Skip to content

armgit / kokon

Public Imported from GitHub
0

Edit photos by describing the change, on your own GPU. A free, open-source alternative to Nano Banana and ChatGPT image editing: photos stay on your computer, no daily limit. One installer for Windows, Mac and Linux sets up the best open model (Qwen-Image 2.1). MCP server for AI agents.

site: the 0.10 film, Prompt+ and LoRA (beta)

Gabriele Lozupone51499cf62c45
diff --git a/site/404.html b/site/404.html
index 86aab54..a56a436 100644
--- a/site/404.html
+++ b/site/404.html
@@ -9,7 +9,7 @@
<link rel="icon" href="/assets/favicon.ico" sizes="any">
<link rel="apple-touch-icon" href="/assets/apple-touch-icon.png">
<link href="https://fonts.googleapis.com/css2?family=Inter:opsz,[email protected],400..700&display=swap" rel="stylesheet">
-<link rel="stylesheet" href="/assets/style.css?v=6">
+<link rel="stylesheet" href="/assets/style.css?v=7">
</head>
<body>
<main class="final" style="min-height:100vh;display:grid;place-items:center;padding:40px 0">
diff --git a/site/assets/kokon-poster.jpg b/site/assets/kokon-poster.jpg
index 8246432..50563e3 100644
Binary files a/site/assets/kokon-poster.jpg and b/site/assets/kokon-poster.jpg differ
diff --git a/site/assets/kokon.mp4 b/site/assets/kokon.mp4
index 43d936d..54f565a 100644
Binary files a/site/assets/kokon.mp4 and b/site/assets/kokon.mp4 differ
diff --git a/site/assets/style.css b/site/assets/style.css
index 2a16687..c068749 100644
--- a/site/assets/style.css
+++ b/site/assets/style.css
@@ -67,7 +67,7 @@ h1 .soft { color: var(--text-3); }
.fine { margin-top: 18px; font-size: 13.5px; color: var(--text-3); }
.showreel { position: relative; margin: 72px auto 0; max-width: 1120px; }
-.showreel .frame { position: relative; z-index: 101; /* above the page grain (body::before): a translucent layer over a playing video greys it in Chrome */ border-radius: 22px; overflow: hidden; background: #000 url("assets/kokon-poster.jpg?v=2") center / cover; box-shadow: 0 0 0 1px var(--line-hi), 0 60px 160px rgba(0,0,0,0.75), 0 0 120px rgba(247,246,242,0.06); aspect-ratio: 16 / 9; }
+.showreel .frame { position: relative; z-index: 101; /* above the page grain (body::before): a translucent layer over a playing video greys it in Chrome */ border-radius: 22px; overflow: hidden; background: #000 url("kokon-poster.jpg?v=3") center / cover; box-shadow: 0 0 0 1px var(--line-hi), 0 60px 160px rgba(0,0,0,0.75), 0 0 120px rgba(247,246,242,0.06); aspect-ratio: 16 / 9; }
.showreel video { width: 100%; height: 100%; object-fit: cover; }
.reel-link { margin: 18px 0 0; text-align: center; font-size: 14px; }
.reel-link a { color: var(--text-3); text-decoration: underline; text-underline-offset: 3px; }
diff --git a/site/index.html b/site/index.html
index 978a956..7a132b9 100644
--- a/site/index.html
+++ b/site/index.html
@@ -17,7 +17,7 @@
<meta property="og:image:alt" content="Kokon logo and tagline">
<meta property="og:url" content="https://kokon.app/">
<meta property="og:type" content="website">
-<meta property="og:video" content="https://kokon.app/assets/kokon.mp4?v=2">
+<meta property="og:video" content="https://kokon.app/assets/kokon.mp4?v=3">
<meta property="og:video:type" content="video/mp4">
<meta property="og:video:width" content="1920">
<meta property="og:video:height" content="1080">
@@ -29,7 +29,7 @@
<meta name="twitter:title" content="Kokon — Run Qwen-Image 2.1 locally, no Python">
<meta name="twitter:description" content="Edit photos by describing the change, on your own GPU. Free and open source for Windows, Mac and Linux.">
<meta name="twitter:image" content="https://kokon.app/assets/og.png">
-<link rel="preload" as="image" href="assets/kokon-poster.jpg?v=2">
+<link rel="preload" as="image" href="assets/kokon-poster.jpg?v=3">
<script type="application/ld+json">
{
"@context": "https://schema.org",
@@ -43,7 +43,7 @@
"url": "https://kokon.app/",
"downloadUrl": "https://github.com/GabrieleLozupone/kokon/releases/latest",
