AI Prompt Assistant - Pro

GITHUB-PROJECT-LINK

Try Online Demo (ver 1.2.0)

Features

This application is provided as an .exe executable file. It is safe to use and does not require administrator privileges to run. The program operates entirely within standard user permissions.

What's New in v1.4.0

Highlights

Interactive Proactive Co-Creator

An intelligent prompt co-creation interface inspired by proactive co-creator research. Rather than just running standard chat, this mode helps you decompose, analyze, and enrich your creative ideas.

• Belief Graph Decomposition — Automatically parses your prompt into structured entities, visual attributes, and relationships.

• Atomic Clarifying Questions — Dynamically generates 3 focused, single-topic questions with multiple choice options to resolve ambiguities.

• Proactiveness Levels — Configure the AI's creativity level (Low, Medium, High). Higher levels infer details and suggest creative additions.

• Global Context & Diversity — The engine avoids assuming Western norms and actively suggests diverse ethnicities, visual styles, and cultural details.

• Reference Image Grounding — Uses attached images as a "source of truth", mapping visible details and asking clarifying questions on intended modifications.

• Interactive UI Cards — Adjust entities, select alternative values, and answer clarifications directly via a rich interactive panel inside the chat bubble.

Unified API Settings Studio

Manage all AI providers from a centralized, beautiful configuration interface.

• Centralized Dashboard — Configure API keys, base URLs, and parameters for all local and cloud-based LLM backends.

• Expandable Cards & Visibility Toggles — Clean layout with expandable cards and toggles to show/hide providers from the sidebar.

• Comprehensive Provider List — Native support for Google Gemini, Vertex AI, OpenAI, Anthropic, Mistral, OpenRouter, Groq, Together, SwiftRouter, NVIDIA NIM, Ollama, LM Studio, Koboldcpp, and the Free Provider.

Prompt Library & Local Seed Service

The prompt library is now more robust, performant, and self-contained.

• Offline Seeding — Bundles and initializes trending prompt libraries (such as Nano Banana, Seedance 2.0, Grok Imagine) directly from local assets on first run.

• Media Download Support — Easily download preview videos and images from prompt galleries straight to your local drive.

• Persistent Storage & Search — Fast local search, sorting, and persistent storage support for prompt library entries.

Agent Workflow & Plan Panel Enhancements

• Rerun from Task — Re-run agent plan execution starting from a specific task when a plan is paused or completed.

• Session Model Picker Override — Honorable per-session model overrides via the Skills toolbar picker during plan execution.

• Retry Failed Commands — Offers a quick "Retry command" option in the Agent Turn card to rerun failed shell commands.

• Skill Detail Panel — Hover over any skill to open a detailed documentation dialog, rendering its SKILL.md and allowing you to click on example prompts to prefill chat.

Model Additions & Chat Tweaks

• Gemini 3.5 Flash — Added support under Google Vertex AI.

• Message Editing & Regeneration — Hover controls in chat bubbles for editing user messages and regenerating assistant responses.

• Auto-Sync and Dependency Updates — Core backend improvements, upgraded dependencies, and robust async sync operations.

---

What's New in v1.3.2

Highlights

Skills Screen

The new Skills screen is the easiest way to run agent skills in AI Prompt Assistant. Open the screen, choose a skill, create a session, and describe the task. The app handles the skill instructions, workspace, approvals, attachments, and prerequisite checks for you.

• Zero Setup Required - Built-in skills are already available, and required skill assets are installed automatically. No manual skill installation is needed.

• Built-In Skills - HyperFrames, GSAP, HyperFrames CLI, HyperFrames Registry, Website to HyperFrames, and Remotion are included.

• Free to Run - Skills can use the Free provider routes, so they can run without paid API keys when a suitable free model is selected.

• Local Model Friendly - Skills can also run through local providers such as Ollama, LM Studio, Koboldcpp, or other OpenAI-compatible local endpoints.

