MultiDiffusion
sd_lite
MultiDiffusion | sd_lite | |
---|---|---|
13 | 15 | |
911 | 18 | |
- | - | |
4.8 | 4.5 | |
8 months ago | about 1 year ago | |
Jupyter Notebook | Python | |
- | MIT License |
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.
MultiDiffusion
-
Opendream: A Non-Destructive UI for Stable Diffusion
For composing this approach works pretty well
https://multidiffusion.github.io/
-
Messing with the denoising loop can allow you to reach new places in latent space. Over 8+ different research papers/Auto1111 extension ideas in a single pipe. Load once and do lots of different things (SD 2.1 or 1.5)
So I've continued to experiment with how many papers I can fit into a single pipe and have them play nicely together. The images below were created by combining the panorama code from omerbt/MultiDiffusion with the ideas from albarji/mixture-of-diffusers. Also turns out nateraw/stable-diffusion-videos can be seen as a special case of a panorama (in latent space rather than prompt space).
- MultiDiffusion Region Control, a prompt on each mask webui extension is out.
-
Hubble Diffusion with MultiDiffusion
Essentially, I fine-tuned Stable Diffusion 2.1 base (the 512x512) model on the ESA Hubble Deep Space Images & Captions dataset I collected from public Hubble images & captions. After around 33,000 training steps, I saved the model and was really impressed by the results. But I really wanted to be able to generate wallpaper-level quality space images, so I stumbled upon MultiDiffusion: a new project for generating massive panorama images using stable diffusion models. I then used hubble-diffusion-2 along with MultiDiffusion to generate each one of these amazing 2560x1536 images. Each image took a little over an hour to generate on a Google Colab T4 GPU. I used the following prompts for each of these images:
- MultiDiffusion: Fusing Diffusion Paths for Controlled Image Generation
-
What is the maximum size a 3090 24gb can produce?
If you need generated and not upscaled 4k for some reason, try something like https://github.com/omerbt/MultiDiffusion
-
[R] [N] "MultiDiffusion: Fusing Diffusion Paths for Controlled Image Generation" enables controllable image generation without any further training or finetuning of diffusion models.
Project: https://multidiffusion.github.io/ Paper: https://arxiv.org/abs/2302.08113 GitHub: https://github.com/omerbt/MultiDiffusion
-
Meet MultiDiffusion: A Unified AI Framework That Enables Versatile And Controllable Image Generation Using A Pre-Trained Text-to-Image Diffusion Model
Quick Read: https://www.marktechpost.com/2023/02/24/meet-multidiffusion-a-unified-ai-framework-that-enables-versatile-and-controllable-image-generation-using-a-pre-trained-text-to-image-diffusion-model/ Paper: https://arxiv.org/abs/2302.08113 Github: https://github.com/omerbt/MultiDiffusion Project: https://multidiffusion.github.io/
-
You to can create Panorama images 512x10240+ (not a typo) using less then 6GB VRAM (Vertorama works too). A modification of the MultiDiffusion code to pass the image through the VAE in slices then reassemble. Potato computers of the world rejoice.
So I haven't made many images with Stable Diffusion despite using it heavily. The reason is I've been messing with the internals of the diffusion pipe, to interfere with the diffusion process in different ways. Todays fun result is based on omerbt/MultiDiffusion for making panoramas.
-
First version of Stable Diffusion was released on August 22, 2022
If we combine Mixture of Diffusers + MultiDiffusion+ Composer+ cross-domain-compositing and probably some more I'm not thinking of.
sd_lite
- List of Stable Diffusion research softwares that I don't think gotten widespread adoption.
- Comparing 5 recent SD distillation methods SSD/LCM/Turbo to find the best option for low-VRAM users (images and statistical analysis included). SD-Turbo scores significantly higher on aesthetics, the boost to SD-21 is remarkable
-
Latent Jitter: a simple method for generating variations on a prompt to composite into a final image. Stacks well with prompt delay and The Stable Artist to give you 4+ options from a single seed/prompt.
