generative-models
diffusers
generative-models | diffusers | |
---|---|---|
21 | 266 | |
22,649 | 22,881 | |
4.4% | 3.8% | |
7.3 | 9.9 | |
about 1 month ago | 4 days ago | |
Python | Python | |
MIT License | Apache License 2.0 |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
generative-models
-
Creating Videos with Stable Video Diffusion
git clone https://github.com/Stability-AI/generative-models.git && cd generative-models
- Show HN: I have created a free text-to-image website that supports SDXL Turbo
- How To Increase Performance Time on MacOS
-
Introducing Stable Video Diffusion: Stability AI's New AI Research Tool for Image-to-Video Synthesis
Generative Models by Stability AI Github Repository
-
image-to-video tutorial
# clone SD repo !git clone https://github.com/Stability-AI/generative-models.git # cd into working directory # the % sets the pwd globally as usually each command is run in a subshell in google colab %cd /content/generative-models/ # installing dependencies !pip install -r requirements/pt2.txt !pip install . # HACK # I was getting ModuleNotFoundError: No module named 'scripts' # This is what ChatGPT suggested (let me know if there is a better way) file_path = '/content/generative-models/scripts/sampling/simple_video_sample.py' new_text = "import sys\nsys.path.append('/content/generative-models')\n\n" with open(file_path, 'r') as file: original_content = file.read() updated_content = new_text + original_content with open(file_path, 'w') as file: file.write(updated_content) # Need to create a checkpoints/ folder - that is where the system looks for weights import os dir_name = 'checkpoints' if not os.path.exists(dir_name): os.makedirs(dir_name) print(f"Directory '{dir_name}' created") else: print(f"Directory '{dir_name}' already exists") # Download weights into checkpoints/ folder from huggingface_hub import hf_hub_download hf_hub_download(repo_id="stabilityai/stable-video-diffusion-img2vid", filename="svd.safetensors", local_dir="checkpoints", local_dir_use_symlinks=False) # I can't remember if this step is needed but it aims to reduce the memory footprint of pytorch # I kept getting CUDA out of memory # I got these instructions from the out of memory error message os.environ['PYTORCH_CUDA_ALLOC_CONF'] = 'max_split_size_mb:512' print(os.environ['PYTORCH_CUDA_ALLOC_CONF']) # Inside of scripts/sampling/simple_video_sample.py you need to make 2 updates 1. input_path (line 26): update to the location of your file (I attached Gdrive so mine was "/content/drive/MyDrive/examples/car.jpeg" 2. decoding_t (line 34): update it to 5. you need to do this for memory preservation (CUDA out of memory). I'm not sure if 5 is the best value but it worked for me # Finally generate the video (output will be in the outputs/ folder) !python scripts/sampling/simple_video_sample.py
-
Stable Video Diffusion
It looks like the huggingface page links their github that seems to have python scripts to run these: https://github.com/Stability-AI/generative-models
- GitHub - Stability-AI/generative-models: Generative Models by Stability AI
-
How does ComfyUI load SDXL 1.0 so VRAM-efficiently? How do I do the same in vanilla python code?
However, when using the example code from HuggingFace or setting up stuff from the StabilityAI/generative-models repo in a jupyter notebook, I end up using 21 GB of VRAM just for running the default pipeline (with no base model output). If I try to run the extra `base.vae.decode(base_latents)` after generation to get unrefined outputs, I get a CUDA out of memory error as it blows past the 24GB of my NVIDIA RTX 3090.
- SDXL 1.0 is out!
-
SDXL 0.9 Anyone having luck NOT centering subjects?
SDXL uses cropping information as part of the conditioning. Images were randomly cropped during training and the coordinates of the crop were included as two integers at the end of the conditioning vector. If you're using ComfyUI you can use the CLIPTextEncodeSDXL node to specify where the upper left corner of the image should appear to be in relation to some hypothetical uncropped image. Here's a figure with examples from the report on SDXL:
diffusers
- StableDiffusionSafetyChecker
- ๐งจ diffusers 0.24.0 is out with Kandinsky 3.0, IP Adapters, and others
-
What am I missing here? wheres the RND coming from?
I'm missing something about the random factor, from the sample code from https://github.com/huggingface/diffusers/blob/main/README.md
-
T2IAdapter+ControlNet at the same time
Hey people, I noticed that combining these two methods in a single forward pass increases the controllability of the generation quite a bit. I was kind of puzzled that sometimes ControlNet yielded better results than T2IAdapter for some cases, and sometimes it was the other way around, so I decided to test both at the same time, and results were quite nice. Some visuals and more motivation here: https://github.com/huggingface/diffusers/issues/5847 And it was already merged here: https://github.com/huggingface/diffusers/pull/5869
-
Won't you benchmark me?
Open Parti Prompts: The better way to evaluate diffusion models (repo)
-
kohya_ss error. How do I solve this?
You have disabled the safety checker for by passing `safety_checker=None`. Ensure that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered results in services or applications open to the public. Both the diffusers team and Hugging Face strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling it only for use-cases that involve analyzing network behavior or auditing its results. For more information, please have a look at https://github.com/huggingface/diffusers/pull/254 .
- Making a ControlNet inpaint for sdxl
-
Stable Diffusion Gets a Major Boost with RTX Acceleration
For developers, TensorRT support also exists for the diffusers library via community pipelines. [1] It's limited, but if you're only supporting a subset of features, it can help.
In general, these insane speed boosts comes at the cost of bleeding edge features.
[1] https://github.com/huggingface/diffusers/blob/28e8d1f6ec82a6...
-
Mysterious weights when training UNET
I was training sdxl UNET base model, with the diffusers library, which was going great until around step 210k when the weights suddenly turned back to their original values and stayed that way. I also tried with the ema version, which didn't change at all. I also looked at the tensor's weight values directly which confirmed my suspicions.
-
I Made Stable Diffusion XL Smarter by Finetuning It on Bad AI-Generated Images
Merging LoRAs is essentially taking a weighted average of the LoRA adapter weights. It's more common in other UIs.
diffusers is working on a PR for it: https://github.com/huggingface/diffusers/pull/4473
What are some alternatives?
background-removal-js - Remove backgrounds from images directly in the browser environment with ease and no additional costs or privacy concerns. Explore an interactive demo.
stable-diffusion-webui - Stable Diffusion web UI
wizmap - Explore and interpret large embeddings in your browser with interactive visualization! ๐
stable-diffusion - A latent text-to-image diffusion model
evernote-ai-chatbot
lora - Using Low-rank adaptation to quickly fine-tune diffusion models.
gping - Ping, but with a graph
invisible-watermark - python library for invisible image watermark (blind image watermark)
graphic-walker - An open source alternative to Tableau. Embeddable visual analytic
automatic - SD.Next: Advanced Implementation of Stable Diffusion and other Diffusion-based generative image models
xgen - Salesforce open-source LLMs with 8k sequence length.
Dreambooth-Stable-Diffusion - Implementation of Dreambooth (https://arxiv.org/abs/2208.12242) by way of Textual Inversion (https://arxiv.org/abs/2208.01618) for Stable Diffusion (https://arxiv.org/abs/2112.10752). Tweaks focused on training faces, objects, and styles.