ZZZ Style LoRA for Qwen-Image-2.1

A Zenless Zone Zero-inspired style LoRA trained on top of Qwen-Image-2.1. The trigger word is zenlesszonezero (optional — the style also transfers without it). Works for both text-to-image and image-to-image (edit) — see Edit results. base model is qwen_image_2.1_int8_convrot.safetensors from comfyui.

---

🎨 Edit Results (img2img)

Prompt used for every pair below:

re render this image in zenlesszonezero_style

The input image is passed to the model as-is; the LoRA restyles it in place — character identity, pose and composition are preserved while palette, lighting, linework and costume rendering shift to the ZZZ look.

Example 0 — Photorealistic input → ZZZ style

_{Photo → anime. Identity, pose, crop and backdrop are held; every rendering decision (linework, cel shading, costume design, colour grade) is replaced by the ZZZ look.}

Example 1 — Hu Tao (Genshin Impact)

_{Input → Output. Same character and pose, ZZZ night-time grade.}

Example 2 — Frieren (Frieren: Beyond Journey's End)

_{Input → Output. Proportions round out, palette cools, background goes painterly.}

Example 3 — Asuna (Sword Art Online)

_{Input → Output. Flat cel shading becomes soft volumetric 3D-ish rendering.}

Full-resolution individual files _{Source images are third-party fan art of characters owned by their respective publishers; used here as style-transfer inputs for demonstration. Generated content is not necessarily typical of the source material in every respect — see [Limitations](#-limitations).}

---

🎨 Text2Image

_{All six images are 1024×1024 and were generated with this LoRA in text-to-image mode. Generated content is not necessarily typical of the source material in every respect — see [Limitations](#-limitations).}

---

📋 Model Details

---

📦 Checkpoint Guide

recommanded text2image: v1.6,v1.7,v1.4

recommanded image edit: v1.7>v1.6

🚀 Usage

Prompt template (text-to-image)

zzz_style, furina from genshin impact, waving hands

Prompt template (edit / img2img)

re render this image in zenlesszonezero_style

Feed the source image as the init/reference image. A denoise strength around 0.5 – 0.7 restyles aggressively while keeping pose and identity; go lower (0.3 – 0.5) for a subtler grade that preserves more of the original linework.

Recommended LoRA scale

---

ComfyUI

Base model files (diffusion_models + text_encoder + vae):

• qwen_image_2.1_int8_convrot.safetensors — diffusion model
• qwen3vl_8b_int8_convrot.safetensors — text encoder
• qwen_image_2.1_vae_bf16.safetensors — VAE

Sampler settings (KSampler):

For edit mode, use QwenImageEdit / QwenImageEditPlus with the input image loaded into Load Image and the prompt above; the same sampler settings apply.

---

Negative prompt

None required — leave it empty. If quality wobbles, try blurry, low quality, jpeg artifacts.

---

⚠️ Limitations

1. The training set is dominated by single-girl portraits, so multi-character scenes and male characters generalise poorly.
2. The base model used for training was int8 quantised (convrot8).
3. Generated content may not be typical of the source material in every respect.
4. In edit mode, very low input resolution (below ~512 px on the short side) yields soft, smeared results — upscale the source first if needed.
5. Strong denoise on edit will drift from the input composition; back the denoise off if identity fidelity matters more than the style shift.

---

📄 License & Attribution

Released under Apache-2.0.

• Base model — Qwen-Image-2.1 (Alibaba Qwen team)
• Training framework — ai-toolkit (ostris)
• Art style reference — the art team of Zenless Zone Zero by miHoYo
• Edit demo inputs — fan art of Genshin Impact (HoYoverse), Frieren: Beyond Journey's End (Kadokawa/Soyobusho) and Sword Art Online (A.C.N.T.), each © its respective owner; Example 0 additionally uses a photorealistic portrait supplied by the uploader

Zenless Zone Zero is a trademark of miHoYo / HoYoverse. This LoRA is a third-party fan work, not affiliated with or endorsed by miHoYo.