---
title: ZImageBase-nvfp4_mixed
canonical_url: "https://www.modelscope.cn/models/ApacheOne/ZImageBase-nvfp4_mixed"
md_url: "https://www.modelscope.cn/models/ApacheOne/ZImageBase-nvfp4_mixed.md"
repository: ApacheOne/ZImageBase-nvfp4_mixed
last_updated: 2026-04-19
license: apache-2.0
model_type:
  - custom
base_model:
  - notaneimu/z-image-base-comfy-fp32
base_model_relation: quantized
parameters: 14.9B
tensor_type:
  - F32
  - U8
  - F8_E4M3
  - BF16
library_name:
  - safetensors
  - pytorch
frameworks:
  - pytorch
language:
  - en
downloads: 147
stars: 1
tags:
  - quantization
  - nvfp4
  - art
---

# ZImageBase-nvfp4_mixed

> ZImageBase-nvfp4_mixed - ApacheOne 在 ModelScope 开源的模型。nvfp4 model infomation

ApacheOne/ZImageBase-nvfp4_mixed 是 ModelScope 魔搭社区上的 14.9B 参数机器学习模型，采用 apache-2.0 许可，基于 notaneimu/z-image-base-comfy-fp32 构建。

- **Repository**: ApacheOne/ZImageBase-nvfp4_mixed
- **License**: apache-2.0
- **Parameters**: 14.9B
- **Base model**: notaneimu/z-image-base-comfy-fp32
- **Tags**: quantization, nvfp4, art
- **Downloads**: 147
- **Stars**: 1
- **Last updated**: 2026-04-19

Source: https://www.modelscope.cn/models/ApacheOne/ZImageBase-nvfp4_mixed

---

## nvfp4 model infomation

`z_image_base_raw_fp8_e4m3fn.safetensors` Fp8 model

BF16 model from here: [Civitai](https://civitai.red/models/2533927/famegrid-zib-checkpoint?modelVersionId=2847800)

`ZImageBase-nvfp4_TurboloraV3.safetensors` runs in comfy with cfg 1 and 8 steps

Note: Model is bigger than fp8, still working on convert proflie


Note: This model has not been personally tested. Feedback is appreciated.

`ZImageBase_nvfp4_FP32.safetensors` Comfy format from : [https://huggingface.co/notaneimu](https://huggingface.co/notaneimu/z-image-base-comfy-fp32)

note: might need to use --fp32-unet flag in ComfyUI

Note: Custom FP32 mixed nvfp4 version of base ZImagebase

note: find out how many offsets we can tune to get better outcomes