Z-Image Turbo 12GB (Hi-Res, Facedetailer, & Sharpen)

This is what I use for most I2V source images. It's a fast & stable ZIT workflow designed for creators who want 2160 × 3240 ultra-sharp results on a 12GB GPU — in ~1 minute per render.

This workflow focuses on clarity, speed, and automation, integrating quality-of-life features like automatic face enhancement, VRAM cleanup, and controlled upscale/sharpen passes. Simple to use, creator-friendly, and capable of extremely high-detail output with consistent style adherence.

Fast Rendering (~1 minute)

Optimized node flow, memory-safe execution, and VRAM cleaning allow high-resolution output even on mid-range hardware. Can be used in batches.

---

Face Detailer (Ultralytics YOLO)

Automatic face detection + enhancement pipeline:

• YOLO face detection

• Precision mask

• Face-only resampling for maximum detail

• Zero manual masking required

---

LoRA-Ready

Full LoRA support with:

• Adjustable strength

• Clean model routing

• Safe scaling for style LoRAs or character LoRAs

Works out of the box with stylistic, realism, anime, or NSFW LoRAs.

---

VRAM & Cache Cleaner

Integrated memory recovery path:

• Clears tensor cache

• Recovers VRAM between stages

• Prevents fragmentation

• Helps maintain <12GB usage even at high resolution

---

2-Stage Upscale + Adaptive Sharpen

Output undergoes:

1. Controlled upscale pass for structure retention

2. Adaptive sharpen for crisp edges and detail emphasis

Final render = 2160 × 3240

---

Performance

GPU: 12GB
Render Speed: ~1 minute per image
Resolution: 2160 × 3240 final (post-upscale)

---

Requirements

• Updated ComfyUI

• ComfyUI-Essentials

• ComfyUI-Impact-pack

• ComfyUI-Impact-Subpack

• ComfyUI-Easy-Use

• Cache-Cleaner

• Rgthree-Comfy

• Checkpoint: unStable Revolution ZIT Bf16: /model/2193942/unstable-revolution-zit

• Clip: qwen_3_4b: https://huggingface.co/Comfy-Org/z_image_turbo/blob/main/split_files/text_encoders/qwen_3_4b.safetensors

• Vae: ae (flux): https://huggingface.co/ffxvs/vae-flux/blob/main/ae.safetensors

• Sam: sam_vit_b_01ec64: https://huggingface.co/datasets/Gourieff/ReActor/blob/main/models/sams/sam_vit_b_01ec64.pth

• Ultralytics: face_yolo8m: https://huggingface.co/xingren23/comfyflow-models/blob/976de8449674de379b02c144d0b3cfa2b61482f2/ultralytics/bbox/face_yolov8m.pt