"installUrl": "https://github.com/GabrieleLozupone/kokon/releases/latest",
- "softwareVersion": "0.9.3",
+ "softwareVersion": "0.10.0",
"releaseNotes": "https://github.com/GabrieleLozupone/kokon/releases/latest",
"softwareHelp": { "@type": "CreativeWork", "url": "https://kokon.app/prompt-guide/", "name": "Prompt guide for Qwen-Image 2.1" },
"applicationCategory": "MultimediaApplication",
@@ -55,9 +55,9 @@
"isAccessibleForFree": true,
"license": "https://www.gnu.org/licenses/agpl-3.0.html",
"offers": { "@type": "Offer", "price": "0", "priceCurrency": "USD" },
- "screenshot": "https://kokon.app/assets/kokon-poster.jpg?v=2",
+ "screenshot": "https://kokon.app/assets/kokon-poster.jpg?v=3",
"image": "https://kokon.app/assets/og.png",
- "featureList": ["Edit photos with a sentence", "Create images from text", "Paint an area to change only that area (inpainting)", "Remove, replace and add objects", "Expand images beyond their borders (outpainting)", "Before/after comparison and version history", "Private mode", "MCP server and local HTTP API for AI agents", "Runs offline on NVIDIA, AMD, Intel and Apple Silicon"],
+ "featureList": ["Edit photos with a sentence", "Create images from text", "Prompt+: a few words become a detailed prompt, written by the model already loaded", "Your own Qwen-Image 2.1 LoRAs (beta)", "Paint an area to change only that area (inpainting)", "Remove, replace and add objects", "Expand images beyond their borders (outpainting)", "Before/after comparison and version history", "Private mode", "MCP server and local HTTP API for AI agents", "Runs offline on NVIDIA, AMD, Intel and Apple Silicon"],
"author": { "@id": "https://gabrielelozupone.github.io/#me" },
"sameAs": ["https://github.com/GabrieleLozupone/kokon", "https://www.npmjs.com/package/kokon-mcp", "https://www.youtube.com/watch?v=ulMVqrhrn-w"]
},
@@ -73,8 +73,8 @@
"@type": "VideoObject",
"name": "Kokon: edit photos by describing the change, on your own GPU",
"description": "Kokon in use: creating an image from a sentence, painting an area to remove a scooter, adding a cat, and six more edits made with Qwen-Image 2.1 on a local GPU.",
- "thumbnailUrl": "https://kokon.app/assets/kokon-poster.jpg?v=2",
- "contentUrl": "https://kokon.app/assets/kokon.mp4?v=2",
+ "thumbnailUrl": "https://kokon.app/assets/kokon-poster.jpg?v=3",
+ "contentUrl": "https://kokon.app/assets/kokon.mp4?v=3",
"uploadDate": "2026-10-11",
"duration": "PT1M25S",
"embedUrl": "https://www.youtube.com/embed/ulMVqrhrn-w"
@@ -92,7 +92,7 @@
<link rel="preconnect" href="https://fonts.googleapis.com">
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
<link href="https://fonts.googleapis.com/css2?family=Inter:opsz,[email protected],400..700&family=JetBrains+Mono:wght@400;500&display=swap" rel="stylesheet">
-<link rel="stylesheet" href="assets/style.css?v=6">
+<link rel="stylesheet" href="assets/style.css?v=7">
<script type="application/ld+json">
{
"@context": "https://schema.org",
@@ -186,6 +186,22 @@
"text": "The beta is not notarized by Apple yet, so macOS blocks it the first time. Try to open it once, then go to System Settings, Privacy & Security, scroll down and click \"Open Anyway\". You only do this once; in-app updates don't trigger it again. If macOS only says the app can't be opened, with just an OK button and no Open Anyway in Settings, run xattr -dr com.apple.quarantine /Applications/Kokon.app in Terminal, then open Kokon."
}
},
+ {
+ "@type": "Question",
+ "name": "What is Prompt+?",
+ "acceptedAnswer": {
+ "@type": "Answer",
+ "text": "An optional helper next to the prompt box. Write a short idea in any language and Prompt+ writes a detailed English prompt from it (looking at your photo for edits), then makes the image; the result keeps both prompts. It runs on Qwen3-VL, the text encoder Qwen-Image 2.1 already loads, inside Kokon's own build of stable-diffusion.cpp, so there is no second model to download. It adds about 2 seconds per image on an RTX 5070 Ti."
