Simple ZImageTurbo workflow for LoRA use. — v1.0 for ZImageTurbo

Use this ZImageTurbo workflow to apply an iterative latent upscale to a normally generated base image, reusing the same model, LoRA(s), and strength to produce an upscaled image.

Generate the base image normally in 8–12 steps, then connect it to Iterative Latent Upscale (Impact Pack).

The described results are stronger LoRA style or character expression, sharper details and textures, and better facial features and micro-details without major composition changes.

The workflow declares these node packs: comfyui_essentials, comfyui-impact-pack, and rgthree-comfy.

Named model files include ae.safetensors (VAE), qwen_3_4b.safetensors (text encoder), Soyjak 2 z.safetensors (LoRA), and z_image_turbo_bf16.safetensors (UNet).

Try an upscale factor of 1.1–2 and denoise of 0.22–0.49; choose 1–3 steps for the effect you want. The author says they usually use 1 step.

Vorschlag

The workflow names Soyjak 2 z.safetensors as a LoRA; confirm that this is the file you want to use.

Vorschlag

No license is listed; check the applicable usage terms before reuse.

Vorschlag