How to use this workflow
Download the JSON above and open it in ComfyUI. The file contains the graph and saved settings; model weights, custom node packages and example input files are separate. Expand its subgraphs to inspect the complete pipeline before running it.
1. Load a reference
Replace the saved LoadImage filename with an image on your machine. ImageScaleToMaxDimension uses Lanczos with a maximum dimension of 1024. Check the reference crop before generation.
2. Inspect the depth guide
The AIO_Preprocessor is set to DepthAnythingV2Preprocessor at 1024. Inspect its preview to confirm that the main foreground and background shapes are represented. A poor depth guide can limit the usefulness of subsequent prompt changes.
3. Configure generation
Expand the Z-Image-Turbo Fun Control to Image subgraph. It requests z_image_turbo_bf16.safetensors, qwen_3_4b.safetensors, ae.safetensors and Z-Image-Turbo-Fun-Controlnet-Union.safetensors. Edit the text prompt to describe the desired result while retaining the structural reference you want.
4. Run and compare
The saved sampler uses 9 steps, CFG 1.5, res_multistep and beta scheduling. Run a baseline before changing these values. SeedVR2 then upscales the generated image, and SaveImage uses the Z-Turbo_Control/Z-Turbo_Control prefix. Compare both the depth preview and final output when assessing structure.
Missing nodes, models or memory errors?
If a node is missing, identify its exact type in the graph and install the matching custom node package from its maintainer. Restart ComfyUI after installation. If a loader cannot find a model, check the filename and loader selection against your installed files. For an out-of-memory error, reduce image dimensions or clip length before retrying. No minimum VRAM or render-time benchmark has been measured for these downloads.
Use the official ComfyUI documentation for installation and node guidance. These instructions describe the bundled JSON, rather than promising compatibility with every ComfyUI version.
Explore related workflows
Z-Image Turbo text-to-image with SeedVR2 upscaling · Qwen Image Edit 2511 · Qwen Image Edit multi-angle camera · LTX 2.3 image-to-video
