models:

Ultra Flux 4k VAE {https://huggingface.co/Owen777/UltraFlux-v1/blob/main/vae/diffusion_pytorch_model.safetensors} OR [ae.safetensors] provided in ComfyUI's default template workflow
Jib Mix ZIT [fp16] {https://civitai.com/models/2231351/jib-mix-zit?modelVersionId=2511897}
qwen_3_4b.safetensors can be downloaded from the base Z-Image-Turbo workflow in ComfyUI
HI8 [@0.04] {https://civitai.com/models/1134895/2000s-analog-core?modelVersionId=2456725}
Olympus [@0.20] {https://civitai.com/models/2212121/olympus-ultrareal?modelVersionId=2490572}

3 step seed variance - same prompt/seeds = same image

custom nodes:

please make sure you delete the example prompt before generating - custom node installation will require restarting ComfyUI multiple times! DEFAULT OUTPUT IS 1920x2160 (960x1080)

more loras can be found here on civitai, remove all of the disabled loras in the stack if they give you errors. experiment with adding your own but make sure to keep the seed fixed while you adjust the individual strength weights one at a time - most loras excel at very low amounts (0.54-0.03)

depending on your hardware you will get very different results

links to required models:

Diffusion Model - https://civitai.com/models/2231351/jib-mix-zit?modelVersionId=2511897

place in ComfyUI/models/diffusion_models

VAE - https://huggingface.co/Owen777/UltraFlux-v1/blob/main/vae/diffusion_pytorch_model.safetensors

place in ComfyUI/models/vae

LoRAs

HI8 [@0.04] {https://civitai.com/models/1134895/2000s-analog-core?modelVersionId=2456725}
Olympus [@0.20] {https://civitai.com/models/2212121/olympus-ultrareal?modelVersionId=2490572}

place in ComfyUI/models/loras

these images are not upscaled, they're all generated by the workflow. 1 generation takes about 2 minutes with 12GB of VRAM

it's easy to crop off the bottom of the image where the generation goes wonky, plan ahead in choosing canvas size or use photoshop's outpainting after cropping for example.

example images no longer accurate - see links above for current workflow LoRA models

Lora Loader Workflow (Z Image Turbo)