Update README.md

f3374ad verified 8 days ago

5 kB

base_model:
  - aoxo/flux.1dev-abliteratedv2
library_name: gguf
pipeline_tag: text-to-image
tags:
  - gguf
  - flux
  - flux.1-dev
  - quantized
  - image-generation
  - text-to-image
  - comfyui
  - comfyui-gguf
  - abliterated
license: fair-noncommercial-research-license

Flux-12B Abliterated V2 — GGUF Q4_K_M

A GGUF Q4_K_M quantization of aoxo/flux.1dev-abliteratedv2, prepared for local inference with ComfyUI and ComfyUI-GGUF.

This repository provides a compact quantized version of the Flux.1-dev Abliterated V2 model while preserving the original model's image-generation capabilities as closely as possible.

This is a quantized conversion, not the original model. Please refer to the original repository for the source model, license, and full model information.

✨ Model

Base model: aoxo/flux.1dev-abliteratedv2

Architecture: FLUX.1-dev / 12B

Format: GGUF

Quantization: Q4_K_M

Task: Text-to-Image

Recommended environment: ComfyUI + ComfyUI-GGUF

🚀 Why GGUF?

GGUF provides a convenient way to run large models locally with reduced memory requirements.

The Q4_K_M quantization provides a strong balance between:

🧠 Model quality
💾 File size
⚡ Inference performance
🖥️ VRAM requirements

GGUF is supported by a growing ecosystem of local inference tools and includes model metadata directly inside the file.

🖥️ ComfyUI

This model is intended primarily for use with ComfyUI through ComfyUI-GGUF.

Requirements

ComfyUI
ComfyUI-GGUF
A compatible Flux workflow
Sufficient system RAM / VRAM for Flux inference

Installation

Install ComfyUI.
Install the ComfyUI-GGUF custom node.
Download the .gguf file from the Files and versions section of this repository.
Place the GGUF model in the appropriate model directory used by your ComfyUI-GGUF setup.
Load a Flux workflow that supports GGUF models.
Select the downloaded Q4_K_M model.

The exact folder may depend on your ComfyUI-GGUF configuration and workflow.

📦 Quantization

This repository currently provides:

Quantization	Format
Q4_K_M	GGUF

Q4_K_M is a 4-bit K-quantization variant designed to provide a useful compromise between model size and quality. Hugging Face documents Q4_K as a 4-bit quantization family, with K-quant variants designed around block/super-block quantization.

🎨 Usage

The model is designed for local image generation.

A typical workflow is:

Prompt
   ↓
Flux Text Encoder
   ↓
Flux GGUF Q4_K_M
   ↓
Sampler
   ↓
VAE
   ↓
Generated Image

Use your preferred Flux.1-dev workflow and replace the transformer/model component with the GGUF version where supported.

⚙️ Performance

Actual performance depends heavily on:

GPU model
Available VRAM
System RAM
ComfyUI configuration
Resolution
Number of sampling steps
CPU/GPU offloading
Flux workflow

Because of this, VRAM and generation-speed figures are intentionally not specified here unless they have been measured on a particular hardware configuration.

🧪 Model Status

This is an unofficial GGUF conversion of the original abliterated Flux model.

The goal of this release is to make the model more accessible to users who want to run it locally with a smaller memory footprint.

If you encounter:

corrupted output,
loading errors,
unexpected artifacts,
incorrect tensor loading,
or other compatibility problems,

please open an issue or leave feedback on the repository.

📜 Credits

Original model

aoxo/flux.1dev-abliteratedv2

https://huggingface.co/aoxo/flux.1dev-abliteratedv2

All credit for the original model and its underlying work belongs to the original authors.

GGUF conversion

anlord

This repository contains the GGUF quantized conversion published by anlord.

⚠️ License

The original model is distributed under:

Fair Noncommercial Research License

Please read and comply with the original model's license before downloading, using, modifying, or redistributing this model.

The license of the original model takes precedence over any assumptions made from this README.

⭐ Feedback

If this model is useful to you, consider leaving a ⭐ on the repository.

Feedback, generation examples, bug reports, and suggestions for additional quantizations are welcome.

🔗 Links

This GGUF: https://huggingface.co/anlord/Flux-12B-abliterated-V2-GGUF
Original model: https://huggingface.co/aoxo/flux.1dev-abliteratedv2
Hugging Face GGUF documentation: https://huggingface.co/docs/hub/gguf

About this release

Flux-12B Abliterated V2 — GGUF Q4_K_M

A community GGUF quantization prepared by anlord for convenient local inference.