GoJun AI Daily

English

简体中文

English

简体中文

Appearance

Spend 1 minute every day to get the latest AI information.

The content includes but is not limited to cutting-edge AI news, AI tools, AI painting, open-source projects, and learning tutorials.

The main feature of the brief is its concise description, but important information is still introduced in detail through independent posts.

Here is the latest AI information for July 2.

Cutting-edge Technology ​

1. Runway has opened Gen3 for use!

Currently, it only supports text-to-video generation, available for paid users. The effects are comparable to Luma and Kelan, each with its own pros and cons.

Runway is mainly too expensive, costing $1 per 5-second video. Kelan is still more attractive.

Official website: https://runwayml.com/

However, today a user achieved character consistency with Gen3, maintaining the same character across multiple generations.

image-20240702231349788

2. Microsoft quietly updated Phi3-mini.

The model's capabilities have been improved in various aspects, such as instruction-following ability and coding ability.

Model download: https://huggingface.co/collections/microsoft/phi-3-6626e15e9585a200d2d761e3

image-20240702231845162

Cutting-edge Technology ​

1. Tencent open-sourced a video generation project with specified actions: MinmicMotion.

Similar to Alibaba's AnimateAnyone, it generates action videos of characters based on specified actions and character images.

The effects seem much better than Alibaba's, with consistent facial features and lip-syncing. It can be used not only for dance videos but also for digital humans.

Project introduction: https://tencent.github.io/MimicMotion/

GitHub: https://github.com/tencent/MimicMotion

Unofficial ComfyUI workflow: https://github.com/AIFSH/ComfyUI-MimicMotion

model architecture

AI Painting ​

1. Someone spent 3 months creating a graphic novel entirely using SD.

The author's production tutorial shared:

  • The painting model is iComix;
  • Mixing famous actors in the prompts to maintain facial consistency;
  • ControlNet Reference model for clothing consistency;
  • ControlNet OpenPose model to control character poses;
  • Using Photoshop to handle dialogue bubbles.

Original post: https://www.reddit.com/r/StableDiffusion/comments/1dpo14t/i_finally_published_a_graphic_novel_made_100_with

image-20240702005718821

Learning Tutorials ​

1. A step-by-step guide to building an AI video generation model from scratch.

Found a tutorial on GitHub where the author explains in detail how to use Python to build a text-to-video generation model from scratch.

It covers the entire process from understanding theoretical concepts to architecture coding, and finally achieving video generation from text prompts.

GitHub: https://github.com/FareedKhan-dev/AI-text-to-video-model-from-scratch

Worth checking out if interested.

AI-text-to-video-model

Open-source Projects ​

1. WebDesignAgent: an agent that helps you build a series of websites.

Supports converting text descriptions, images, and visual cues into fully functional, beautifully designed websites, simplifying the website creation process with AI.

Detailed introduction: https://t.zsxq.com/3EIE3

img