ReImagineX ✦

ReImagineX is a fluid, high-performance web gallery and image-to-image variations interface. Heavily inspired by the clean aesthetics and user interactions of Grok's (xAI) "Imagine" tool, it provides a seamless workflow to generate, iterate, upscale, and organize image lineages locally using a ComfyUI backend and DeepSeek (with OpenRouter fallback) LLM prompt refinement.

---

✦ Previews

Main Gallery	Prompt & Details	Variation Editor

---

✦ Key Features

• Linear & Lineage-based Navigation: Organize generations by lineage. Click on a root image to explore all its child variations and iterations.
• Natural Touch Gestures (Mobile optimized):
  • Swipe Up anywhere on the image view to pull up the Prompt & Quick Actions drawer.
  • Swipe Down to close the image viewer and return to the gallery grid (or slide the drawer back down).
  • Swipe Left/Right on the image viewport to slide transition between completed images.
• Premium Fluid Animations:
  • Staggered Card-Dealing Intro: Cards fan out dynamically with 3D rotation, scaling, and slide-in cascading when loading the gallery.
  • Zoom-Fade Opening: Opening/closing full images scales the viewport smoothly.
  • Swipe Transitions: Swiping left/right slides the old image out and the new image in dynamically.
• Direct Drawer Controls:
  • Upscale (2MP): Runs the prompt-based Qwen 2.1 upscale pass on the parent's base image, targeting 2 megapixels without external ESRGAN weights.
  • Reprocess: Resubmits the exact same configuration and prompt to generate a new variation.
• Bulk Select & Delete: Delete multiple images at once, cascading recursively to all their lineage descendants.
• Auto-Cancellation on interrupt: Deleting or canceling a pending generation sends a signal to ComfyUI to immediately cancel the job and clear the execution queue.
• Dynamic Configuration: Adjust API keys, ComfyUI URLs, UNet/diffusion models, VAE, CLIP, and prompt expansion prompts on the fly via the built-in Settings Panel.

---

✦ System Architecture & Setup

ReImagineX runs as a fast Python web server (FastAPI) acting as a orchestrator/proxy to your local ComfyUI instance.

1. Prerequisites

Before running the server, verify you have:

• Python 3.12+
• uv (high-performance Python package installer)
• A running ComfyUI instance.

2. Local Installation

Clone the repository and run dependencies setup using uv:

# Install dependencies and setup venv
uv sync

3. Environment Variables Configuration

Create a .env file in the root directory (you can copy .env.example):

# ComfyUI backend URL
COMFY_URL=http://127.0.0.1:8001

# Data directory for local images and state JSON
DATA_DIR=gallery_data

# Network binding (0.0.0.0 exposes the gallery on your local network / Wi-Fi)
REIMAGINEX_HOST=0.0.0.0
REIMAGINEX_PORT=8888

# LLM API keys (DeepSeek is the default provider, OpenRouter is the fallback)
DEEPSEEK_API_KEY=your_api_key_here
OPENROUTER_API_KEY=your_api_key_here

Note

Settings saved from the Settings Panel are written to DATA_DIR/config.json, which is applied after this file and therefore takes precedence once it exists. Use the panel to change models/prompts after the first run; use .env for the API keys and network binding.

4. Running the Web App

You can quickly boot the server by running the startup script:

./start.sh

Alternatively, launch the local development web server manually:

uv run python -m src.server

The application will be accessible locally at http://localhost:8888. Because the server binds to 0.0.0.0 by default, it is also reachable from other devices on your network at http://<your-lan-ip>:8888. Set REIMAGINEX_HOST=127.0.0.1 to restrict it to this machine, or pass --host/--bind and --port explicitly.

---

✦ ComfyUI Backend Setup

Tip

A pre-configured API workflow template is available in the project root as workflow_api.json. You can load this file into your ComfyUI interface. If you are missing any custom nodes, use the ComfyUI Manager and click Install Missing Custom Nodes to install all required dependencies automatically.

For the generation workflows to validate successfully, your ComfyUI environment must meet the following configuration requirements:

Required Models

Ensure the following models are downloaded and placed in the respective folders under your ComfyUI directory:

Model File	Target Subfolder	Type	Description
qwen_image_2.1_int8_convrot.safetensors	models/diffusion_models/ or models/unet/	Diffusion (UNet)	Qwen Image 2.1 diffusion model
qwen3vl_8b_int8_convrot.safetensors	models/clip/	CLIP	Qwen 3 VL 8B text/vision encoder
qwen_image_2.1_vae_bf16.safetensors	models/vae/	VAE	Qwen Image 2.1 VAE
PornMaster_QI2.1_Low_Resolution_Slider_V1.safetensors	models/loras/	LoRA	Qwen 2.1 Resolution Slider (used at -3.0 weight for upscale)

Generation & Upscaling Architecture

1. Qwen Image 2.1 Edit Workflow: Uses native ComfyUI UNET/CLIP/VAE loaders alongside TextEncodeQwenImage21, QwenImage21Cache and sampler nodes (workflow_api.json).
2. Qwen 2.1 Prompt-Based 2MP Upscaler: Uses workflow_upscale_api.json with high-detail DSLR restoration prompt and resolution slider LoRA (-3.0 weight) to reconstruct micro-textures and upscale images to 2 Megapixels without external ESRGAN weights.

---

✦ User Guide & Workflow

1. Upload a Base Image: Click the Upload button in the header to import your starting image. This becomes the "Root" of a new lineage tree.
2. Generate Variations:
   • Click the root image in the gallery to enter its lineage view.
   • Click the Create Variation button (btnOpenPrompt pen icon) in the top-right corner of the modal.
   • Enter your prompt, configure options (Generation Resolution, Bypass LLM), and click Generate Variation.
3. Refining Lineage: Swipe left or right through completed variations. Swipe up to access prompt details and perform quick upscaling or reprocessing tasks on the fly.
4. Delete and Clean: Toggle Select Multiple (btnToggleSelection checklist icon) in the header to purge unwanted lineage branches recursively.