+ }
+ },
+ {
+ "@type": "Question",
+ "name": "Can I use LoRAs?",
+ "acceptedAnswer": {
+ "@type": "Answer",
+ "text": "Yes, in beta: your own LoRA files made for Qwen-Image 2.1 (.safetensors). Load them in Settings, LoRA and turn them on for each project with the LoRA button next to the prompt, up to 3 at a time. LoRAs for older Qwen-Image versions don't work with 2.1. Tested on Windows so far; there is no LoRA catalog yet, because the popular LoRAs we tested didn't improve on what Turbo does with a good prompt."
+ }
+ },
{
"@type": "Question",
"name": "Does it work offline?",
@@ -256,7 +272,7 @@
<div class="showreel rv">
<div class="glow"></div>
<div class="frame">
- <video id="reel" src="assets/kokon.mp4?v=2" poster="assets/kokon-poster.jpg?v=2" muted loop playsinline preload="metadata" aria-label="Kokon in use: creating an image, painting an area to remove a scooter, and six more edits"></video>
+ <video id="reel" src="assets/kokon.mp4?v=3" poster="assets/kokon-poster.jpg?v=3" muted loop playsinline preload="metadata" aria-label="Kokon in use: Prompt+ writing a prompt from a few words, painting an area to remove a scooter, and five more edits"></video>
<button class="sound" id="reel-btn" type="button" aria-label="Watch the film from the start with sound">🔊 Watch with sound</button>
</div>
<p class="reel-link"><a href="https://youtu.be/ulMVqrhrn-w" rel="noopener">Watch in 4K on YouTube</a></p>
@@ -273,6 +289,7 @@
<li><svg><use href="#intel"/></svg>Intel Arc</li>
<li><svg><use href="#apple"/></svg>Apple Silicon</li>
<li><svg><use href="#win"/></svg>Windows</li>
+ <li><svg><use href="#linux"/></svg>Linux</li>
</ul>
</div>
</section>
@@ -347,7 +364,7 @@
<div class="icon"><svg viewBox="0 0 24 24"><path d="M21 15v4a2 2 0 0 1-2 2H5a2 2 0 0 1-2-2v-4M7 10l5 5 5-5M12 15V3"/></svg></div>
<h3>No Python. One installer.</h3>
<p>A small native app, about 55 MB. The models download and are checked on first run, with resume.</p>
- <div class="vis"><div class="big-num">35<small>MB installer</small></div></div>
+ <div class="vis"><div class="big-num">55<small>MB installer</small></div></div>
</article>
<article class="card b-wide rv d1">
<div class="icon"><svg viewBox="0 0 24 24"><path d="M4 17l6-6-6-6M12 19h8"/></svg></div>
@@ -355,6 +372,18 @@
<p>A built-in MCP server and a local HTTP API let Claude Code, Cursor or any MCP client create and edit images on your machine, for free and offline.</p>
<div class="vis tools" style="margin-top:auto"><code>generate_image</code><code>edit_image</code><code>inpaint_image</code><code>outpaint_image</code><code>upscale_image</code></div>
</article>
+ <article class="card b-mid rv">
+ <div class="icon"><svg viewBox="0 0 24 24"><path d="M12 3l1.9 5.1L19 10l-5.1 1.9L12 17l-1.9-5.1L5 10l5.1-1.9zM19 16l.8 2.2L22 19l-2.2.8L19 22l-.8-2.2L16 19l2.2-.8z"/></svg></div>
+ <h3>Prompt+ writes the prompt.</h3>
+ <p>Type a few words, in any language. Prompt+ turns them into a detailed prompt, looking at your photo for edits, then makes the image. It uses Qwen3-VL, which Qwen-Image 2.1 already loads: no second model to download.</p>
+ <div class="vis tools" style="margin-top:auto"><code>sunny italian street, café, red scooter</code><code>A 35mm photograph of a sunny cobblestone street in a small Italian coastal town…</code></div>
+ </article>
+ <article class="card b-mid rv d1">
+ <div class="icon"><svg viewBox="0 0 24 24"><path d="M12 2l9 5-9 5-9-5zM3 12l9 5 9-5M3 17l9 5 9-5"/></svg></div>
+ <h3>Your own LoRAs. <span class="beta">BETA</span></h3>
+ <p>Load a Qwen-Image 2.1 LoRA once in Settings, then turn it on or off for each project with one click. Kokon keeps its own prepared copy; your file is never changed. Tested on Windows so far.</p>
+ <div class="vis chips"><span class="chip">Quality</span><span class="chip">Prompt+</span><span class="chip on">LoRA</span></div>
+ </article>
