## Model links
text_encoders
qwen_3_4b.safetensors
Model Storage Location
```
📂 ComfyUI/
├── 📂 models/
│ ├── 📂 text_encoders/
│ │ └── qwen_3_4b.safetensors
│ ├── 📂 loras/
│ │ └── qinglong_detailedeye_z-imageV2(comfy).safetensors
│ ├── 📂 diffusion_models/
│ │ └── z_image_turbo_bf16.safetensors and z_image_turbo-Q4_K_S.gguf
│ └── 📂 vae/
│ └── ae.safetensors
```
Read the notes in the workflow for more help, or join the Discord community:
🎮 Join Discord : https://discord.gg/DZEenb5wGc
🔗 Get the Pro Version of This Workflow with More tweaks and enhancements on My Patreon.
Description
🆘 Having Issues?
🎮 Join Discord : https://discord.gg/DZEenb5wGc
🔗 Get the Pro Version of This Workflow with More tweaks and enhancements on My Patreon.
FAQ
Comments (6)
worked great but is it possible to use a non realistic model or only realistic ones? seemed like all the color got washed out when i was expecting it to give that anime type effect.
Tested Only with Realistic Models, but it will work with Anime too.
Try this :
Load anime Model and in the XL Lora Loader add the DMD2 Lora at 0.8 Strength.
That it.
@Stable_Yogi ty I will give it a go in the morning.
The workflow, well-designed and functional, works perfectly with an RTX 3070 Ti 8GB. Once the models are loaded and the first image is generated (which takes the most time), generating subsequent images takes between 90 and 100 seconds (which is reasonable). Ultimately, it's a matter of preference, but I generally prefer the image generated by Z-Image to the one modified by SDXL (characters, eyes, fingers). Another observation, which is in no way your fault, as it's consistent with all the Z-Image workflows I've tested so far: a high-resolution image generated directly by Z-Image seems more efficient in terms of quality and detail than gradual upscaling. Best regards.
Thanks for the feedback
Could you please share the workflow? I couldn’t fully figure out how to set up SDXL detailers for higher-quality and sharper generation. Or is this already considered post-processing?