The DALL-E3 in the ChatGPT paid version and API fee usage is considered to be commercially available, but when using DALL-E3 on the free version of Micosoft Copilot, the licence is not explicitly stated, and the mainstream interpretation is that commercial use is not possible.

In this workflow, using the SDXL(※Default version) and Flux1 Schnell(※Latest version) model, which can be used commercially, the LLM creates prompts from simple Japanese instructions and generates images based on these prompts... and then the LLM generates images based on the prompts... This time, we used ComfyUI to create an image generation AI with functionality almost similar to the paid version of DALL-E3.

We have also created several nodes using the OLLAMA application so far,

This time, too, I made use of this and designed the LLM part to be self-contained in the local environment. In other words, the concept is to make the LLM part free of charge (apart from the cost of the PC itself and electricity)... The concept is to make the LLM part of the system free of charge (apart from the cost of the PC itself and electricity).

For the base model, download the 'Boltning Realistic HYPER D' model from Civitai's website here. You can change it to your own preference.

The SDXL Hyper is a high-speed SDXL model developed by ByteDance, a Chinese company well known for its TokTok products, and can generate even higher quality images than the SDXL Lightning model developed by the company.

The default settings of KSampler are also optimised for SDXL Hyper.

In 24 September 2024, I added the latest version fixed Flux1 Schnell.

As shown in the diagram below, enter simple your native language. instructions in the 'CR Text' field and click 'Queue Prompt', and the LLM built on the ollama application will create the prompt and generate the image.

However, the LLM must correspond to your native language.

In DALL-E3, four images are generated for one prompt, but in this node, two images are created.