• Add Your Own Skills - Use Add skill in the Skills panel and paste an owner/repo value or a full GitHub URL. The app finds the skill bundle and installs it into your local skills library.

• Agent Workflow Tools - Sessions get their own working folders, file attachments, requirement checks, command approval controls, safe-command auto approval, and plan-and-execute mode.

Libraries Tab

The prompt-library experience has been upgraded from a single Nano Banana gallery into a unified Libraries tab that supports multiple prompt sources in one place.

• Multi-Source Browser — switch between Nano Banana Pro, Seedance 2.0, GPT Image 1.5, SeeDream 4.5, Gemini 3, and Grok Imagine from the same screen

Free Provider

A new Free provider that connects to a public relay — no API key, no account, no setup required.

• 5 Routes — Groq, Ollama, Pollinations, Nvidia NIM, and Gemini; switch with a single dropdown in the sidebar

• Searchable Model Picker — type to filter the full model list returned by the selected route; tap a row to select (single-select radio style)

• Image Support — images are sent as standard OpenAI multipart content; Gemini and Pollinations routes generally have the best multimodal support

• Default Provider — selected automatically for new installs so users can start chatting with zero configuration

Local Enhancer Gemma 4 Support

• Added Gemma 4 E4B

• Added Gemma 4 26B A4B

• Gemma model loading, status reporting, and download checks are now integrated into the same Local Enhancer flow as the existing models

PromptFill Template Studio

• Added native PromptFill browsing, editing, and variable-filling workflows inside the desktop app

• Included imported PromptFill categories, banks, and templates with inline variable editing and AI Smart Terms support

• Added PromptFill media preview support for template images and videos

Multi-Provider Support

• Free - Zero-configuration cloud provider via the public G4F relay — no API key required. Choose from Groq, Ollama, Pollinations, Nvidia, and Gemini routes with a searchable model picker

• Ollama - Local model hosting with keep-alive configuration

• LM Studio - Local OpenAI-compatible API with model unloading

• Koboldcpp - Local inference server

• Google Gemini - Cloud API with image and video support

• OpenAI - GPT, reasoning, image, and video-capable models with per-provider API key support

• Anthropic - Claude chat models via a dedicated Anthropic provider

• Mistral - Mistral chat models and supported image tooling

• OpenRouter, Groq, Together, SwiftRouter, NVIDIA - OpenAI-compatible cloud and gateway providers with configurable base URLs

• Custom Providers - Add any OpenAI-compatible endpoint from the API settings screen, then fetch models and use it like a built-in provider

• Provider Visibility Controls - Show only the providers you want in the sidebar while keeping all provider settings available in the API settings dialog

• Veo Video Generation - Powered by Google Video FX for professional cinematic results

• Image Studio - High-quality image generation using Gemini 3 and Imagen 4 models

• Local Enhancer - Self-contained GGUF-based prompt enhancer. No third-party software required — models and the llama.cpp runtime are downloaded automatically on first useImage Studio - High-quality image generation using Gemini 3 and Imagen 4 models

Core Capabilities

• Multi-Model Execution - Run queries against multiple models simultaneously

• Image Analysis - Upload and analyze multiple images with drag-and-drop support

• Video Analysis - Full Google Files API integration with resumable uploads (Google only)

• Chat Interface - Conversation-based interaction with streaming responses

• Bulk Analysis - Batch process entire folders of images

• System Prompt Builder - Generate prompts with 11 caption types, 30 length options, and 25 extra options

• Prompt Director Pro - AI image/video prompt writing helper with model-aware dropdowns for style, camera, lighting, composition, and video movement

• Nano Banana Prompt Library - Curated prompt gallery with search, category filters, image thumbnails, and one-click copy or send-to-Image-Studio

• PromptFill Template Authoring - Build reusable prompt systems with banks, variables, template tags, and media previews

Local Enhancer (new)

• No Setup Required -prompt-enhancement backend is bundled inside the app. No Ollama, no third-party tools needed.

