JoyAI-Image-Edit-ComfyUI (NVFP4 Quantized)

This repository provides NVFP4 quantized weights derived from the original BF16 weights of jdopensource/JoyAI-Image-Edit-ComfyUI. The conversion enables efficient usage of the JoyAI Image Edit model within ComfyUI.

Quantization Details

The quantization was performed using:

Quantization configuration file: config_nvfp4.json

Two weight variants are provided:

Filename	Description
`joyai_image_edit_nvfp4_uncalib.safetensors`	Uncalibrated (zero‑point free) NVFP4 weights
`joyai_image_edit_nvfp4.safetensors`	Calibrated NVFP4 weights

The calibrated weights were generated using CALIB_DATA.json – a calibration dataset produced by running only 3 editing tasks.

Important Notes

To use this quantized version, you must either:

Switch to the official pull request submitted by JoyAI:
Comfy-Org/ComfyUI#14428

or
Use my custom node repository:
judian17/ComfyUI-JoyAI-Image-Edit

Purpose

Reduce memory footprint and potentially improve inference speed on GPUs that support NVFP4.
Maintain compatibility with the original model’s behavior while offering a compressed alternative.
Provide an easy integration path for ComfyUI users.

Usage

Download the desired weight file (joyai_image_edit_nvfp4.safetensors or joyai_image_edit_nvfp4_uncalib.safetensors) from this repository.
Place the weights into the appropriate ComfyUI model directory (e.g., ComfyUI/models/joyai/).
Ensure your ComfyUI installation includes the required custom node or the specific PR.
Load the model using the corresponding node – it will automatically recognise the NVFP4 format.

License

Please refer to the original model’s license for terms of use.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support