diffusers VS invisible-watermark

I'm missing something about the random factor, from the sample code from https://github.com/huggingface/diffusers/blob/main/README.md

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

Open Parti Prompts: The better way to evaluate diffusion models (repo)

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 .

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...

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.

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

I'm not sure your online tool is working. I tried it with the watermarked example image from https://github.com/ShieldMnt/invisible-watermark, and your tool returned that it did not detect a watermark:

The public availability and quality of LLMs and stable diffusion have been an unprecedented disaster for spam mitigation largely because there is no effective way to determine if this content was created and posted by a human being. Particularly with text content, the amount of information present is so small that I don't believe there is a way to definitively analyze it and concretely say whether or not it was generated by an LLM. The only potential way to do so that I can think of would be to check every comment against the output of each LLM service provider, but that's a futile endeavor because you can go back to inserting typos and substitutions, reorder the text or omit some of it, mash multiple outputs together, or even self-host an LLM and skip all the bullshit from the start. At least the images and videos being created by stable diffusion can be watermarked reasonably well.

i found an article but i don't understand it... is there any video tutorial of anything?

Creating invisible watermark encoder (see https://github.com/ShieldMnt/invisible-watermark)...

Tools to add this kind of watermarking are publicly available or could be reimplemented by in house developers if they don't like FOSS licenses.

Most major AI generators embed invisible watermarks into the images so that they can detect them and avoid training on generated imagery later. I know Stable Diffusion uses this python library to do it: https://github.com/ShieldMnt/invisible-watermark I haven't bothered to look up others but they have similar steps.

It uses the same library as Stable Diffusion (https://github.com/ShieldMnt/invisible-watermark) without giving credit in its github repository which does not even contain the sources of its 3 lines of code. This watermark doesn't protect anything, it would be necessary that the robots that retrieve the images from the internet make the effort to read the watermark to not add them in their dataset (best case scenario, totally utopian). The repository is suspicious and could be a way to install malware.

- Stable diffusion uses https://github.com/ShieldMnt/invisible-watermark by default unless you check "Do not add watermark to images" in settings

"This script incorporates an invisible watermarking of the outputs, to help viewers identify the images as machine-generated."