"
] | 2023-07-19T22:38:59Z | 2023-07-26T15:15:04Z | 2023-07-26T15:15:04Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/diffusers | 1,812,834,669 | I_kwDOHa8MBc5sDalt | 4,164 | https://github.com/huggingface/diffusers/issues/4164 | https://api.github.com/repos/huggingface/diffusers/issues/4164 | Massive SD XL memory spike | Issue with sd-xl massive memory spike on second run
first run is fine, but second run spikes memory to full vram and then it stays constant | closed | completed | false | 16 | [
"stale",
"SD.Next"
] | [
"patrickvonplaten"
] | 2023-07-19T22:39:35Z | 2023-10-30T15:09:36Z | 2023-10-30T15:09:36Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/diffusers | 1,812,835,263 | I_kwDOHa8MBc5sDau_ | 4,165 | https://github.com/huggingface/diffusers/issues/4165 | https://api.github.com/repos/huggingface/diffusers/issues/4165 | Ability to interrupt generate process | Ability to interrupt generate process | closed | completed | false | 14 | [
"wip",
"SD.Next"
] | [
"DN6",
"patrickvonplaten"
] | 2023-07-19T22:40:17Z | 2024-01-04T06:55:54Z | 2024-01-04T06:55:54Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/diffusers | 1,812,838,832 | I_kwDOHa8MBc5sDbmw | 4,166 | https://github.com/huggingface/diffusers/issues/4166 | https://api.github.com/repos/huggingface/diffusers/issues/4166 | IF M breaks | Tried DeepFloyd/IF-I-M-v1.0 on nVidia A6000 with 48GB VRAM, it loads ok, but fails on inference:
Exception: CUDA out of memory. Tried to allocate 1536.45 GiB (GPU 0; 47.54 GiB total capacity; 16.54 GiB already allocated; 30.45 GiB free; 16.76 GiB reserved in total by PyTorch)
site-packages/diffusers/models/attent... | closed | completed | false | 5 | [
"stale",
"SD.Next"
] | [
"patrickvonplaten"
] | 2023-07-19T22:44:17Z | 2023-10-30T15:09:33Z | 2023-10-30T15:09:33Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/diffusers | 1,812,903,980 | I_kwDOHa8MBc5sDrgs | 4,167 | https://github.com/huggingface/diffusers/issues/4167 | https://api.github.com/repos/huggingface/diffusers/issues/4167 | A1111 <> Diffusers Scheduler mapping | # A1111 <> Diffusers Mapping
| A1111 / K-Diffusion | Diffusers | Notes |
| --- | --- | --- |
| DPM++ 2M | [Multistep DPM-Solver](https://huggingface.co/docs/diffusers/main/en/api/schedulers/multistep_dpm_solver) | |
| DPM++ 2M Karras | [Multistep DPM-Solver](https://huggingface.co/docs/diffusers/main/en/api/sched... | closed | completed | false | 29 | [] | [
"yiyixuxu"
] | 2023-07-19T23:56:50Z | 2024-05-21T11:31:25Z | 2023-09-12T17:39:05Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/diffusers | 1,813,035,322 | I_kwDOHa8MBc5sELk6 | 4,168 | https://github.com/huggingface/diffusers/issues/4168 | https://api.github.com/repos/huggingface/diffusers/issues/4168 | different inpainting in stable diffusion and diffusers | In stable diffusion we can use like original Stable Diffusion checkpoint. like realisticVisionV30_v30VAE.safetensors to do inpaint, but in diffusers's inpainting pipeline we must select corresponding inpainting model which means "realisticVisionV30_v30VAE-inpainting" to do the inpaint. Is there a way to use the origina... | closed | completed | false | 2 | [
"stale"
] | [] | 2023-07-20T02:34:10Z | 2023-08-27T15:03:26Z | 2023-08-27T15:03:26Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Laidawang | 85,244,566 | MDQ6VXNlcjg1MjQ0NTY2 | User | false |
huggingface/diffusers | 1,813,160,138 | I_kwDOHa8MBc5sEqDK | 4,169 | https://github.com/huggingface/diffusers/issues/4169 | https://api.github.com/repos/huggingface/diffusers/issues/4169 | Replace Command-Line Arguments with config.yaml Settings in train.py files | I'm trying to use SDXL Lora and Dreambooth to fine-tuning SDXL.
