Tensor Works Reference Forge — Z-Image Img2Img Workflow

Tensor Works Reference Forge is a ready-to-launch Z-Image Turbo img2img environment for creators who already train character LoRAs and want a cleaner way to produce social-ready images in the cloud.

You do not rebuild ComfyUI, hunt for custom nodes, or manage private GitHub tokens. Launch the Tensor Works RunPod template, upload your LoRA, load the included workflow, drop in a reference image, and generate.

The stack is preconfigured for character LoRA workflows: ComfyUI, required custom nodes, automatic Z-Image model download, JupyterLab file access, and persistent-volume caching so you are not re-downloading everything every session.

WHAT YOU CAN DO

• Copy poses, framing, and vibe from a reference photo onto your AI character

• Generate realistic images with Z-Image Turbo

• Use your own character LoRA and trigger word

• Upload LoRAs through JupyterLab or pull them from Hugging Face with one terminal command

• Iterate quickly on cloud GPUs without maintaining a local install

WHO IT IS FOR

Built for AI creators, persona builders, agencies, and content teams who already have a character LoRA and want a reliable cloud workflow for image-to-image production.

Works best with a trained character LoRA (.safetensors), a clear trigger word, and a clean reference image (pose / scene you want copied).

WHAT YOU GET

• TensorWorks_Z_Image_Reference_Forge.json — the full ComfyUI workflow

• TensorWorks_Reference_Forge_Setup_Guide.html — professional step-by-step setup guide

• Direct RunPod template link for one-click deploy

• Access to Tensor Works Discord support

WHAT YOU NEED

• A RunPod account and credits

• A GPU with 48 GB VRAM or more (recommended)

• Your character LoRA

• A reference image to copy

RECOMMENDED HARDWARE

• Best: RTX PRO 6000 (96 GB), RTX 6000 Ada, A6000, L40S

• Volume: 120 GB+ persistent network volume at /workspace

• Lower VRAM cards may fail at higher resolutions or require reduced settings

IMPORTANT NOTES

• RunPod usage is billed separately through your RunPod account

• First launch on a new volume downloads Z-Image models (one-time per volume)

• Output quality depends on your LoRA, reference image, trigger word, and settings

• Character LoRA training is not included — you bring your own LoRA

SUPPORT

Questions or stuck on deploy? Join our Discord:

https://discord.gg/kwZvdUKE6