⚡ Quick start
Add commands to any image request:
a lone lighthouse on a cliff /storm /dramaticlighting /85mm /8k /16:9 /9:16 /clean /mj
You get one finished prompt per ratio, in Midjourney's format:
── 16:9 ──
a lone lighthouse on a cliff, 85mm portrait framing, flattering compression, creamy background,
violent storm, dark roiling clouds, lightning, wind, dramatic lighting, strong contrast between
light and shadow, 8K resolution, extreme fine detail --ar 16:9 --no text, watermark, logo,
signature, caption, username
── 9:16 ──
(same prompt) --ar 9:16 --no text, watermark, logo, signature, caption, username
Swap /mj for /kling, /flux, /dalle, /gemini, /sd or /flow and the same request comes out in that tool's style.
No slashes needed either: "Create me a cinematic album double exposure, size 16:9 and 9:16" works.
🖼️ Before and after
Each pair starts from the same plain idea. The right-hand image adds commands.
Cafe portrait: a woman at a cafe vs a woman at a cafe /portraitpro /goldenhour /smile /sitting
Product shot: a ceramic mug vs a ceramic mug /productshot
Landscape: a forest path vs a forest path /foggyforest /goldenhour /24mm
Tokyo at night: a street in Tokyo vs a street in Tokyo /nightcity /rainynight /neonlights /cinematic
Floating castle (fantasy): a castle on a floating island vs a castle on a floating island /epicfantasy /goldenhour /volumetriclight /mattepainting
Sports car (automotive): a red sports car on a coastal road vs a red sports car on a coastal road /automotivephotography /sunset /lowangle /motionblur
Modern house (architecture): a modern house on a hillside vs a modern house on a hillside /architecturalphotography /goldenhour /wideangle /clean
Perfume (product): a glass perfume bottle on a marble table vs a glass perfume bottle on a marble table /productphotography /softlighting /macro /clean
Ramen (food): a bowl of ramen vs a bowl of ramen /foodphotography /warmtones /closeup /shallowdepth
Fisherman (portrait): an elderly fisherman on a boat vs an elderly fisherman on a boat /environmentalportrait /goldenhour /85mm /highdetail
Living room (interior): a living room vs a living room /scandinavianinterior /naturalwindowlight /wideangle /clean
Left: the plain idea. Right: the same idea with the commands shown above. Made by the project author with AI image tools (tool and seed not recorded), one result per prompt, shown as made.
The commands side is the expander's output for each request (the lighthouse one is the /mj form without flags; the others use /gemini sentences):
a lighthouse on a cliff, 24mm wide lens, environmental context with natural perspective, violent storm, dark roiling clouds, lightning, wind, dramatic lighting, strong contrast between light and shadow
a woman at a cafe. Warm natural smile. Seated pose. 85mm portrait lens, flattering compression, creamy background. Aperture f/1.4, razor-thin focus plane, heavy background blur. Beautiful bokeh, creamy out-of-focus highlights. Portrait photograph, subject-centered, flattering framing. Soft diffused lighting, gentle shadows, flattering glow. Golden hour, warm low-angle sunlight, long soft shadows. Avoid extra fingers, missing fingers, deformed hands, extra limbs, malformed anatomy, distorted face, cross-eyed.
a ceramic mug. 85mm portrait lens, flattering compression, creamy background. Shallow depth of field, sharp subject, blurred background. Product photography, clean hero shot, clean background, commercial lighting. Professional studio setup, controlled lighting, clean backdrop. Soft diffused lighting, gentle shadows, flattering glow. Tack-sharp focus on the subject. Avoid text, watermark, logo, signature, caption, username.
a forest path, 24mm wide lens, environmental context with natural perspective, foggy forest, misty trees, quiet depth layers, golden hour, warm low-angle sunlight, long soft shadows
- The prompts above are the wording the example images were made with. Since then the lens and camera phrases were reworded to describe the look (for example "85mm portrait framing") instead of naming equipment, because some models drew the lens itself into the picture.
- Images were generated by the project author with AI image tools. The exact tool, version and seed were not recorded, and some images carry the tool's own corner mark.