and, I feel uncomfortable when setting a lot of argparsers in terminal.
so, I think use config.yaml file and just set this config.yaml file to argparser.
than, when training, py file will read this YAML file and set variables in it.
ex)
#config.yaml... | closed | completed | false | 11 | [
"stale"
] | [] | 2023-07-20T04:56:32Z | 2023-08-27T15:03:25Z | 2023-08-27T15:03:25Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | hwjeon98 | 102,274,521 | U_kgDOBhiV2Q | User | false |
huggingface/diffusers | 1,813,360,357 | I_kwDOHa8MBc5sFa7l | 4,170 | https://github.com/huggingface/diffusers/issues/4170 | https://api.github.com/repos/huggingface/diffusers/issues/4170 | HyperDreamBooth Support | found and interesting paper from google [https://hyperdreambooth.github.io/](https://hyperdreambooth.github.io/), I think you all already know about that. but I think it will be great idea to introduce it in diffusers, if we start to work on this project, it only needs few images to train a simple few KB size model. an... | closed | completed | false | 1 | [] | [] | 2023-07-20T07:35:09Z | 2023-07-20T10:15:13Z | 2023-07-20T10:15:13Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | tronizx | 87,223,249 | MDQ6VXNlcjg3MjIzMjQ5 | User | false |
huggingface/diffusers | 1,813,381,143 | I_kwDOHa8MBc5sFgAX | 4,171 | https://github.com/huggingface/diffusers/issues/4171 | https://api.github.com/repos/huggingface/diffusers/issues/4171 | Incorrect render of Reference page from HF Diffusers home page | ### Describe the bug
On clicking the "Reference" button on the [Diffusers homepage](https://huggingface.co/docs/diffusers/index), I get the following text returned
Not sure if this is intended.
### Reproducti... | closed | completed | false | 2 | [
"bug"
] | [] | 2023-07-20T07:48:03Z | 2023-08-17T22:22:35Z | 2023-08-17T22:22:35Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | suvadityamuk | 70,141,886 | MDQ6VXNlcjcwMTQxODg2 | User | false |
huggingface/diffusers | 1,814,052,862 | I_kwDOHa8MBc5sID_- | 4,173 | https://github.com/huggingface/diffusers/issues/4173 | https://api.github.com/repos/huggingface/diffusers/issues/4173 | FABRIC non-training based iterative image improvement | **Is your feature request related to a problem? Please describe.**
FABRIC: PERSONALIZING DIFFUSION MODELS WITH ITERATIVE FEEDBACK
In an era where visual content generation is increasingly driven by machine learning,
the integration of human feedback into generative models presents significant
opportunities for ... | closed | completed | false | 5 | [
"community-examples",
"Good second issue"
] | [] | 2023-07-20T13:52:43Z | 2023-08-28T07:10:57Z | 2023-08-28T07:10:57Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | AmericanPresidentJimmyCarter | 110,263,573 | U_kgDOBpJ9FQ | User | false |
huggingface/diffusers | 1,814,576,622 | I_kwDOHa8MBc5sKD3u | 4,176 | https://github.com/huggingface/diffusers/issues/4176 | https://api.github.com/repos/huggingface/diffusers/issues/4176 | Mandatory authentication for diffusers | **Is your feature request related to a problem? Please describe.**
At the moment some HF diffusion models can be leveraged inside a notebook without mandatory authentication. The growing popularity of these models could increase the likelihood of potential misuse. In that case HF ethical guidelines would be violated:
... | closed | completed | false | 4 | [
"stale"
] | [] | 2023-07-20T18:30:35Z | 2023-08-31T15:04:30Z | 2023-08-31T15:04:30Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | giuliocn | 57,756,052 | MDQ6VXNlcjU3NzU2MDUy | User | false |
huggingface/diffusers | 1,814,626,353 | I_kwDOHa8MBc5sKQAx | 4,178 | https://github.com/huggingface/diffusers/issues/4178 | https://api.github.com/repos/huggingface/diffusers/issues/4178 | Unloading LoRA weights from an SD pipeline doesn't work with XFormers | ### Describe the bug
When running `unload_lora_weights` from a SD pipeline that has been previously loaded with XFormers attention processors, the unloading has no effect on the attention processors.