The full details of how to do this are available on Github: latent jitter · thekitchenscientist/sd_lite but I will explain the idea briefly here. I have read this could be done with perlin or simplex noise but the code was too complex for my taste. This gets the job done with only minor modifications to the standard pipe.
-
"SEGA: Instructing Diffusion using Semantic Dimensions": Paper + GitHub repo + web app + Colab notebook for generating images that are variations of a base image generation by specifying secondary text prompt(s). In this example, the secondary text prompt was "smiling". See comment for details.
I did successfully swap the effiel tower for the burj Khalifa but that required additional steps https://github.com/thekitchenscientist/sd_lite/wiki/latent-jitter
-
Are there any sure-fire 100% SFW models for Stable Diffusion? Project for kids
I use it in my pipe as a general image beautifier. https://github.com/thekitchenscientist/sd_lite/wiki/safe-latent-diffusion
-
Messing with the denoising loop can allow you to reach new places in latent space. Over 8+ different research papers/Auto1111 extension ideas in a single pipe. Load once and do lots of different things (SD 2.1 or 1.5)
The pipe is available at sd_lite/pipeline_stable_diffusion_multi.py (github.com) it combines:
-
Comparison of new UniPC sampler method added to Automatic1111
This community has published many XY plots of CFG versus steps. https://github.com/thekitchenscientist/sd_lite/wiki/recommended The consistent theme is low CFG, lower steps; high CFG, more steps. UniPC can reach convergence in as few as 8 steps, so I increased by 1/3 to account for more complex prompts needing longer
-
Create Panorama images of ANY size using less then 6GB VRAM, also x6-10 speed-up and added support for batch mode! A modification of MultiDiffusion. Potato computers of the world rejoice. SD2.0 768 model gives fastest creation of larger sizes but the VAE image slicing means no VRAM spike.
the pipeline is available from github.com and is called in the usual way. The Technique requires the DDIM scheduler.
-
Img2Img as a side-scrolling enhancer - more pictures in the comments
https://github.com/thekitchenscientist/sd_lite is where the code is. Version 1 of the multi-pipe is limited to images 512 high or wide but any number on the other dimension
-
You to can create Panorama images 512x10240+ (not a typo) using less then 6GB VRAM (Vertorama works too). A modification of the MultiDiffusion code to pass the image through the VAE in slices then reassemble. Potato computers of the world rejoice.
Not to be deterred I hacked together some code to blend it all back together after the VAE but before the final colour balance. The pipe code is available on github. thekitchenscientist/sd_lite
What are some alternatives?
stable-diffusion-webui-two-shot - Latent Couple extension (two shot diffusion port)
sd-dynamic-prompts - A custom script for AUTOMATIC1111/stable-diffusion-webui to implement a tiny template language for random prompt generation
sd-webui-controlnet - WebUI extension for ControlNet
sd-dynamic-thresholding - Dynamic Thresholding (CFG Scale Fix) for Stable Diffusion (StableSwarmUI, ComfyUI, and Auto WebUI)
mixture-of-diffusers - Mixture of Diffusers for scene composition and high resolution image generation
safe-latent-diffusion - Official Implementation of Safe Latent Diffusion for Text2Image
Diffusion-Models-Papers-Survey-Taxonomy - Diffusion model papers, survey, and taxonomy
ziplora-pytorch - Implementation of "ZipLoRA: Any Subject in Any Style by Effectively Merging LoRAs"
stable-diffusion-videos - Create 🔥 videos with Stable Diffusion by exploring the latent space and morphing between text prompts
erasing - Erasing Concepts from Diffusion Models
openpose-editor - Openpose Editor for AUTOMATIC1111's stable-diffusion-webui
Concurrent-gif2gif - Experimental Automatic1111 Stable Diffusion WebUI extension, concurrent frame rendering