victarry / stable-dreambooth Goto Github PK

original inversion has a trouble with that, it synhetsized mutations of the trained subjects unles you overfit and they you cant edit style anymore and composition that much.
Is it really like dreambooth and it retains identity ?

I have a similar question AttributeError: 'StableDiffusionPipeline' object has no attribute 'parameters' diffusers 0.15.0 can you help me solve it?

I was able to successfully train a model after chaning the diffusers version and changing batch size to 2 but when running inference on the output I only get reconstructions of the training images

Error is:

Traceback (most recent call last):
  File "train.py", line 206, in <module>
    train_loop(config, model, noise_scheduler, optimizer, train_dataloader)
  File "train.py", line 129, in train_loop
    latents = model.vae.encode(imgs).mode() * 0.18215
AttributeError: 'AutoencoderKLOutput' object has no attribute 'mode'

It seems related to the diffusers library not running on the gpu? I am in an environment with an a6000 though

Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
RuntimeError: CUDA out of memory. Tried to allocate 4.00 GiB (GPU 0; 47.46 GiB total capacity; 44.29 GiB already allocated; 862.56 MiB free; 45.46 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

How to get this working on an EC2 instance? any steps/tutorials?
Thank you.

File "train.py", line 206, in
train_loop(config, model, noise_scheduler, optimizer, train_dataloader)
File "train.py", line 131, in train_loop
noisy_latents = noise_scheduler.add_noise(latents, noise, timesteps.cpu().numpy())
File "/HPS/EgofaceTrial/work/anaconda3/envs/stable-diffusion/lib/python3.8/site-packages/diffusers/schedulers/scheduling_ddpm.py", line 303, in add_noise
timesteps = timesteps.to(original_samples.device)
AttributeError: 'numpy.ndarray' object has no attribute 'to'

It might be helpful to explain the needed images to train a new model.

The partial example with some images in data/dogs/instance is more confusing than it helps. Would it be possible for you to include an example training dataset?

Hi, similar to https://github.com/huggingface/diffusers/tree/main/examples/textual_inversion, would you be interested in adding example colab notebook for training and inference users the diffusers pipeline?

Maybe you guys should cooperate?

https://github.com/XavierXiao/Dreambooth-Stable-Diffusion

for text in datasets:
with torch.no_grad():
images = model(text, height=512, width=512, num_inference_steps=50)["sample"] ///
sample key work error in conda activate enviroment