<h1 id="moss-image-prompt-enhancer">MOSS-Image Prompt Enhancer</h1><p>An Ollama model that turns short image ideas into detailed prompts for the MOSS-Image (<a target="_blank" rel="ugc" href="https://github.com/OpenMOSS/MOSS-Image">https://github.com/OpenMOSS/MOSS-Image</a>) text-to-image model.</p><p>It is based on the official moss-image-prompt-enhancer skill (<a target="_blank" rel="ugc" href="https://github.com/OpenMOSS/MOSS-Image/tree/main/skills/moss-image-prompt-enhancer">https://github.com/OpenMOSS/MOSS-Image/tree/main/skills/moss-image-prompt-enhancer</a>), packaged as a local model so it can run offline without an agent.</p><h2 id="what-it-does">What it does</h2><p>Give it a rough idea, get back only the enhanced prompt (no preamble, no explanations).</p><p><strong>Caption</strong> (default), Photos, portraits, landscapes, illustrations, paintings - One cohesive English prose description</p><p><strong>Layout</strong>, Posters, infographics, multi-panel images, exact text placement - A JSON draft with pixel bounding boxes</p><h3 id="behavior:">Behavior:</h3><ul><li><p><strong>Keeps your intent:</strong> subject, style, counts, exclusions, and exact visible text are preserved.</p></li><li><p><strong>Avoids filler:</strong> no quality-tag lists, no negative prompts, no invented slogans or watermarks. Minimalist requests stay minimalist.</p></li><li><p><strong>People always get an age and an ethnicity.</strong> If you give them, they are used exactly. If not, they are inferred from context (a Seoul night market gives Korean, "grandmother" gives elderly). With no clues the model picks one.</p></li><li><p><strong>Honors an explicit mode</strong> ("Layout mode, 1024x1536: ...", "Caption mode: ...") and an explicit canvas size.</p></li></ul><h3 id="run-directly-from-ollama">Run directly from Ollama</h3><p>To run the model directly from <a target="_blank" rel="ugc" href="http://Ollama.com">Ollama.com</a>, use the following command:</p><p></p><pre><code>ollama pull DutchIT/moss-image-prompt-enhancer
ollama run DutchIT/moss-image-prompt-enhancer "a woman reading in a cafe"</code></pre><h3 id="install">Install</h3><pre><code>git clone https://github.com/royen99/moss-image-prompt-enhancer
cd moss-image-prompt-enhancer
ollama create moss-prompt-enhancer -f Modelfile</code></pre><p>The first run downloads the base model. Re-run the same command after editing the <code>Modelfile</code>; unchanged layers are reused.</p><h3 id="usage">Usage</h3><pre><code>ollama run moss-prompt-enhancer "a woman reading in a cafe"</code></pre><p>Or through the API:</p><pre><code>curl http://127.0.0.1:11434/api/generate -d '{
"model": "moss-prompt-enhancer",
"prompt": "a street food vendor at a night market in Seoul",
"stream": false
}'</code></pre><p>Paste the result into MOSS-Image (for example in ComfyUI) as the positive prompt.</p><h3 id="caption-mode">Caption mode</h3><p>Just describe the image:</p><p><em>a grandmother knitting by a window, watercolor style</em></p><p></p><h3 id="layout-mode">Layout mode</h3><p>Ask for it explicitly, or use a poster, infographic, or multi-panel idea. State the canvas size if you care about it:</p><p><em>Layout mode, 1024x1536: a jazz concert poster. Title "SUMMER JAZZ NIGHT", below it "Friday 8 PM\nBlue Door Cafe". A saxophone player in the middle.</em></p><p></p><p>The model returns a <strong>draft</strong> with plain pixel boxes <code>bbox_coord_space: "abs"</code>, <code>bbox_format: "xyxy"</code>, <code>[x_min, y_min, x_max, y_max]</code> from the top-left). MOSS-Image expects centered <code>yxyx</code> coordinates, so the draft must be compiled before use. The official skill ships the compiler:</p><p></p><p>Get <code>compile_prompt.py</code> from the [skill's <code>scripts</code> folder](<a target="_blank" rel="ugc" href="https://github.com/OpenMOSS/MOSS-Image/tree/main/skills/moss-image-prompt-enhancer/scripts">https://github.com/OpenMOSS/MOSS-Image/tree/main/skills/moss-image-prompt-enhancer/scripts</a>). It rejects invalid or out-of-bounds boxes. Use the width and height from the draft as the image size when generating. The compiled prompt itself does not contain the dimensions.</p><p>The model is not asked to do the coordinate arithmetic because small models get it wrong.</p><p>## Usage in ComfyUI</p><p>For usage within ComfyUI I would recommend using the simple <code>comfyui-ollama</code> node without any special settings. Simply connect a input textbox to the prompt connector (no system prompt input needed).</p><p></p><h3 id="usage-in-comfyui">Usage in ComfyUI</h3><p>For usage within ComfyUI I would recommend using the simple <code>comfyui-ollama</code> node without any special settings. Simply connect a input textbox to the prompt connector (no system prompt input needed).</p><p><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/5a08c684-aca6-43ba-a77b-375d64dd0959/original=true/5a08c684-aca6-43ba-a77b-375d64dd0959.jpeg" /></p>

Files with SHA256: 7936dedd074578ba3ab76a24d518713c98b6008d02899dc662cd4f81280fe16e