matheus58457/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:Oct 3, 2026License:otherArchitecture:Transformer Featherless Exclusive Cold

matheus58457/Qwen-Image-2.1-PE-I2I-Heretic is a 9 billion parameter image-editing prompt rewriter, derived from Qwen/Qwen-Image-2.1-PE-I2I. This model specializes in transforming short edit instructions and input images into detailed English edit prompts, with its refusal behavior significantly reduced through Heretic directional ablation. It maintains the original architecture and parameter count, making it suitable for non-commercial image-to-image editing applications where detailed prompt generation is crucial and reduced refusal is desired.

Loading preview...

Overview

matheus58457/Qwen-Image-2.1-PE-I2I-Heretic is a community-derived 9 billion parameter model based on Qwen/Qwen-Image-2.1-PE-I2I. It functions as an image-editing prompt rewriter, taking a concise edit instruction and one or more input images to generate a detailed English edit prompt. A key differentiator is the removal of refusal behavior using Heretic directional ablation, making it more permissive for certain content compared to its source model.

Key Capabilities

  • Image-to-Image Prompt Rewriting: Transforms brief edit instructions and visual input into comprehensive prompts for image editing.
  • Reduced Refusal Behavior: Ablated using Heretic to significantly lower refusal rates (9/100 compared to 100/100 for the original on harmful behaviors), while maintaining core prompt rewriting functionality.
  • Maintains Original Performance: Preserves the bf16 precision, shapes, and parameter count of the original Qwen-Image-2.1-PE-I2I, ensuring similar performance in prompt generation.
  • System Prompt Integration: Requires system_prompt.txt for defining output format, which is included and unmodified from the source.

Good for

  • Creative Image Editing Workflows: Ideal for applications requiring detailed prompt generation for image-to-image editing where the original model's refusal behavior might be restrictive.
  • Exploring Content Generation: Useful for researchers and developers experimenting with models that have reduced content moderation, particularly in image editing contexts.
  • Integration with Image Editing Apps: Designed to be used with image editing applications that benefit from enhanced and detailed prompts derived from user instructions and input images.

Finetunes
1 models