### Reproduction
```py
pipe = StableDiffusionPipeline.from_pretrained(
model_name,
torch_dtype=torch.float... | closed | completed | false | 0 | [
"bug"
] | [] | 2023-07-20T18:59:42Z | 2023-07-21T08:59:22Z | 2023-07-21T08:59:22Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | isidentical | 47,358,913 | MDQ6VXNlcjQ3MzU4OTEz | User | false |
huggingface/diffusers | 1,815,075,487 | I_kwDOHa8MBc5sL9qf | 4,183 | https://github.com/huggingface/diffusers/issues/4183 | https://api.github.com/repos/huggingface/diffusers/issues/4183 | Bug: controlnet generate image error when guidance_scale <= 1 | ### Describe the bug
I found that an error occurs when guidance_scale <= 1: "Sizes of tensors must match except in dimension 1. Expected size 1 but got size 2 for tensor number 1 in the list."
After reviewing the code, I found that when guidance_scale <= 1, the shape[0] of the sample is 1, while the shape[0] of the... | closed | completed | false | 2 | [
"bug"
] | [
"yiyixuxu"
] | 2023-07-21T02:54:48Z | 2023-07-24T08:10:16Z | 2023-07-24T08:10:16Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | chuck-ma | 74,402,255 | MDQ6VXNlcjc0NDAyMjU1 | User | false |
huggingface/diffusers | 1,815,088,087 | I_kwDOHa8MBc5sMAvX | 4,184 | https://github.com/huggingface/diffusers/issues/4184 | https://api.github.com/repos/huggingface/diffusers/issues/4184 | About the temperature parameter | Is there a way to adjust the temperature parameter when generating images? | closed | completed | false | 9 | [] | [] | 2023-07-21T03:15:22Z | 2023-07-27T05:06:24Z | 2023-07-27T05:06:23Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | suzukimain | 131,413,573 | U_kgDOB9U2RQ | User | false |
huggingface/diffusers | 1,815,088,676 | I_kwDOHa8MBc5sMA4k | 4,185 | https://github.com/huggingface/diffusers/issues/4185 | https://api.github.com/repos/huggingface/diffusers/issues/4185 | controlnet with sdxl infer black images even after rebasing #4038 | ### Describe the bug
I have been tightly following our amazing https://github.com/huggingface/diffusers/pull/4038, I got the new code and tried training for 10000 steps, training works however validation images are all black. I assume with #4038, we should have fixed the black image issue. Any clues?
### Reproduc... | closed | completed | false | 14 | [
"bug"
] | [] | 2023-07-21T03:16:21Z | 2023-10-04T17:39:53Z | 2023-07-25T00:33:44Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | yutongli | 12,973,127 | MDQ6VXNlcjEyOTczMTI3 | User | false |
huggingface/diffusers | 1,815,105,757 | I_kwDOHa8MBc5sMFDd | 4,186 | https://github.com/huggingface/diffusers/issues/4186 | https://api.github.com/repos/huggingface/diffusers/issues/4186 | Stable Diffusion XL Inpaint not working with refiner | ### Describe the bug
Stable Diffusion XL Inpaint not working with refiner. When doing inpainting, I get an unexpected keyword argument 'mask_image' error.
This could be because the refiner is not expecting a `mask_image` parameter.
