No need to spend money to bring the robot home, the open source AI suite satisfies you

Artificial intelligence is so popular, but it’s a pity. Services are either not allowed to be used, blocked, or neutered. Workers in the new era are very confused.

While you are busy with various internal tests and tinkering with accounts, the open source AI industry is already quite prosperous.

In fact, with a PC, you can build various services without being connected to the Internet, and take the robot home. The only thing that makes people feel helpless is that if you want to play happily, you still need a more powerful graphics card.

I have just experienced the baptism of mining, and now I have to accept the lash of AI. Graphics cards (especially N-cards) are really bad at the wrong time.

(Note: The pictures in this article are all drawn by AI)

Robot Painting

stable-diffusion -webui

When it comes to robot painting, the most popular one is of course Stable Diffusion. We generally use webui as the front-end interface.

https://github.com/AUTOMATIC1111/stable-diffusion-webui

The installation of webui is very simple. You only need to clone the contents of the above warehouse and install according to the documentation.

Support:

1. Windows, run webui-user.bat
2. Linux & Mac, run webui.sh

depends The file will be downloaded automatically.

The basic functions of Stable Diffusion are as follows:

1. Generate pictures according to the prompt words (spells), which can be combined in any way.
2. Generate similar images based on the provided images and prompt words.
3. The image is partially fine-tuned and new elements are added.
4. Support mission training, having your own virtual wife is not a dream.
5. Supports a lot of extensions.
6. Massive models.

Other recommendations

In fact, for computer painting, a stable-diffusion-webui is enough. If you want to use it well, you need more resources.

1. Massive models, free download.

https://civitai.com/ You can download large models of various styles from Station C without registration. For example, to draw a girl in Hanfu, you only need to download a checkpoint model.

https://huggingface.co/ Of course, you can also get more models from the github of the machine learning industry.

Commonly used models include: Chilloutmix, Counterfeit, YesMix, RealDosMix, etc.

2. Generated prompt words

To obtain better images, prompt words are the most critical.

https://stablediffusionweb.com/prompts On the official website of sd, you can see a large number of similar prompt words. You can find what you like and fine-tune it.

https://tag.redsex.cc/ For those who are not good at English, here are some key words for drawing character models. Of course, there are R18 restricted vocabulary. After unlocking, you can wirelessly generate S-pictures without having to visit any website to expose your identity.

https://civitai.com/ It is also the C station. Most people also share the prompt words when sharing pictures. You can also look for them here.

3. Drawing workflow

In the webui interface, the content disappears after refreshing. If you want to save your own workflow, try https://github.com/comfyanonymous/ComfyUI.

If you have used blender and other software, the experience of using ComfyUI will be very smooth.

4. NovelAI

In addition, I have to mention NovelAI. In fact, it is also developed based on SD, but the two-dimensional pictures generated are more beautiful, so it has become an independent software.

However, webui has more complete content and more models, and many two-dimensional models are also very effective.

AI Chat

ChatGPT detonated the field of artificial intelligence. But in fact, SD in the field of graphics developed earlier and took the open source route.

Now, more GTP open source models have been released. We can definitely build one locally.

Just like the llama.cpp I shared last time.

For a person who speaks Chinese, this is obviously not enough. So there are the following two.

ChatGLM-6B

ChatGLM-6B is an open source conversational robot released by the Knowledge Engineering and Data Mining Group of Tsinghua University. Although this model is not large, it is similar to ChatGPT. It is optimized for dialogue and has a good user experience.

https://github.com/THUDM/ChatGLM-6B

Combined with model quantization technology, users can deploy it locally on consumer-grade graphics cards (a minimum of 6GB of video memory is required at the INT4 quantization level). Of course, if you have a Macbook m1 pro, this can also run quickly.

Chinese-LLaMA-Alpaca

Didn’t the llama I shared last time only support English? Here, there is one that supports Chinese. All of these are evolved from Meta's LLama model. It can be said that Meta Company has also made a lot of efforts in the field of artificial intelligence open source.

https://github.com/ymcui/Chinese-LLaMA-Alpaca

显然，中文模型相对于英文模型，也很小。但回答常识性的问题，已经耍你家里那什么小度小爱小E，千八百条街了。

作者简介：小姐姐味道  (xjjdog)，一个不允许程序员走弯路的公众号。聚焦基础架构和Linux。十年架构，日百亿流量，与你探讨高并发世界，给你不一样的味道。

The above is the detailed content of No need to spend money to bring the robot home, the open source AI suite satisfies you. For more information, please follow other related articles on the PHP Chinese website!