Electron + Vue 3 · Open Source

Myriad Frame

A free, open-source desktop front-end for stable-diffusion.cpp. Manage chat sessions, download and select models, and drive image/video generation through a bundled sd-cli runtime — all in one place.

Download Release View on GitHub
What it does

Everything you need to generate images locally

A clean, JSON-driven desktop app that keeps all behavior easy to configure.

💬

Session & chat management

Organize generations into chat sessions with per-session sampling, negative prompt, and image-persistence settings that save automatically.

🧩

Model discovery & downloads

Browse a searchable catalog of quantized diffusion, VAE, and encoder models from curated Hugging Face and CivitAI sources — download and manage them without leaving the app.

⚡

Fast sd-cli runtime

Runs on stable-diffusion.cpp with CUDA/Vulkan or CPU support. Recommended runtime is auto-selected per platform and GPU.

🖼️

References & image input

Attach reference images for style adaptation or use a previous output as an init image. Fullscreen lightbox previews every generated image.

🎨

Inpainting & outpainting

Paint a mask over an image to regenerate regions, or expand the frame outward — all with an in-canvas editor, no external image tools required.

🔌

LoRA support

Browse, download, and apply LoRA adapters per generation with a management UI and per-session apply mode.

🎛️

JSON-driven config

Model lists, sampling defaults, resolutions, and runtime URLs are all driven by JSON files in configurations/ — edit behavior without touching code.

🌓

Themed UI

Dark-first interface with a theme picker (Default, Halloween, Christmas) and a polished, resizable layout.

🔐

Per-source auth

Store tokens for private Hugging Face and CivitAI repos via a built-in Tokens manager — never hand-edited, never logged.

Getting started

How generation works

From prompt to pixel in four steps.

STEP 01

Pick a model

Choose a quantized diffusion model (and optional VAE/encoder) from the searchable catalog.

STEP 02

Configure

Set a resolution, sampling defaults, and any reference images or prompts.

STEP 03

Generate

The main process spawns sd-cli, streams step progress, and writes results to data/output/.

STEP 04

Preview

Click any output to open a fullscreen lightbox. Regenerate, edit, or save.

Video demo

Inpainting workflow

Paint a mask over an image and regenerate the masked region — no external editors.

Screenshots

Take a look

A focused, dark-first workspace built for model-heavy workflows.

Model picker dialog
Model picker
Model management dialog
Model management
Sampling settings
Sampling settings
Improved sampling settings
Improved sampling settings
Negative prompt override
Negative prompt override
Appearance settings
Appearance settings
Unified resolution selector
Resolution selector
Qwen Image 2.1 generation
Qwen Image 2.1 — generation
Qwen Image 2.1 editing
Qwen Image 2.1 — editing
Krea 2 Turbo generation
Krea 2 — generation
Runtime & model providers

Built on open, powerful tools

Runtime: stable-diffusion.cpp
Hugging Face 20
Unsloth Abenzerps Black Forest Labs Qwen City 96 huaweilin Comfy-Org Rebel AI Stability AI CircleStone Labs RunDiffusion Krea Unmods Diffusers leejet comfyanonymous QuantStack OpenMOSS-Team Vantage with AI Lightricks
CivitAI 17
Euge_ Lykon JRD_2212 Kranos94 captain_daddy angelomaiota Wackiii annguyentruongandrew142 Darkessence arrssenne valpat1988108 theevildarkmage126 cyberthron tigerart Maxed_Out WereCatf CagliostroLab
Build from source

Run it yourself

Myriad Frame is free and open-source. Clone the repo and build locally.

# Install dependencies (downloads the Electron binary)
npm install

# Run in development
npm run dev

# Build a production bundle to out/
npm run build

Requires Node.js 20+. A GPU (NVIDIA CUDA or Vulkan) is recommended; CPU-only runs are supported but slower.