Neura Market
PromptEnhancer logo

PromptEnhancer

Free

[CVPR 2026] PromptEnhancer is a prompt-rewriting tool, refining prompts into clearer, structured versions for better image generation.

Image GeneratorsFreeFree tier
Inputs: textOutputs: text
Type
Open Source
Visit Website GitHub (3.7k stars)

About PromptEnhancer

PromptEnhancer is a universal prompt-rewriting framework designed to improve the alignment of text-to-image (T2I) diffusion models without modifying their weights. It uses a Chain-of-Thought (CoT) approach, trained via a two-stage process: first supervised fine-tuning (SFT) on prompt-reprompt pairs, then reinforcement learning (GRPO) guided by a dedicated reward model called AlignEvaluator. The AlignEvaluator provides fine-grained feedback based on a taxonomy of 24 key points covering common T2I failure modes (e.g., attribute binding, negation, compositional relationships). PromptEnhancer significantly enhances image-text alignment on models like HunyuanImage 2.1 and includes a new human preference benchmark.

Key Features

Universal framework: enhances any pretrained T2I model without weight modification
Chain-of-Thought (CoT) prompt rewriting trained via supervised fine-tuning and reinforcement learning
AlignEvaluator reward model providing fine-grained feedback based on a 24-point taxonomy of T2I failure modes
Two-stage training: SFT initialization followed by policy alignment using GRPO
Improves image-text alignment in attribute binding, negation, and compositional relationships
Includes a new high-quality human preference benchmark for T2I alignment research

Pros & Cons

Pros
  • Works with any pretrained T2I model without requiring model modifications
  • Uses structured Chain-of-Thought reasoning for more precise prompt rewriting
  • Reinforcement learning aligns rewriter with downstream image-text alignment
  • Fine-grained reward model based on comprehensive failure mode taxonomy
Cons
  • Requires training the AlignEvaluator reward model, adding complexity
  • Benchmark and experiments focused on HunyuanImage 2.1; generalization to other models may vary
  • Online demo appears to be placeholder only (Lorem ipsum text)

Best For

Improving fidelity of text-to-image generation for complex user promptsEnhancing attribute binding and compositional relationships in generated imagesAddressing negation and complex semantic requirements in image generationServing as a drop-in improvement for existing T2I models without retraining

Alternatives to PromptEnhancer

ComfyUI

The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface.

Awesome-Prompt-Engineering

This repository contains a hand-curated resources for Prompt Engineering with a focus on Generative Pre-trained Transformer (GPT), ChatGPT, PaLM etc

AI-Youtube-Shorts-Generator

A python tool that uses GPT-4, FFmpeg, and OpenCV to automatically analyze videos, extract the most interesting sections, and crop them for an improved viewing experience.

awesome-gpt4o-images

Awesome curated collection of images and prompts generated by GPT-4o and gpt-image-1. Explore AI generated visuals created with ChatGPT and Sora, showcasing OpenAI’s advanced image generation capabili

awesome-generative-ai

A curated list of Generative AI tools, works, models, and references

awesome-ai-painting

AI绘画资料合集（包含国内外可使用平台、使用教程、参数教程、部署教程、业界新闻等等） Stable diffusion、AnimateDiff、Stable Cascade 、Stable SDXL Turbo

FAQ

What is PromptEnhancer?
PromptEnhancer is a universal prompt-rewriting framework that enhances any pretrained text-to-image model by rewriting user prompts using Chain-of-Thought reasoning, trained via reinforcement learning guided by a dedicated alignment reward model.
How does PromptEnhancer work?
It uses a two-stage training process: first, a Chain-of-Thought rewriter is initialized via supervised fine-tuning on prompt-reprompt pairs. Second, the rewriter is refined using reinforcement learning (GRPO) where the AlignEvaluator reward model scores generated images for alignment, optimizing the rewriter to produce prompts that maximize image-text matching.
Does PromptEnhancer modify the T2I model?
No. PromptEnhancer is a universal framework that enhances any pretrained T2I model without requiring changes to its weights. It only rewrites the input prompt.
What are the key improvements over other methods?
Unlike model-specific fine-tuning or CLIP-based rewards, PromptEnhancer decouples the rewriter from the generator and uses a fine-grained reward model based on a 24-point taxonomy of common T2I failure modes, leading to better handling of attribute binding, negation, and compositional relationships.