SynthLab — Multi-Model Image Generator

A multi-model AI image generation system with CLIP-based quality verification. Runs one prompt through multiple Stable Diffusion models sequentially, with automatic fallback on errors and quality scoring against a reference image.

Built on ComfyUI (backend) + Flask (web UI).

Features

Multi-model pipeline — one prompt through all enabled models (SDXL, SD 3 Medium, PixArt Sigma, AuraFlow, Kolors, HunyuanDiT, SD 1.5, and more)
CLIP verification — automatic quality scoring against reference image and text prompt
Model fallback — skips to next model on errors, retries on low quality
Web UI — configure prompts, models, output path, and monitor progress in-browser
Custom output directory — save generated images anywhere

Requirements

Windows 10/11
NVIDIA GPU with CUDA support (tested on RTX 4050 Laptop, 6GB VRAM)
Python 3.10+

Setup

cd "C:\path\to\SynthLab"
.\setup.ps1

This clones ComfyUI, creates the folder structure, and installs dependencies.

Add Models

Place model files in:

Type	Folder
Checkpoints (.safetensors)	`comfyui\models\checkpoints\`
UNET / GGUF	`comfyui\models\unet\`
Text encoders (CLIP, T5)	`comfyui\models\text_encoders\`
VAE	`comfyui\models\vae\`
LoRAs	`comfyui\models\loras\`

Usage

Double-click start.bat → opens the web UI at http://127.0.0.1:5000

Enter a positive/negative prompt
Upload a reference image (optional, for CLIP scoring)
Select models and set CLIP threshold
Click Start Generation

Folder Structure

start.bat              ← Launch ComfyUI + Web UI
setup.ps1              ← One-time setup
models.json            ← Model registry (names, workflows, VRAM)
scripts/
  app.py               ← Flask web UI + generation worker
  run_all_models.py    ← CLI batch runner
  templates/index.html ← Web UI frontend
  static/              ← Favicon, assets
workflows/             ← ComfyUI API-format workflow JSONs
prompts/               ← Prompt lists
output/raw/            ← Generated images
comfyui/               ← ComfyUI engine (gitignored)

License

MIT

Downloads last month: -

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support