CogVideo VS diffusers

Additionally two open source demo models [CogVideo[(https://github.com/THUDM/CogVideo) by a groups of cs students a model by [Antonia Antonova](https://antonia.space/text-to-video-generation) and have presented their own innovative methods of generating video from text

The second thing that will happen is the appearance of AI video and audio. Google has shown two programs for video generation, one which is fairly high quality and the other which can make long videos with several scenes. Meta has also demonstrated their own. We've already seen other projects like CogVideo, as well as many others that are currently being worked on. It's likely that these techniques will become so refined that over the next year or two, they'll have a similar boom to image generation programs. And eventually, they'll have a similar application in video editing, once coherence is adequate enough. Select a person's shirt, and it stays that for the remainder of the scene. Change an actor's hairstyle in real time, or add characters that didn't exist into a scene and let the computer figure out the desired level of realism. This'll revolutionize VFX to a degree where making an effects heavy will be less about wrangling complex toolsets and more about making aesthetic choices of style and placement.

Check out https://github.com/THUDM/CogVideo - progress is being made on coherent video generation.

Characters and dialogue are effectively solved, just look at GPT-3.

The entity behind StableDiffusion is also supporting generative music art, so let's see what is coming out of that: https://www.harmonai.org/

We are currently far away from generating a production quality movie with AI, but I don't think it's going to be nearly as long as a lifetime. In my opinion, we'll have high quality AI shorts within the decade.

CogVideo and a CogView web app.

A better comparison to AI music generation would be video generation, which has not improved much since i saw first examples some years ago. The last iteration is stuff like CogVideo and this is only able to generate 4 second videos with mid-strong artifacts.

github: https://github.com/THUDM/CogVideo

GitHub repo.

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