Intel AI Playground Hands-on Play: Locally import third-party SDXL models to generate images and use AI chatbots

The AI ​​​​Playground launched by Intel has functions such as AI image generation, image editing, image enhancement, and chatbots We can experience its functions and performance through the AI ​​PC that has been built on Core Ultra 258V.

Can mount third-party SDXL modules

Intel AI Playground is a free downloadable resource, but requires a computer equipped with a Core Ultra-H series processor, Core Ultra 200V, or an Arc graphics card or graphics chip with more than 8GB of display memory. The current version provides three main functions: Create, Enhance, and Answer, which can perform AI image generation, image enhancement, Inpaint modification, image style modification, and Outpaint image enhancement, and which robots can establish chat through large language modules (LLM).

Further reading:
Intel Lunar Lake performance measurement: Asus Zenbook S 14 slightly increased performance, AI, battery life doubled!
Article Series Article Series Table of Contents Stable Release AI Grafting User Manual

Readers can go to the AI ​​Playground Github projectRelease pageFind the installation file of the corresponding version according to the hardware When you execute each function for the first time after installation, the program will automatically download the corresponding model.

In the image generation part, in addition to using the model preset by the program, you can also download the Checkpoint model of Stable Diffusion 1.5 or SDXL, Pony Diffusion, and a small LoRA model apply.

For example, readers may refer to “Tips for installing and using Pony Diffusion“The article, toCivitaI websiteDownload the model and store it in the corresponding custom path, and then you can use it in the AI ​​Playground.

After opening the AI ​​Playground program, users can use the AI ​​image generation function on the Create page, and click the gear icon in the upper right corner to display the model, resolution , iterative steps, random seed, set CFG control weight , generation size, and negative of the generated image.

With a Core Ultra 258V processor Asus Zenbook S 14For example, using the built-in display chip to generate an SDXL image with a resolution of 1024 x 1024 and 25 iterations, one image can be generated in about 38 seconds, which is considered acceptable in terms of speed.

▲ Introduction video of the AI ​​Playground function on Intel’s official channel.

▲ For example, the author uses a Core Ultra 200V series processor code named Lunar Lake, and can select the “LNL” version when downloading. If you are using a Core Ultra 1 series processor, you can download the “MTL” version, and for an Arc graphics card, select the “dGPU” version.

▲ After installing and running the program, you can use the AI ​​image generation function on the Create page The program will download the corresponding basic AI model the first time it is run put to death

▲ Users can also place third-party Checkpoint and LoRA models in Stable Diffusion 1.5, SDXL, Pony Diffusion and other formats into the corresponding packages and call them in the AI ​​Playground.

▲ Checkpoint and LoRA modules can be specified in the table. Although the official documentation says that only one group of LoRA can be used at a time, the author was not able to load it normally, probably because of a bug in the program.

▲ In addition, settings such as resolution, iteration steps, random seed, CFG control weight, generation size, negative prompts, etc. can be done in the table.

▲ Then go back to the main screen, enter the advanced password and click “Generate”.

▲ The AI ​​Playground is equipped with the personaStyle_v2Pony model and does not use sample graphics called LoRA.

▲AI function example of playground image generation function.

The chatbot also works offline

AI Playground Response page is a chatbot function similar to GhatGPT such as planes, and all All the data is kept on the local device, and there is no concern about data leakage.

The answer is also very simple to usePhi-3-Mini-4K-Instruct), then enter the content of the conversation in the prompt field, and the chatbot will respond based on the data in the model.

1

▲ The Answer function can choose from several large-scale language models It is recommended to choose lightweight ones.Phi-3-Mini-4K-InstructTo achieve faster response speed, the system will automatically load the model the first time it is used.

2

▲ After the user enters a question in the prompt field, the chatbot will respond based on the data in the model.

▲ Example of chatbot action functionality.

The author still encountered some minor problems using the AI ​​Playground, such as the inability to automatically download the image generation module (it can still be manually installed), or -ability to apply LoRA, and after installing the manually created device. image resolution value, you must press the Enter key to apply it When you enter quick words in the image generation function, you cannot use the Enter key to wrap the line (but you can text which contains external line breaks), and when loading the model to generate an image, you must be connected to the Internet (it can work normally when disconnected from the Internet after loading) ), but fortunately these problems are not serious and can be corrected through a software update.

Since AI Playground is free software and its operation method is very simple, interested readers may want to download and experience various AI activities for themselves.