Kiri.art

Kiri.art is a free, open-source Progressive Web App (PWA) for AI image generation, built with React, Next.js, and Material UI. Developed by Gadi Cohen, it provides a zero-setup web interface for Stable Diffusion and SDXL models that runs in any browser without requiring a local GPU. The platform targets artists, designers, and hobbyists who want to generate, edit, and upscale AI images through an intuitive mobile-first interface, as well as developers who need a customizable open-source frontend for their own diffusion backends.

Key Features

• Text-to-image generation — Enter text prompts to generate images using Stable Diffusion or SDXL 1.0, with full control over inference steps, guidance scale, dimensions, and random seeds. SDXL support enables native 1024x1024 resolution outputs with improved composition and detail.
• Image-to-image and inpainting — Use an existing image as a structural reference for new generations, or selectively edit regions of an image with AI-generated content using the inpainting tool.
• Instruct Pix2Pix — Edit images using natural language instructions such as "make it look like a watercolor painting" without manually masking regions.
• High-quality upscaling — Enhance generated images up to 4K resolution using Real-ESRGAN_x4plus, with optional GFPGAN face restoration for portraits.
• Compel prompt syntax — Advanced prompt weighting via the Compel library lets users emphasize or de-emphasize specific terms for finer control over generation results.
• Multiple models and schedulers — Compatible with Stable Diffusion, SDXL, Waifu Diffusion, and other community checkpoint models. Supports DPMSolverMultistep, LMSDiscrete, DDIM, PNDM, EulerAncestral, and EulerDiscrete schedulers to balance speed and quality.
• NSFW filtering — A configurable safety checker can be applied per request for environments that require content moderation.
• Progressive Web App — Installable on desktop and mobile devices with a responsive Material UI design and native-app-like behavior.
• Internationalization — Supports English, Hebrew, Japanese, and Persian through Next.js i18n routing and Lingui translations.

Use Cases

• Artists and designers generating AI visuals for creative projects, concept art, and rapid prototyping without managing local hardware.
• Hobbyists experimenting with Stable Diffusion or SDXL models through a browser-based interface, using daily free credits or affordable cloud GPU services.
• Web developers who need a customizable, MIT-licensed open-source frontend for their own diffusion backend or Docker container.
• Small studios and teams sharing a hosted image generation interface with user accounts and credit-based usage tracking.

Pricing

All registered users receive 20 free image generation credits daily. Additional credits can be purchased in packs for higher usage. Users with their own NVIDIA GPU hardware can self-host the full stack for free under the MIT license with no usage limits.

Deployment and Architecture

The frontend is a Next.js application deployable to Vercel, Netlify, or any Node.js host. The backend is the companion docker-diffusers-api container, which wraps HuggingFace Diffusers and exposes a REST API for text-to-image, image-to-image, upscaling, and inpainting pipelines. Users without local GPU hardware can connect the frontend to serverless GPU platforms such as RunPod, with cold-start times reduced by pre-bundling model weights at container build time. The full stack is MIT-licensed and actively maintained on GitHub.