- One result per prompt, shown as made. This is a set of examples, not a benchmark: results vary by tool and seed.
- Aspect-ratio commands (
/16:9, /9:16) were not applied by the tools used, so the outputs are square or 3:4.
- The two cafe pairs are separate runs of the same request.
Prompts for the "More examples" (Gemini/ChatGPT sentence format from the expander):
a street in Tokyo. City at night, lit windows, streetlights, urban glow. Rainy night, wet reflective surfaces, glowing lights in rain. Vivid neon lights, glowing signs, colorful light spill. Cinematic look, movie-still composition, filmic lighting and color.
a castle on a floating island. Wide-angle perspective, expansive field of view. Fantasy world, magical, epic imaginative setting. God rays, visible beams of light through atmosphere. Golden hour, warm low-angle sunlight, long soft shadows. Volumetric lighting, light beams in hazy air. Matte painting, grand cinematic environment. Highly detailed, intricate fine details.
Aspect ratio: 16:9
a red sports car on a coastal road. Low-angle shot, viewed from below looking up, imposing perspective. Motion blur, sense of speed and movement. Automotive photography, sleek car, glossy reflections, dynamic angle. Sunset, warm orange-pink sky, low sun.
a modern house on a hillside. Wide-angle perspective, expansive field of view. Architectural photography, clean lines, precise perspective. Golden hour, warm low-angle sunlight, long soft shadows. Avoid text, watermark, logo, signature, caption, username.
a glass perfume bottle on a marble table. Product photography, clean hero shot, clean background, commercial lighting. Macro photography, extreme magnification. Soft diffused lighting, gentle shadows, flattering glow. Avoid text, watermark, logo, signature, caption, username.
a bowl of ramen. Close-up shot, subject fills the frame. Shallow depth of field, sharp subject, blurred background. Food photography, appetizing, styled plating, fresh textures. Warm tones, golden amber hues.
an elderly fisherman on a boat. 85mm portrait framing, flattering compression, creamy background. Environmental portrait, subject shown in their meaningful surroundings. Golden hour, warm low-angle sunlight, long soft shadows. Highly detailed, intricate fine details.
a living room. Wide-angle perspective, expansive field of view. Scandinavian interior, light wood, white, cozy minimalism. Natural window light, soft directional daylight. Avoid text, watermark, logo, signature, caption, username.
✨ What you get
| |
|---|
| 🎯 Consistent wording | /goldenhour always means the same phrase, not whatever you remember to type. |
| 🧩 Right format per tool | Midjourney flags, Stable Diffusion negative prompt, Kling negative field, Flux and Gemini sentences. |
| 🛡️ Guard rails | Max 8 style modifiers, one winner for conflicting commands (/4k vs /8k), unknown words kept as text. |
| 📐 Many ratios, many prompts | Ask for 16:9 and 9:16 together and get one prompt for each. |
| 🗣️ Plain English | Style words and ratios are recognised without slashes; filler like "create me a" is dropped. |
| 💾 Your own recipes | Save a bundle as /mylook and reuse it. Plus 14 built-in presets such as /productshot. |
🧰 Commands
403 in total. Browse them all in the cheat sheet.
| Group | Count | Examples |
|---|
| Style | 288 | /cinematic /goldenhour /droneview /chiaroscuro /oilpainting /cyberpunk |
| Camera and lens | 13 | /85mm /f1.4 /tiltshift /macrolens |
| Video motion | 10 | /dolly /orbit /handheld /slowmotion |
| Couple and wedding | 16 | /couple /prewedding /wedding /holdinghands /sunsetcouple |
| Subject controls | 29 | /age30s /smile /walking /suit /longhair |
| Reference images | 6 | /sameface /samestyle /samepose |
| Aspect ratios | 10 | /16:9 /9:16 /square /4:5 (also /16x9) |
| Avoid-lists | 9 | /clean /nodeform /nopeople /notext /nogear |
| Presets | 14 | /productshot /linkedinheadshot /moviepost /youtubethumb /albumcover |
| Target tools | 8 | /mj /sd /flux /dalle /gemini /flow /kling /generic |
- Order: subject, then angle, camera, light, style, quality last.