Printing the `refiner` shows that it inherits from `StableDiffusionXLImg2ImgPi... | closed | completed | false | 3 | [
"bug"
] | [] | 2023-07-21T03:44:27Z | 2023-08-25T06:21:32Z | 2023-07-22T04:27:10Z | CONTRIBUTOR | null | 20260407T133413Z | 2026-04-07T13:34:13Z | darhsu | 35,377,472 | MDQ6VXNlcjM1Mzc3NDcy | User | false |
huggingface/diffusers | 1,815,239,825 | I_kwDOHa8MBc5sMlyR | 4,189 | https://github.com/huggingface/diffusers/issues/4189 | https://api.github.com/repos/huggingface/diffusers/issues/4189 | Inpainting/controlnet-inpaint affects areas outside masked area | ### Describe the bug
Inpainting seems to subtly affect areas outside the masked area. After many generations, the effect becomes very noticeable. Was unsure if I am somehow using it wrong since all I could find about this was this old issue
https://github.com/divamgupta/diffusionbee-stable-diffusion-ui/issues/269... | closed | completed | false | 3 | [
"bug",
"stale"
] | [] | 2023-07-21T06:38:02Z | 2023-08-28T15:03:55Z | 2023-08-28T15:03:55Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | Metal079 | 7,731,400 | MDQ6VXNlcjc3MzE0MDA= | User | false |
huggingface/diffusers | 1,366,053,650 | I_kwDOHa8MBc5RbFMS | 419 | https://github.com/huggingface/diffusers/issues/419 | https://api.github.com/repos/huggingface/diffusers/issues/419 | [Question] Maybe the sample code in Readme is outdated? | ## Desc
Maybe the sample code in Readme is outdated? (🤦♂️Or my operation is faulty. I copied and ran multiple examples directly, all with similar errors`AttributeError: 'dict' object has no attribute 'images'`.
## Env
**python** v3.10
**diffusers** v0.2.4
**huggingface-hub** v0.9.1
**transformers** v4.2... | closed | completed | false | 5 | [] | [] | 2022-09-08T10:58:43Z | 2022-09-08T23:30:41Z | 2022-09-08T23:30:41Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | QIN2DIM | 62,018,067 | MDQ6VXNlcjYyMDE4MDY3 | User | false |
huggingface/diffusers | 1,815,412,594 | I_kwDOHa8MBc5sNP9y | 4,192 | https://github.com/huggingface/diffusers/issues/4192 | https://api.github.com/repos/huggingface/diffusers/issues/4192 | So why is from_ckpt very different from from_pretrained with the arguments? | More of a question to learn. But I've noticed that from_ckpt and from_pretrained is very different.
So I'm using the StableDiffusionPipeline for additional context.
from_ckpt
```
cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
resume_download = kwargs.pop("resume_download", False)
force_download = kwargs... | closed | completed | false | 7 | [
"stale"
] | [] | 2023-07-21T08:44:49Z | 2024-01-26T12:23:48Z | 2023-08-28T15:03:54Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | JemiloII | 5,192,788 | MDQ6VXNlcjUxOTI3ODg= | User | false |
huggingface/diffusers | 1,815,753,418 | I_kwDOHa8MBc5sOjLK | 4,194 | https://github.com/huggingface/diffusers/issues/4194 | https://api.github.com/repos/huggingface/diffusers/issues/4194 | VAE not work,when use StableDiffusionPipeline.from_single_file | ### Describe the bug
```
pipe = StableDiffusionPipeline.from_single_file("xx.safetensors",
torch_dtype=torch.float16, load_safety_checker=False, controlnet=False,vae=AutoencoderKL.from_pretrained("stabilityai/sd-vae-ft-mse",torch_dtype=torch.float16).to("cuda")
... | closed | completed | false | 16 | [
"bug"
] | [] | 2023-07-21T12:48:24Z | 2023-08-08T10:43:14Z | 2023-07-25T00:00:25Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | zhaoyun0071 | 35,762,050 | MDQ6VXNlcjM1NzYyMDUw | User | false |
huggingface/diffusers | 1,816,103,186 | I_kwDOHa8MBc5sP4kS | 4,197 | https://github.com/huggingface/diffusers/issues/4197 | https://api.github.com/repos/huggingface/diffusers/issues/4197 | Feature Request: enable_model_cpu_offload for MPS | https://github.com/huggingface/diffusers/blob/5e80827369272f4b24af51573466aba24454b068/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_upscale.py#L160
enable_model_cpu_offload doens't work for MPS because it's currently expecting CUDA parameters only.
