Here’s what’s happening in AI this week! From breakthroughs to industry trends, we’re covering the biggest updates shaping the future of artificial intelligence.
Have a quick look at our AI news this week series,
1. Step Video: The Opensource Video Generation Model
- This model can generate very consistent, realistic, high-fidelity video from the text prompt.
- You can visit Step Video’s Hugging face and see the demo videos by clicking on the link. If you want to use it locally then download the model and use it from there
- Things to note and why it is better than other models:
- Previous Video generators had problems generating complex things like the fingers of humans but this doesn’t have that kind of problem.
- It can generate very complex scenes.
- It can also generate Text inside the video. Take a look at the Steve Jobs video; it has good, consistent text in the background.
- Dance video, people eating food, violin playing everything is fine.
- Pro: It is Completely Opensource
- Con: It requires a GPU with a minimum of 80GB of VRAM to generate a good video.
- You can use their web version but it is a Chinese app and requires mobile number signup. So, if you are okay giving your personal shit then you may sign up and use it.
How to use it Free?
If you trust China and their data policies then Here is a Web link for Step Video Generation. It is in Chinese language so use Google Translate. You can use it for Free there without downloading any model.
2. SynCD: Text to Image Customization FREE & Opensource
Forget the Product Photographers. You can edit consistent character and change the full background ever you want. Even edit the input image with specific modifications.
- What this can do?
So, you give any reference image for example you have a picture of a specific customized Toy in your room. So upload the image of it. Give it a prompt such as Your Toy on a Mount Everest. Then this tool can generate an image of your Toy on Mount Everest. - Here is the official SynCD website with all the codes and papers.
How to use it FREE?
You just have to click here to go to the official Hugging Face Demo of SynCD. Yes, it is Free to use.
Now just start if. Upload the image and give a small prompt that is it. See the Demo below. I uploaded Lemon as a character and typed prompt of Sunset and Coconut Tree. It created the same Lemon art on the beach.
3. ByteDance Phantom: Deepfakes and Advertising
It can generate video from a single image and text prompt. It can create hyper-realistic deepfake videos.
Here is the official link of ByteDance Phantom.
GitHub Link of Bytedance Phantom Video Generator (As of now, when I am writing this article they have only uploaded a research paper, models will be published soon for public use)
- It makes insanely accurate videos from just one reference image.
- It also works on products as well for example they have added a demo video of a person running in Nike shoes (only a single Nike shoes picture is given)
- So it will going to revolutionize the Advertising industry. You won’t need a professional photographer and a whole lot of editing team to record a commercial advertisement
- It works with Animals, Cartoon images, and Disney Pixar animated characters as well.
- You can also create a video with multiple reference images for example a specific person wearing specific clothes. So add two images and it will generate the video. Oh my god! this is so awesome. It will save a lot of cost.
How to use it Free?
Right now you have to there is NO option to use it on Hugging Face. Continuously visit the official site that I gave you for any updates
4. ART (Anonymous Region Transformer): Multi-Layer Image Generation
It is a tool that will replace Canva in the future. I am not talking about literal replacement. I am just saying that it can replace it because it can generate multi-layer images with just a single prompt.
So, if you are using Canva and creating some good illustrations. Then this could be the lifesaver. Let’s see if this can be the future of Canva or not.
How to use it Free?
Here is the official page link of ART: Anonymous Region Transformer
Note: When I was writing the blog their official Demo was showing an error so I was not able to record the generation.
5. Skyreels V1: Another Opensource AI Video Generator
Actually Skyreels is the finetuned model of the HunyuanVideo model. They are saying that it is finetuned on HD TV shows and other high-quality videos.
- As of now, one can only generate 544 x 960-pixel videos
- It can only generate a 3-4 second long video.
How to use it?
They are saying that it requires 24GB of VRAM. You can use Skyreels Playground (online platform) but it is paid. Just make the account and start using it.
But if you want to use it free of cost then you have to download it locally from HuggingFace and run it with your own GPU.
6. Alibaba WAN 2.1 Video Generator: It is Opensource
So, Alibaba a Chinese company launched their Text to Video generator. It also supports Image-to-video generation. Currently, Alibaba’s WAN 2.1 model can be used from their Qwen GPT which is a competitor of ChatGPT.
- Just look at the Demos, they can generate professional Dance Videos with complex motions
- You can generate Professional fighting scenes with proper expressions
- You can Replace any character from the video for example replace Lion with Bear.
- You can smoothly transition one video to another video for example from realistic to animation.
How to use it Free?
You can register and use their video generator, image generator, and GPT from a single platform. Here is the link for Alibaba’s Qwen GPT.
If you want to use Alibaba’s WAN 2.1 locally on your device. you can download it from GitHub.
Conclusion:
That is it for today’s article. I will write an updated article soon for the next AI news.
