Qwen-Image is more sensitive to prompts. If conflicting prompts are used, it will degrade the image generation quality. Please keep this in mind, everyone.

---

To better showcase the characteristics of the LoRA, subsequent LoRA cover sample images will use identical parameters, making it easier for everyone to compare.

---

When running online, please set the CFG value between 2 and 5. The lower the CFG value, the more details, but the lower the prompt adherence.

When running offline, you can consider using the distilled acceleration model for Qwen-Image. Below are the resource links:

1. Accelerated Workflow (4 Steps): https://www.liblib.art/modelinfo/6fbc9a0d81394360812307cc59f4ac44

2. Qwen-Image Distilled Acceleration Model: https://www.liblib.art/modelinfo/cfe5c999b01f4e0ca3531819217c8d71

---

Current LoRA style: Poster

Reprinted from: StarDream｜Qwen-Image｜Photography Style｜Poster