Learn the most basic operation of discodiffusion

The goal is

Disco Diffusion It's a AI Drawing tools , One sentence can make AI Draw a picture that matches the description , You can see many amazing results on the Internet .

I also want to try this magical technology , The goal of this article is to learn the basic operation of using it .
The main reference tutorial is ：Quick & Easy Tutorial for Disco Diffusion Google Colab (Make your own pictures!). - YouTube

Because it is in Google Colab On , So if you know nothing about it , It is best to have a brief understanding of the relevant basic concepts （ see The previous ）

0. Basic information

The Disco Diffusion Of GoogleColab The address is ：
Disco Diffusion v5.4 [Now with Warp] - Colaboratory
（ The version of the original tutorial is 4.1, But I made a mistake after trying , So it ended up with this 5.4 edition ）

---

And then , You can copy it to your own GoogleDrive In the network disk .

Then you can open it in your network disk . The follow-up operation of this article is also done in your own network disk .

---

just as The previous said , It is Jupyter Notebook, So all you have to do is configure the basic settings in this document , Then run the document .

The directory structure of the document is as follows ：

The next step is to make some basic settings .

1. Set up ： name 、 They count 、 Size

These settings are in Settings in

Here are three basic parameters note ：

bathch_name： The name of this run , It is also the name that will be saved in the cloud disk later

steps： Number of steps to run . The more steps, the more detail , Examples given in the tutorial ：

Of course , The higher the number of steps, the longer it will take , So you need to find the most appropriate number of steps for a picture .

width_height： Picture length and width . Must be 64 Multiple .

2. Set up ： Save interval

This is set at Settings Medium ExtraSettings in ：
intermediate_saves Save the control center several times . For example, I fill in here 9, In addition to seeing the last result on my cloud disk , You can also see the 9 Zhang .

3. Set up ： The core Prompts

It's in Settings Medium Prompts in .

Of course , He is the core . For this parameter , The description in the document is ：

text_prompts： A description of what you’d like the machine to generate. Think of it like writing the caption below your image on a website. A description of what you want the machine to generate . It can be imagined that , If this picture appears on the website , So what is its title .

for example , I am here ：
“a huge panda is destroying alien base”, “Technology and Fantasy”
“ Giant pandas are destroying alien bases ”,“ Technology and fantasy ”

4. Set up ： Show interval , Number of batches

These settings are in Diffuse in ：

display_rate ： Just a few steps apart Colab It shows that
n_batches： How many batches are calculated in total （ That is, output several completely different pictures ）

5. Began to run ！

At the top , find “ All running ” The button ：

Then some prompts will pop up .
such as

There are also some man-machine verifications

There's a connection GoogleDrive A hint of ：

When it's all done , Start running the code in this document .
It will take a few minutes to wait for the data that the installation depends on .
Then you can see the progress of the calculation at the bottom ：

6. result

After completion of operation （ About experienced 40 minute ）, You can find the output pictures in your network disk ：

（ among partials Folders store intermediate processes ）

Evaluate the result ：
Feeling “ Alien base ” The feeling is that ,“ Panda ” This “ huge ” You can also see the characteristics of , Black and white can also be seen , But the panda's own details , A little too abstract （ laugh ）.

summary

0. This is based on GoogleColab Of , So it's best to understand the basic concepts ,（ see The previous ）
1. To DiscoDiffusion Of GoogleColab Address , Copy it to your own GoogleDrive In the network disk .
2. Open... In your network disk DiscoDiffusion copy , Make the most basic settings , The core setting is prompts.
3. Click on the run , Then you can do it in your own GoogleDrive See the result in the network disk .

Other references

Sina weibo @Simon_ aven There are a lot of relevant information in , This tutorial is also seen here .
Artificial intelligence painting tools Disco Diffusion Introductory tutorial - Bili, Bili It is also a similar Chinese tutorial , You can refer to .