How to Build a Custom Midjourney Style Model with GPT‑Powered Prompt Engineering

By calling model code, the generated images maintain high stylistic consistency and aesthetic quality.

The tutorial explains how the powerful MJ model is created, using a simple example and offering a free style model library at the end.

Prompt Engineering with GPT : Convert the Chinese theme "gift box, Christmas" into detailed English prompts, breaking down possible visual elements. Generate five distinct prompt groups, each with at least ten keywords and the style tag Minimalist . The prompt format follows

subject + details + tags + parameters

, where parameters are fixed as

--ar 3:4 --c 25.0 --s 750 --v 5.2

.

Example prompt:

Gift Boxes, details, brilliant colors, full sun, award‑winning, photograph shot with Kodak Portra 800, a Hasselblad 500C, 55mm f/1.8 lens, extreme depth of field, available light, high contrast, Ultra HD, HDR, DTM, 8K --ar 3:4 --c 25.0 --s 750 --v 5.2

After generating prompts with GPT, adjust them on a prompt‑tool platform, then copy the final prompts into Discord for model training.

Training steps in Discord:

Enter the command

/tune xxx

.

Paste the prompt, removing any extra command text to avoid errors.

Confirm the training dataset size (32 images is sufficient; 64 or 128 for richer styles, with longer training time).

Click

submit

.

Confirm and wait a few minutes.

Once training completes, a link appears to open a new webpage for model training.

Select 1‑5 high‑quality images for the MJ style model; using too many images may produce odd styles.

The resulting model excels at expressing overall texture and style but is not suited for fine‑grained local adjustments; for special needs, combine MJ with Stable Diffusion.

All style model codes are stored in an online spreadsheet, and readers are encouraged to share their own style codes in the comments for community exchange.