</div>
</div>
</section>
@@ -490,6 +519,8 @@
</details>
<details><summary>How do I write good prompts for Qwen-Image 2.1?</summary><p>Describe the scene, not just the subject: who, doing what, where, in what light. Never write &ldquo;no X&rdquo; (it adds X), put text in quotes, and for edits name the exact thing to change. The <a href="prompt-guide/" style="text-decoration:underline">prompt guide</a> shows each rule with a vague and a clear prompt side by side.</p></details>
<details><summary>Does it run on Linux?</summary><p>Yes, in beta, on x86-64: a .deb for Ubuntu, Debian and Mint and an .rpm for Fedora, from the <a href="https://github.com/GabrieleLozupone/kokon/releases/latest" rel="noopener" style="text-decoration:underline">latest release</a>. Install it with <code>sudo apt install ./Kokon_&lt;version&gt;_amd64.deb</code> or <code>sudo dnf install ./Kokon-&lt;version&gt;-1.x86_64.rpm</code>. NVIDIA cards need the proprietary driver; AMD and Intel cards use Vulkan. A headless server with the interface in your browser, then Docker, come next.</p></details>
+ <details><summary>What is Prompt+?</summary><p>An optional helper next to the prompt box. Write a short idea in any language and Prompt+ writes a detailed English prompt from it (looking at your photo for edits), then makes the image; the result keeps both prompts. It runs on Qwen3-VL, the text encoder Qwen-Image 2.1 already loads, inside Kokon&#x27;s own build of stable-diffusion.cpp, so there is no second model to download. It adds about 2 seconds per image on an RTX 5070 Ti.</p></details>
+ <details><summary>Can I use LoRAs?</summary><p>Yes, in beta: your own LoRA files made for Qwen-Image 2.1 (.safetensors). Load them in Settings &rarr; LoRA and turn them on for each project with the LoRA button next to the prompt, up to 3 at a time. LoRAs for older Qwen-Image versions don&#x27;t work with 2.1. Tested on Windows so far; there is no LoRA catalog yet, because the popular LoRAs we tested didn&#x27;t improve on what Turbo does with a good prompt.</p></details>
<details><summary>Does it work offline?</summary><p>Yes. After setup has downloaded the models, everything runs offline.</p></details>
</div>
</div>
diff --git a/site/llms.txt b/site/llms.txt
index 98afd88..bf53114 100644
--- a/site/llms.txt
+++ b/site/llms.txt
@@ -2,12 +2,12 @@
> Kokon is a free, open-source desktop app for editing and creating photos with AI on your own computer. You describe the change in plain words ("remove the scooter", "make it a snowy night") and Qwen-Image 2.1 does it locally on your GPU. There is nothing to set up: one installer detects the hardware, downloads and verifies the right model, and picks the resolution and the VRAM/RAM split for every job. Nothing is uploaded.
-Key facts (version 0.9.3, October 2026):
+Key facts (version 0.10.0, October 2026):
- Model: Qwen-Image 2.1 Turbo (official 8-step version, 4-bit GGUF) with Qwen3-VL 8B as the text encoder, run through stable-diffusion.cpp. About 12 GB of models, downloaded once and checked with SHA-256.
- Platforms: Windows 10/11 with an NVIDIA GPU (12 GB VRAM or more, CUDA; 8 GB in beta, about 1-megapixel images). AMD and Intel Arc (12 GB+) through Vulkan, beta. Apple Silicon Macs on macOS 15 or newer (Metal), beta: 16 GB works slowly, 32 GB recommended. Linux x86-64 (.deb for Ubuntu/Debian/Mint, .rpm for Fedora; CUDA or Vulkan), beta.
- Speed: RTX 5070 Ti about 24 s per edit at 1.5K and 36 s for a new image at native 2K; RTX 5070 12 GB about 12 s per 1-megapixel image; M2 Pro about 4 minutes per image.
-- Features: create from text, edit with a sentence, paint an area and Change/Remove/Replace/Add only there, combine several images ("image 1", "image 2"), expand beyond the borders (outpainting), AI upscaling, versions with undo and before/after compare, private mode, works offline after setup.
