feat: Add multilingual prompt optimizer with LangGraph agent support

[ChristopheZhao]

What I'm trying to accomplish

This PR aims to enhance the text-to-image generation workflow by adding intelligent prompt processing capabilities. Specifically, it:

1. Makes the WebUI more accessible to non-English speakers by automatically detecting and translating prompts
2. Improves image generation quality through prompt optimization using LangGraph-based agent workflows
3. Provides a seamless experience that works within the existing txt2img interface

Summary of changes in code

• Added scripts/txt2img_prompt_optimizer.py - A new script that:

  • Implements a Script class that integrates with the WebUI's txt2img tab
  • Uses LangGraph to create an agent-based workflow for prompt processing
  • Detects non-English text and translates it to English
  • Optimizes prompts to improve generation quality while preserving intent
  • Handles API key management through environment variables
  • Provides graceful fallbacks when optional dependencies are missing

• Updated requirements.txt to include:

  • python-dotenv for environment variable management
  • langgraph for building the agent workflow

• Updated requirements_versions.txt with specific versions:

  • Added compatible versions of new dependencies
  • Ensured version compatibility with existing dependencies

• Updated .gitignore to exclude:

  • .env files containing sensitive API keys

Issues fixed

This PR addresses the feature request in Issue #4576, which requested multilingual prompt support but was previously marked as "not planned".

The implementation:

1. Adds multilingual support through automatic translation of non-English prompts
2. Goes beyond the original request by also implementing prompt optimization
3. Integrates seamlessly with the existing txt2img interface without requiring changes to the core pipeline

Screenshots/videos:

Here's a demonstration of how our system handles backend translations and their effectiveness for prompts in various languages. We will use 'a kitten under a pine tree' as a prompt to test the effects across different languages.

Chinese (simplified):

• backend

• frontend

Japanese;

• backend

• frontend

French;

• backend

• frontend

Spanish;

• backend

• frontend

Vietnamese.

• backend

• frontend

Kiswahili

• backend

• frontend

• And, of course, English prompts are also automatically optimized.
  • backend

• frontend

Checklist:

• I have read contributing wiki page
• I have performed a self-review of my own code
• My code follows the style guidelines
• My code passes tests