Could we change the code to also use this f... | closed | completed | false | 2 | [
"stale"
] | [] | 2023-07-21T16:47:27Z | 2023-08-29T15:03:58Z | 2023-08-29T15:03:58Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | soten355 | 8,527,497 | MDQ6VXNlcjg1Mjc0OTc= | User | false |
huggingface/diffusers | 1,816,126,756 | I_kwDOHa8MBc5sP-Uk | 4,198 | https://github.com/huggingface/diffusers/issues/4198 | https://api.github.com/repos/huggingface/diffusers/issues/4198 | mat1 and mat2 must have the same dtype (https://github.com/huggingface/diffusion-models-class/blob/main/unit3/01_stable_diffusion_introduction.ipynb) | ### Describe the bug
giving error:
```
---------------------------------------------------------------------------
RuntimeError Traceback (most recent call last)
[<ipython-input-45-1dbcf63417f5>](https://localhost:8080/#) in <cell line: 14>()
13 # Model prediction:
14 w... | closed | completed | false | 0 | [
"bug"
] | [] | 2023-07-21T17:08:30Z | 2023-07-22T06:09:44Z | 2023-07-22T06:09:44Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | andysingal | 20,493,493 | MDQ6VXNlcjIwNDkzNDkz | User | false |
huggingface/diffusers | 1,816,193,091 | I_kwDOHa8MBc5sQOhD | 4,199 | https://github.com/huggingface/diffusers/issues/4199 | https://api.github.com/repos/huggingface/diffusers/issues/4199 | Downloading real regularization images (Custom Diffusion Cat training example) | ### Describe the bug
The provided scripts stop at 98% (196 / 200).
### Reproduction
pip install clip-retrieval
python retrieve.py --class_prompt cat --class_data_dir real_reg/samples_cat --num_class_images 200
### Logs
```shell
(custom) user@Puget-233195:/ssd1/workspace/diffusers/examples/custom_diffusion$ pytho... | closed | completed | false | 4 | [
"bug"
] | [] | 2023-07-21T18:07:48Z | 2023-07-25T00:11:49Z | 2023-07-25T00:11:49Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | taeyeopl | 50,736,858 | MDQ6VXNlcjUwNzM2ODU4 | User | false |
huggingface/diffusers | 1,816,246,715 | I_kwDOHa8MBc5sQbm7 | 4,200 | https://github.com/huggingface/diffusers/issues/4200 | https://api.github.com/repos/huggingface/diffusers/issues/4200 | Add flash attention 2 | Flash attention 2 helps in faster speeds, here is birch labs implementation in diffusers
https://gist.github.com/Birch-san/4315701264b72bb72e8eac5a529ee93a
@patrickvonplaten | closed | completed | false | 3 | [] | [] | 2023-07-21T18:47:56Z | 2023-07-25T01:10:36Z | 2023-07-25T01:10:36Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | adhikjoshi | 11,740,719 | MDQ6VXNlcjExNzQwNzE5 | User | false |
huggingface/diffusers | 1,816,313,735 | I_kwDOHa8MBc5sQr-H | 4,202 | https://github.com/huggingface/diffusers/issues/4202 | https://api.github.com/repos/huggingface/diffusers/issues/4202 | Decode latent to smaller size than 512x512? | Hello. I am currently conducting a research on image manipulation with Stable Diffusion, which involves CLIP score calculation.
I do start with latent of size 64x64, pass it through decoder (and get 512x512 image), resize it to 224x224 and then pass to CLIP.
However, there is one drawback: since I am calculating grad... | closed | completed | false | 2 | [] | [] | 2023-07-21T19:52:11Z | 2023-07-24T02:27:50Z | 2023-07-24T02:27:50Z | NONE | null | 20260407T133413Z | 2026-04-07T13:34:13Z | GLivshits | 47,557,983 | MDQ6VXNlcjQ3NTU3OTgz | User | false |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.