+- Features: Prompt+ (a few words in any language become a detailed prompt, written by the Qwen3-VL text encoder already loaded, inside Kokon's own build of stable-diffusion.cpp, so no second model), your own Qwen-Image 2.1 LoRAs (beta, tested on Windows), create from text, edit with a sentence, paint an area and Change/Remove/Replace/Add only there, combine several images ("image 1", "image 2"), expand beyond the borders (outpainting), AI upscaling, versions with undo and before/after compare, private mode, works offline after setup.
- For AI agents: built-in MCP server (generate_image, edit_image, inpaint_image, outpaint_image, upscale_image, ...) and a local HTTP API. `npx -y kokon-mcp`; listed in the official MCP Registry as io.github.GabrieleLozupone/kokon.
- License: the app is AGPL-3.0. The Qwen-Image 2.1 weights are under the Qwen Research License, which does not allow commercial use.
- Not yet: a headless server / Docker image (planned); code signing (Windows SmartScreen and macOS Gatekeeper warn on first run).
diff --git a/site/mcp-image-generation/index.html b/site/mcp-image-generation/index.html
index 43f6446..2031a78 100644
--- a/site/mcp-image-generation/index.html
+++ b/site/mcp-image-generation/index.html
@@ -22,7 +22,7 @@
<link rel="preconnect" href="https://fonts.googleapis.com">
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
<link href="https://fonts.googleapis.com/css2?family=Inter:opsz,[email protected],400..700&family=JetBrains+Mono:wght@400;500&display=swap" rel="stylesheet">
-<link rel="stylesheet" href="../assets/style.css?v=6">
+<link rel="stylesheet" href="../assets/style.css?v=7">
<script type="application/ld+json">
{
"@context": "https://schema.org",
diff --git a/site/prompt-guide/index.html b/site/prompt-guide/index.html
index 535500c..c3520b7 100644
--- a/site/prompt-guide/index.html
+++ b/site/prompt-guide/index.html
@@ -26,7 +26,7 @@
<link rel="preconnect" href="https://fonts.googleapis.com">
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
<link href="https://fonts.googleapis.com/css2?family=Inter:opsz,[email protected],400..700&family=JetBrains+Mono:wght@400;500&display=swap" rel="stylesheet">
-<link rel="stylesheet" href="../assets/style.css?v=6">
+<link rel="stylesheet" href="../assets/style.css?v=7">
<script type="application/ld+json">
{
"@context": "https://schema.org",
diff --git a/site/qwen-image-2-1-local/index.html b/site/qwen-image-2-1-local/index.html
index af483eb..bfb2a98 100644
--- a/site/qwen-image-2-1-local/index.html
+++ b/site/qwen-image-2-1-local/index.html
@@ -22,7 +22,7 @@
<link rel="preconnect" href="https://fonts.googleapis.com">
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
<link href="https://fonts.googleapis.com/css2?family=Inter:opsz,[email protected],400..700&family=JetBrains+Mono:wght@400;500&display=swap" rel="stylesheet">
-<link rel="stylesheet" href="../assets/style.css?v=6">
+<link rel="stylesheet" href="../assets/style.css?v=7">
<script type="application/ld+json">
{
"@context": "https://schema.org",
diff --git a/site/sitemap.xml b/site/sitemap.xml
index 662f8f3..c04d36f 100644
--- a/site/sitemap.xml
+++ b/site/sitemap.xml
@@ -2,12 +2,12 @@
<urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9" xmlns:video="http://www.google.com/schemas/sitemap-video/1.1" xmlns:image="http://www.google.com/schemas/sitemap-image/1.1">
<url>
<loc>https://kokon.app/</loc>
- <lastmod>2026-10-11</lastmod>
+ <lastmod>2026-10-20</lastmod>
<video:video>
- <video:thumbnail_loc>https://kokon.app/assets/kokon-poster.jpg?v=2</video:thumbnail_loc>
+ <video:thumbnail_loc>https://kokon.app/assets/kokon-poster.jpg?v=3</video:thumbnail_loc>
<video:title>Kokon: edit photos by describing the change, on your own GPU</video:title>
<video:description>Kokon in use: creating an image from a sentence, painting an area to remove a scooter, adding a cat, and six more edits made with Qwen-Image 2.1 on a local GPU.</video:description>
- <video:content_loc>https://kokon.app/assets/kokon.mp4?v=2</video:content_loc>
+ <video:content_loc>https://kokon.app/assets/kokon.mp4?v=3</video:content_loc>
<video:duration>85</video:duration>
</video:video>
</url>