- At most 8 style modifiers (reference commands do not count). Extras are dropped, quality first.
- One winner per slot, the last one given: resolution, realism, focal length, aperture, age, expression, pose, gaze, hair length.
- Several ratios in one request produce one prompt per ratio.
- Unknown
/words are kept as plain text and reported.
- Video-only commands (
/dolly, /orbit...) are ignored for still images, and the reply says so.
- Reference commands need an image you own or have permission to use.
| Preset | Bundle | Default ratio |
|---|
/productshot | product photography, studio, soft light, shallow depth, 85mm, sharp, clean | 1:1 |
/linkedinheadshot | headshot, studio, soft light, 85mm, f1.4, natural colors, clean | 4:5 |
/albumcover | cinematic, dramatic light, centered, film grain, color graded | 1:1 |
/youtubethumb | close-up, dramatic light, high contrast, vibrant, sharp | 16:9 |
/moviepost | cinematic, dramatic light, anamorphic, teal and orange, film grain | 2:3 |
/streetportrait | street photography, candid, 35mm, golden hour, film grain | 3:2 |
/foodhero | food photography, macro close-focus, window light, soft light, shallow depth | 4:5 |
/realestate | real-estate photography, 14mm, window light, high detail | 16:9 |
/fashioneditorial | fashion, editorial, dramatic light, 85mm, color graded | 4:5 |
/cyberpunkcity | cyberpunk, night city, rainy night, neon, anamorphic | 21:9 |
/epicfantasy | fantasy, matte painting, god rays, wide angle, high detail | 16:9 |
/portraitpro | portrait, 85mm, f1.4, soft light, bokeh, natural colors | 3:4 |
/couplepro | couple, romantic, candid, 85mm, soft light, golden hour, natural colors | 4:5 |
/weddingclassic | wedding, 50mm, soft light, natural colors, color graded, clean | 3:2 |
Your own ratio always overrides a preset's default.
📦 Install per platform
Everything is pre-built in dist/. Pick your tool:
| Tool | What to use | How |
|---|
| Claude Code (plugin) | dist/marketplace | claude plugin marketplace add geevarghesekthomas84-sys/image-style-skill then claude plugin install image-style@image-style-market |
| Claude Code (skill only) | dist/image-style/ | Copy the folder to ~/.claude/skills/image-style/ |
| Claude.ai | dist/platforms/claude-ai/image-style.zip | Settings → Capabilities → Skills → Upload skill |
| ChatGPT | dist/platforms/chatgpt/ | Create a Custom GPT. Paste instructions.txt, upload image-style-knowledge.md, enable Image Generation. No GPT access? Paste custom-instructions-short.txt into Custom instructions. |
| Gemini | dist/platforms/gemini/ | Create a Gem. Paste instructions.txt, add image-style-knowledge.md as knowledge. |
| ComfyUI | dist/platforms/comfyui/ComfyUI-ImageStyle.zip | Unzip into ComfyUI/custom_nodes, restart, then add node image-style, Image Style Prompt. Details: comfyui/ComfyUI-ImageStyle. |
| Google Flow, Kling, anything else | dist/platforms/flow-kling-anywhere/expander.html | Save the file and open it in a browser (works offline). The file has the Generate image and Copy-and-open buttons. |
Plugin commands are namespaced (/image-style:goldenhour a quiet harbor). The skill also understands bare tokens in normal messages.
Claude.ai has no image generator, so it returns the final prompt for you to paste.
🌐 Online prompt generators
Free, no sign-up: Midjourney · Flux · Stable Diffusion · ChatGPT and DALL-E · Gemini · Google Flow · Kling. New to prompts? Read how to write AI image prompts.
💻 The expander page
Google Flow and Kling have no skill or plugin system, so there is a single offline web page that does the same job. A copy runs online at glogee.online.
