This workflow is designed to give you full control over image generation, whether you’re working with ControlNet, LoRAs, and/or upscaling.

It uses a dual sampler setup with the RES4LYF pack to fix the quality issues ControlNet has when paired with LoRAs.

Main Features

Simple generation

Standard Z-Image Turbo image generation

Fun-ControlNet V2.1 (2026)

Supports multiple control conditions (Canny, Depth, Pose, MLSD, HED, Scribble, and Gray).

SAM3 segmentation

For high-accuracy character/object pose reference isolation.

LoRA fix

LoRA hooks for variable LoRA strengths during generation. The dual sampling process injects your LoRAs at a low strength during the first steps to guide the model without degrading the image, while the full strength of your LoRAs is applied in the final steps, resulting in a high-quality image with strong character consistency.

You get 8 LoRA slots at your disposal.

Lora used for this demo: Woman037 by Huslyo123

Detail control

Smoothness control and Detail Daemon give you full control over the details in your final image.

4K Upscaler

Ultra high-quality upscaling with SeedVR2.

Detailed instructions are contained within the workflow itself:
- Yellow nodes are input and configuration nodes you can change to suit your needs.
- Red nodes are instructions and helpful notes