prithivMLmods/Qwen-Image-2.1-PE-T2I-Heretic-v1.0

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

prithivMLmods/Qwen-Image-2.1-PE-T2I-Heretic-v1.0 is a 9-billion parameter text-to-image prompt-rewriting model based on Qwen-Image-2.1-PE-T2I, which itself is a fine-tuned Qwen3.5-VL-9B. This model specializes in transforming brief, multilingual image requests into detailed, expanded English prompts with a recommended aspect ratio, outputting a structured JSON object. It significantly reduces refusal rates (from 97% to 22%) compared to its base model, making it highly effective as a preprocessing step for downstream image generators like the Qwen-Image-2.1 DiT-based model.

Loading preview...

Overview

prithivMLmods/Qwen-Image-2.1-PE-T2I-Heretic-v1.0 is a specialized 9-billion parameter model designed for text-to-image prompt rewriting. Built upon the Qwen-Image-2.1-PE-T2I architecture, which is a fine-tuned Qwen3.5-VL-9B, this model acts as a crucial intermediary for image generation pipelines. Its primary function is to take concise, often multilingual, user requests for images and expand them into rich, detailed English prompts, complete with an optimal aspect ratio, formatted as a JSON output.

Key Capabilities

  • Multilingual Prompt Expansion: Transforms short image descriptions in any language into elaborate English prompts.
  • Aspect Ratio Recommendation: Automatically suggests a wh_ratio (e.g., 16:9) suitable for the generated image, directly usable by Diffusers pipelines.
  • Structured Output: Produces a JSON object containing the rewritten_prompt and wh_ratio for seamless integration with image generators.
  • Reduced Refusal Rate: Significantly improves usability by lowering the refusal rate from 97% (in the original Qwen-Image-2.1-PE-T2I) to just 22%.

Good For

  • Preprocessing for Image Generation: Ideal for use cases where user input for image creation is brief or in non-English languages, requiring detailed prompts for optimal results from models like Qwen-Image-2.1.
  • Improving User Experience: Enables users to provide simple requests without needing to craft complex, detailed prompts themselves.
  • Integration with Diffusers: Designed to output directly compatible JSON for QwenImage21Pipeline in Diffusers.

Finetunes
1 models