It includes a searchable command browser (click to add), one-click copy, and a panel showing exactly what each command added:
Share links: when the page is served from a web address (not opened as a local file), a Copy link button makes a URL like ...#r=a lighthouse /storm /24mm /16:9&t=mj that reopens the same request and target for anyone. Your own saved recipes are expanded into plain commands first, so the link works for others.
Also in the page: Copy and open buttons (Flow, Kling, ChatGPT, Gemini, Midjourney) and your own saved recipes.
Type your usual commands, give them a name (for example mylook) and click Save these /commands as a recipe.
Then /mylook works like any built-in command and shows under the Mine tab, with a delete button.
Recipes are stored in your browser only. Open the page through http://127.0.0.1:8787/app (run python imagegen.py) so the browser keeps them reliably; pages opened as plain files may not keep saved data.
The Generate image button tries your own API first, then a free third-party service (often rate-limited).
Your key never goes into the web page: a small local helper holds it.
$env:OPENAI_API_KEY = "sk-..." # and/or $env:GEMINI_API_KEY = "..."
python imagegen.py
Then open http://127.0.0.1:8787/app and click Generate image. Images are billed to your provider account.
Provider model names change often. Override with OPENAI_IMAGE_MODEL (default gpt-image-2) or GEMINI_IMAGE_MODEL (default gemini-3.1-flash-image) if one is rejected. Google Imagen has been shut down.
🧩 ComfyUI node
Use the same commands inside ComfyUI. The Image Style Prompt node takes a request like a lone lighthouse /storm /24mm /16:9 /clean and outputs a positive prompt, a negative prompt, a width and a height, plus a report of what each command did. Wire positive and negative into your text encoders and width and height into Empty Latent Image. A second node outputs one item per ratio, so a workflow runs once for each (for example /16:9 /9:16).
It also adds /command autocomplete and a live preview to the request box, plus an Upload & Convert node: upload an image and save it as SVG (traced to flat-colour vectors, best for logos and flat art), PNG, JPG, WebP, BMP, TIFF or GIF. It only writes text and numbers (no model is loaded) and needs no extra Python packages. Install and wiring notes are in comfyui/ComfyUI-ImageStyle. A ready-made Z-Image-Turbo workflow with the node already wired in is here: image-style-z-image-turbo.json (drag it onto the ComfyUI canvas). Also: image-style-upload-to-svg.json and image-style-z-image-turbo-svg.json (same workflow plus SVG export). Tested in ComfyUI 0.39.0, not yet on older versions.
📋 Cheat sheet
dist/cheatsheet.html lists every command with its wording. Searchable, and prints cleanly (Ctrl+P).
🔧 Edit and rebuild
Data lives in src/:
| File | Format |
|---|
modifiers.txt | name|category|phrase |
aspect.txt | name|ratio|pixels|note (note may list aliases: ...) |
negatives.txt | name|terms to avoid |
models.txt | name|label|formatting rules |
presets.txt | name|label|tokens|default ratio |
python build.py # regenerate dist/
python test.py # names, presets, outputs, page script
python run_claude_tests.py # end-to-end through `claude -p` (needs a signed-in CLI)
Names are lowercase a-z, 0-9 and . only, and must be unique across all files.
Bump VERSION in build.py after changes so Claude Code sees the plugin update.
GitHub Actions runs build.py and test.py on every push.
✅ Status: what is tested
| Area | State |
|---|
| Data, build, plugin validation, page script | Tested automatically on every push |
| Claude Code skill and plugin | 9 end-to-end cases pass through real claude -p runs (run_claude_tests.py) |
| Expander page (parsing, ratios, presets, recipes, cap, copy) | Tested by hand in a browser |
imagegen.py | Checked against mock OpenAI and Gemini servers only, not yet with a real key |
| Claude.ai, ChatGPT and Gemini uploads | Not yet tested in the real apps |
| Output formats for Flow, Kling, Midjourney, Flux | Written from general knowledge, not benchmarked. Please report outputs that look wrong. |
No command names a living artist or studio. test.py enforces a small blocklist.
☕ Support
SlashFrame is free and open source. If it saves you time, you can buy me a coffee or scan the code:
📄 License
MIT
