darrellbest/Qwen-Image-2.1-PE-I2I-Heretic

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 22, 2026License:otherArchitecture:Transformer0.0K Featherless Exclusive Cold

darrellbest/Qwen-Image-2.1-PE-I2I-Heretic is a 9 billion parameter community derivative of Qwen/Qwen-Image-2.1-PE-I2I, fine-tuned as an image-editing prompt rewriter. This model processes short edit instructions and input images to generate detailed English edit prompts, with its refusal behavior significantly reduced through Heretic directional ablation. It is specifically designed to enhance image-to-image editing workflows by providing more descriptive prompts.

Loading preview...

Model Overview

This model, darrellbest/Qwen-Image-2.1-PE-I2I-Heretic, is a 9 billion parameter community-derived version of Qwen-Image-2.1-PE-I2I. It functions as an image-editing prompt rewriter, taking concise edit instructions and one or more input images to produce detailed English prompts for image manipulation. A key differentiator is the application of Heretic directional ablation, which significantly reduces the model's refusal behavior while maintaining its core functionality.

Key Capabilities

  • Prompt Rewriting: Transforms short, high-level edit instructions into comprehensive, specific prompts suitable for image editing applications.
  • Refusal Reduction: Achieves a substantial reduction in refusal rates (from 100/100 to 9/100 in tests) compared to the original model, particularly for chat-style harmful instructions, without compromising its ability to rewrite image edits.
  • Image-to-Image (I2I) Focus: Specifically designed for image-to-image editing workflows, distinguishing it from text-to-image (T2I) prompt rewriters.
  • Qwen Research License: Distributed under the Qwen Research License, permitting non-commercial use.

When to Use This Model

This model is ideal for developers and researchers who require an image-editing prompt rewriter with reduced refusal tendencies. It is particularly useful in applications where the original Qwen-Image-2.1-PE-I2I might exhibit unwanted refusal behaviors, especially when dealing with diverse or potentially 'edgy' image subjects. It ensures that detailed prompts are generated consistently, enhancing the control over image editing processes.

Finetunes
3 models