custom-diffusion
diffusers
custom-diffusion | diffusers | |
---|---|---|
11 | 266 | |
1,787 | 22,763 | |
2.0% | 3.3% | |
5.6 | 9.9 | |
5 months ago | 6 days ago | |
Python | Python | |
GNU General Public License v3.0 or later | 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.
custom-diffusion
-
ELITE: new fine-tuning technique that can be trained in less than a second
I think https://github.com/adobe-research/custom-diffusion
-
What's the best technology for training faces these days?
I attempted Custom Diffusion which, too, did not yield anywhere near as photorealistic face outputs as Dreambooth.
- [Discussion] Stable Diffusion Models with Subject/Keyword References
-
Suggestions for creating prompts with two people that you've added via fine-tuning?
Curious if anyone has better experiences? I've also been trying Adobe's Custom Diffusion code, which probably works the best out of Textual Inversion and Dreambooth, but the code they provide is super buggy and very hard to use with Automatic1111.
-
Version 0.1.0 of LoRA released! (alternative to Dreambooth, 3mb sharable files)
Well, actually it is https://github.com/adobe-research/custom-diffusion. It's Adobe, so this is probably only once in the lifetime :P
-
How would I go about creating an app like "Lensa" with Stable Diffusion?
A Few Comments - The "caveat" I mentioned above is that there are a few apps that are running Stable Diffusion locally on the user's device. Apple recently released a tool to convert Stable Diffusion models to CoreML. CoreML is their proprietary format for machine learning models, and runs insanely well on Apple devices with a Neural Engine (like newer Macs, iPhones, and iPads). However, this technology is in its infancy, almost certainly isn't capable of "training" a Stable Diffusion model, and isn't anywhere near as fast as running Dreambooth or Stable Diffusion on powerful servers. In the long-run, it might be possible to do all of this processing on a user's device, but it's likely that we're a long ways away from that. - Dreambooth itself isn't that hard to run and play around with yourself, nor is it that hard to integrate into an automated server pipeline, though what it does under the hood is pretty amazing. Dreambooth isn't the only way to "train" a Stable Diffusion model with custom photos, and other people/companies (like Adobe), have found other ways to create amazing AI-generated images with user-provided photos (see their Custom Diffusion GitHub). - Given how long Lensa has been around, and that they're pretty decently funded (they raised $6m in 2019, I believe), it's very likely they they've developed their own in-house way of training Stable Diffusion models, just like the Adobe reference above. But if any of us were to build an app that works like Lensa, a starting point would probably be to use Dreambooth since it's well-built out and easy to integrate, and get similar results. - A very popular way to run Dreambooth is to use a Google Colab notebook, like this one from TheLastBen's GitHub. Since the vast majority of us don't have super powerful GPU cards in our computer, and are just playing around with Dreambooth/Stable Diffusion, the Google Colab notebook lets you go step-by-step to "setup" an environment for doing Dreambooth, but runs it on Google's super-powerful servers. The cool thing about Google Colab, besides not having to have a super-powerful computer yourself to still do Dreambooth with great performance, is that you can look through the code of how the Google Colab notebook works, and that could be a foundation for an engineer to learn how to implement Dreambooth in your own scripts that run on the "Backend Server" for your app.
-
How does tiktok’s AI portrait filter work?
If it is Stable Diffusion-based, I'd guess a few things. Given the small size that would be needed to handle this for every possible user, I wonder if it's using something that isn't Dreambooth based, like Adobe's [Custom Diffusion(https://github.com/adobe-research/custom-diffusion), or some in-house variant of that, that's able to basically generate a small file that could be processed for each user.
-
How to get the smallest models or portions of Dreambooth-trained models for a specific subject
Adobe Research also has "Custom Diffusion" out: https://github.com/adobe-research/custom-diffusion. It's got a similar goal of ~megabytes sized outputs. Warning that it's got a proprietary license.
- Custom Diffusion - Adobe Research
- What's the difference between custom-diffusion and Dreambooth?
diffusers
- StableDiffusionSafetyChecker
- 🧨 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?
lora - Using Low-rank adaptation to quickly fine-tune diffusion models.
stable-diffusion-webui - Stable Diffusion web UI
sd-webui-additional-networks
stable-diffusion - A latent text-to-image diffusion model
LoRA - Code for loralib, an implementation of "LoRA: Low-Rank Adaptation of Large Language Models"
configDB - A database like , dynamic config file generator to enable more customization in your python project , from games to ML notebooks and everything between.
invisible-watermark - python library for invisible image watermark (blind image watermark)
WaveDiff - Official Pytorch Implementation of the paper: Wavelet Diffusion Models are fast and scalable Image Generators (CVPR'23)
automatic - SD.Next: Advanced Implementation of Stable Diffusion and other Diffusion-based generative image models
LLM-Adapters - Code for our EMNLP 2023 Paper: "LLM-Adapters: An Adapter Family for Parameter-Efficient Fine-Tuning of Large Language Models"
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.