Qwen3.5 9B Heretic 2.1 – INT8 / W4A8 ConvRot
About
This is a quantized and converted version of Qwen3.5-9B-Heretic, originally developed by bambocher.
Intended Use
This model is primarily designed for use as a text-processing component in image-generation pipelines, rather than as a standalone image-generation model.
1. Text Encoder
Suitable for workflows that use Qwen3.5 as a text encoder, such as:
Moss Image
Other compatible Qwen3.5-based workflows
2. Prompt Enhancer
Can be used as an alternative prompt enhancer for image-generation models, including:
Qwen Image 2.1
MiniMax H3 Image
Other workflows using Qwen-based prompt enhancement
Why Use This Model?
If you are using an official model for prompt enhancement and occasionally encounter responses similar to:
"I am unable to generate content involving *** in this manner or *** explicit descriptions as it violates safety policies."
You may want to try this model as an alternative prompt enhancer.
The Heretic-based model is designed to reduce refusal behavior compared with the original model, making it potentially more suitable for creative prompt enhancement workflows.
Actual results may vary depending on the workflow, prompt, and inference settings.
Credits & Original Model
Original model: bambocher / Qwen3.5-9B-heretic
Original model based on Qwen3.5-9B and modified using Heretic v2.1.0.
All credit for the original model and its underlying work goes to the respective creators.
My contribution: INT8 ConvRot and W4A8 ConvRot conversions.
