# Z-Image Turbo FP8/BF16 Community Workflow

## Summary

Community `Z-Image Turbo` text-to-image workflow packaged with a same-day embedded ComfyUI graph. It is a practical fast baseline rather than a new control trick.

## Model Family

- `zimage`

## Status

- `imported-only`

## Source

- Hugging Face model page: <https://huggingface.co/Saravutw/ZImageTurbo-NSFW-FP8-bf16>
- Workflow file: <https://huggingface.co/Saravutw/ZImageTurbo-NSFW-FP8-bf16/raw/main/z_image_turbo12stepwrokflow.json>
- Original author: `Saravutw`
- Date imported: `2026-08-26`

## Developer notes

- The workflow uses the standard ComfyUI core nodes plus `ModelSamplingAuraFlow`.
- The graph targets `z_image_turbo_bf16.safetensors`, `qwen_3_4b.safetensors`, and `ae.safetensors` from the official Z-Image Turbo bundle.
- The embedded graph keeps the familiar turbo settings: low CFG, `dpmpp_sde`, `simple` scheduler, and 12 steps.

## Our notes

- This is worth keeping because it is a direct workflow payload, not just a model card.
- On an `RTX 3070`, it should be a comfortable local test path for fast image generation.
- It does not look like a major leap over the official Z-Image Turbo example, but it is a clean same-day variant worth having in the repo.

## Required custom nodes

- none beyond current core ComfyUI support for the embedded graph

## Required model files

- `models/text_encoders/qwen_3_4b.safetensors`
- `models/diffusion_models/z_image_turbo_bf16.safetensors`
- `models/vae/ae.safetensors`

## Notes

- The workflow can be loaded by dragging the JSON into ComfyUI.
- Good as a fast local baseline, but not yet a substitute for stronger control or edit workflows.