• Quantization Backends - GGUF and Quanto INT8 supported for flexible VRAM usage.

• Audio Understanding for Video Modes -, the enhancer analyzes attached video audio locally can incorporate dialogue, ambience, music, and sound effects into the rewritten prompt.

PromptFill

1. Switch to the PromptFill tab in the top navigation bar

2. Browse templates by search, type, and tag filters in the left sidebar

3. Select a template to open it in the visual editor

4. Fill variables by clicking any inline chip in the template body

5. Use Smart Terms in the picker dialog to generate AI suggestions for the current variable in the context of the full template

6. Add custom values when no existing bank option fits your use case

7. Edit template content with the Edit button to work directly with raw {{variable}} placeholders

8. Use AI Smart Split from the AI tools menu to turn a plain prompt into a reusable template with extracted variables

9. 

Image Studio (Generation & Editing)

• Text to Image - Create stunning visuals from descriptive prompts

• Image to Image - Use reference images to guide style, composition, and content

• Advanced Resolution - Select between 1K, 2K, and 4K output (model dependent)

• Aspect Ratio Control - Standard 1:1, Landscape 16:9, or Portrait 9:16 support

• Improve Prompt: Use the ✨ wand icon to have an LLM enhance your base prompt for better results

Veo Video Generation

• Text to Video - Generate high-quality cinematic videos from text prompts

• Image to Video - Use start and end images to guide video generation

• Extend Video - Automatically extend existing videos by extracting the last frame and generating a continuation, then seamlessly merging them with FFmpeg

• Prompt Enhancement - Built-in LLM-powered rewriter that uses attached images and video frames to create highly detailed cinematic prompts

• Advanced Controls - Configure aspect ratio (16:9, 9:16) and resolution (720p, 1080p, 4K)

SVG Generator (new)

• Text to SVG - Describe any object, icon, or scene and generate a fully self-contained SVG vector graphic

• Animated SVG - Toggle to Animated mode to produce CSS-animated SVGs with looping @keyframes effects

• Reference Image - Attach an image as a visual reference; the AI recreates it in vector format

• Export Options - Download as SVG, PNG, GIF, Animated PNG (APNG), MP4 (H.264), or MOV (lossless)

https://civitai.com/posts/27364776

Prompt Director Pro

A built-in prompt writing helper . Supports 9 AI models across image and video generation:

First Time Setup

1. Open the sidebar (hamburger menu icon)

2. Select API Provider (Ollama, LM Studio, Koboldcpp, or Google)

3. Configure provider settings:

   • For local providers: Set API base URL (default ports: Ollama=11434, LM Studio=1234, Koboldcpp=5001)

   • For Google: Enter API key

4. Fetch models using the "Fetch Models" button

5. Select one or more models from the available list

6. Choose or create a system prompt

Chat Interface

1. Upload images (optional): Click "Add Images" or drag-and-drop

2. Upload videos (Google only): Click "Add Videos"

3. Enter your message or click "Analyze Image(s)" for media-only analysis

Bulk Analysis

System Prompt Builder

Nano Banana Prompt Library

Recommended Models

Local Models (Ollama/LM Studio)

• Llama JoyCaption Alpha One (12GB VRAM) - Best for system prompt builder

• Gemma 3 27B (24GB VRAM) - High quality, complex scenes

• Gemma 3 12B (8GB VRAM) - Balanced performance

• Qwen2.5-VL-7B (8GB VRAM) - Excellent detail and instruction following

• LLaVA 1.6 (8GB VRAM) - Popular open-source option

Google Gemini & Imagen Models (Cloud)

• imagen-4.0-generate-001 - Latest Imagen 4 model for photorealistic results

• gemini-3-pro-image-preview - High-quality reasoning + image generation

• gemini-3-flash-image-preview - Fast, lightweight image generation

• gemini-2.5-flash-image - Efficient and reliable image creation

• gemini-3-flash-preview - Fast text/vision analysis

• gemini-3-pro-preview - Best quality text/vision analysis