37 questions
-3
votes
0
answers
26
views
T2I Adapter with SDXL produces only black images when conditioning on COCO-WholeBody skeletons (no error, loss decreasing) [closed]
I'm trying to fine-tune a T2I Adapter (full_adapter_xl) on COCO-WholeBody skeleton images, using Hugging Face Diffusers 0.33.0.dev0 + Stable Diffusion XL (stabilityai/stable-diffusion-xl-base-1.0).
...
0
votes
0
answers
81
views
IP-adapter plus face model not working as expected
I came from these two links,
https://huggingface.co/h94/IP-Adapter-FaceID
https://stable-diffusion-art.com/consistent-face/
They all mentioned I can preserve face id with the controlnet model.
So I ...
- 23.7k
3
votes
1
answer
10k
views
ImportError: cannot import name 'cached_download' from 'huggingface_hub'
huggingface_hub==0.27.1
diffusers==0.28.0
I am getting this error:
Traceback (most recent call last): File "/data/om/Lotus/infer.py", line 11, in <module>
from diffusers.utils ...
- 1,037
0
votes
1
answer
1k
views
ModuleNotFoundError: No module named 'diffusers.models.unet_2d_blocks'
when I use the diffusers in the https://github.com/alvinliu0/HumanGaussian project, I got the error :
Traceback (most recent call last):
File "launch.py", line 239, in <module>
...
- 1
0
votes
0
answers
125
views
Diffusers pipeline Instant ID with Ipadapter
I want to use an implementation of InstantID with Ipadapter using Diffusers library.
So far I got :
import diffusers
from diffusers.utils import load_image
from diffusers.models import ControlNetModel
...
- 502
0
votes
0
answers
73
views
Differences in no of ResNet blocks in up blocks and no of channels for Unet2D model of diffusers
I have been reading about Unets and Stable diffusion and want to train one. I understand the original architecture for unets and how its channels, height and width evolve over down blocks and up ...
- 21
2
votes
1
answer
288
views
Huge memory consumption with SD3.5-medium
I have a g4dn.xlarge AWS GPU instance, it has 16GB memory + 48GB swap, and a Tesla T4 GPU Instance with 16GB vRAM.
According to the stability blog, it should be sufficient to run SD3.5 Medium model.
...
- 23.7k
0
votes
0
answers
314
views
Stable Diffusion 3.5 Turbo extremely slow using diffusers library
Running example code directly from the huggingface stable diffusion 3.5 site link and I am getting extremely slow run times, averaging 90 seconds per iteration. For reference when I use Stable ...
1
vote
1
answer
273
views
Cannot merge Lora weights back to the Flux Dev base model
I have a Flux-Dev base model which has been trained with the LoRA technique using the SimpleTuner framework (https://github.com/bghira/SimpleTuner/blob/main/documentation/quickstart/FLUX.md). The ...
- 21
0
votes
0
answers
64
views
why unet forward takes whole GPU memory in every denoising loop
trying to write some toy example code of stable diffusion denoising without diffuser lib.
in diffusers examples :
https://huggingface.co/docs/diffusers/stable_diffusion
we just use the pipe style to ...
- 1,255
0
votes
0
answers
69
views
Run a pretrained Image-2-Image model using diffusers without CUDA?
I need to run a pretrained SD model locally for a small proof of concept project generating images out of outer images. I want to use the python diffusers library for this but if there are better ...
- 31
0
votes
1
answer
669
views
Flux.1 Schnell image generator issue, GPU resources getting exhausted after 1 prompt
So, I tried to train a prompt based image generation model using FLUX.1-schnell. I used Lightning AI Studio (an alternate to Google Colab), that helped me to access to L40 GPU, that came with 48gb ...
0
votes
0
answers
392
views
huggingface diffusers inference for flux in fp16
the Hugginface Flux documentation, links to this comment, describing how to run inference in fp16:
https://github.com/huggingface/diffusers/pull/9097#issuecomment-2272292516
it says:
FP16 ...
- 2,533
3
votes
2
answers
1k
views
Issue loading FluxPipeline components
import torch
from diffusers import FluxPipeline
pipe = FluxPipeline.from_pretrained('C:\\Python\\Projects\\test1\\flux1dev', torch_dtype=torch.bfloat16)
pipe.enable_sequential_cpu_offload()
prompt = ...
- 31
1
vote
1
answer
239
views
Shapes mismatch while training diffusers/UNet2DConditionModel
I am trying to train diffusers/UNet2DConditionModel from scratch. Currently I have error on unet forwarding: mat1 and mat2 shapes cannot be multiplied (288x512 and 1280x512). I noticed that mat1 first ...
- 11