322 composable visual commands for AI image generation and editing across modern AI coding agents.
Visual Commands is an open Agent Skill that teaches AI coding agents—including OpenAI Codex, Claude Code, Google Antigravity, and compatible Agent Skills environments—how to interpret and execute a compact, composable visual command language.
Instead of writing repetitive paragraphs of prompt engineering, control image transformations with short, composable slash commands:
/xray
/cinematic
/night
/360view
/explodedview
/remove
/productshot
Compose commands together:
/xray /cinematic /night
/360view /productshot /blackbg
/explodedview /technical /whitebg /16:9
Visual Commands translates these compact directives into deterministic, art-directed image generation and editing instructions for the image tools available in your host environment while preserving source fidelity by default.
Visual Commands adheres to the open Agent Skills specification (SKILL.md), allowing the exact same command registry and composition rules to be used across multiple agent platforms:
| Environment | Skill Discovery | Image Execution | Recommended Location |
|---|---|---|---|
| OpenAI Codex | Supported | Depends on host image tools | $skill-installer or ~/.codex/skills/ |
| Claude Code | Supported | Depends on host image tools | ~/.claude/skills/ or .claude/skills/ |
| Google Antigravity | Supported | Depends on host image tools | ~/.gemini/config/skills/ or .agents/skills/ |
| Compatible Agent Environments | Supported via SKILL.md |
Host-dependent | Standard Agent Skills directory |
ℹ️ Note on Execution: Visual Commands provides the visual language parsing, prompt normalization, and composition logic. Actual image generation and editing is performed by the image-generation and editing tools provided by your host environment.
Visual Commands cleanly separates visual intent specification from agent execution:
flowchart TD
A[User Input:\nImage + Slash Commands] --> B[Visual Commands Skill:\nSKILL.md + Parser Engine]
B --> C[Unified Visual Directive:\nResolved Conflicts & Preserved Fidelity]
C --> D[Host AI Agent:\nCodex / Claude Code / Antigravity]
D --> E[Host Image Tooling:\nGenerate / Edit Output]
E --> F[Finished Image Result]
- Visual Command Layer: The user provides compact slash commands (
/xray /cinematic /night). - Agent Skill Normalization: The skill normalizes aliases, extracts parameters, and resolves conflicts using canonical rules in
commands.json. - Host Agent Execution: The host coding agent (Codex, Claude Code, Antigravity) receives a precise, unified visual prompt.
- Tool Generation: The host environment's image tools produce or edit the visual output.
-
Install the skill in your preferred agent environment (see Installation).
-
Attach or reference an image (for edits) or describe a subject (for new generation).
-
Enter one or more commands:
/xray /cinematic /night -
Your agent interprets the commands through Visual Commands and executes the image task using the available host image tools.
Complex visual transformations typically require verbose prompts to maintain consistent lighting, framing, background isolation, and subject fidelity. Visual Commands packages proven visual intent into memorable, reusable directives:
| Instead of writing | Type |
|---|---|
| "Preserve the exact product identity and proportions, produce eight consistent angles, use commercial studio lighting, and place every view on a seamless black background..." | /360view /views 8 /productshot /blackbg |
| "Create an exploded view of this assembly showing all internal components separated along their axis on a clean white background in 16:9 format..." | /explodedview /technical /whitebg /16:9 |
| "Relight this scene to nighttime with cinematic grading while preserving camera position and subject details..." | /cinematic /night /sameangle |
Commands are additive by default across different functional categories:
/xray /cinematic /night
/xray→ Structural transformation/cinematic→ Art direction & grading/night→ Environmental lighting condition
All three instructions combine into a single, unified generation/editing operation.
When two or more commands belong to the same mutually exclusive single-value category, the last explicit command wins:
/day /night → night
/blackbg /whitebg → white background
/frontview /rearview → rear view
/9:16 /16:9 → 16:9
Additive conditions within compatible categories do not conflict (for example, /night /rain /fog or /cinematic /luxury /productshot).
Master commands define the high-level presentation structure or multi-view layout of the output image. Modifier commands style the master output rather than overriding its structural format:
/360view /productshot /blackbg
(Generates an 8-angle 360° turnaround where every view receives the commercial product lighting and seamless black background treatment.)
/360view— Multi-angle 360-degree turnaround presentation/turnaround— Front, 3/4, side, and rear multi-view sequence/orthographic— Top, front, side orthographic technical views/explodedview— Exploded mechanical/product assembly view/disassembly— Disassembled parts layout/assembly— Stepwise assembly presentation/beforeafter— Side-by-side original and transformed comparison/contactsheet— Grid array of curated viewpoints/detailviews— Multi-panel close-up detail inspection/componentmap— Organized parts map with clean visual hierarchy/materialboard— Material and finish swatch board/colorways— Multi-colorway product palette sheet/variants— Design variation comparison sheet/storyboard— Cinematic multi-frame narrative layout/sequence— Step-by-step visual sequence/timelapse— Progressive time/state transformation/progression— Staged evolution or construction sequence/blueprint— Architectural/engineering blueprint schematic/schematic— Clean technical schematic presentation/layers— Cutaway layered structure display/grid4— 2×2 quadrant presentation grid/grid9— 3×3 multi-view presentation grid/diptych— Two-panel side-by-side layout/triptych— Three-panel progressive layout/panorama— Ultra-wide panoramic scene composition/product360— Commercial product turntable sequence
Visual Commands includes 322 verified commands across 13 functional categories:
| Category | Count | Description | Representative Commands |
|---|---|---|---|
| Look & Style | 65 | Art direction, aesthetic treatments, rendering styles | /cinematic, /editorial, /luxury, /minimal, /cyberpunk, /vintage, /productshot, /technical, /photoreal, /noir |
| Camera & Angle | 37 | Camera positioning, angles, viewpoints, multi-view capture | /360view, /frontview, /isometric, /birdseye, /closeup, /droneview, /heroangle, /threequarter, /topview, /views |
| Materials | 35 | Surface materials, finishes, textures, tactile properties | /carbonfiber, /brushedmetal, /glass, /wood, /matte, /ceramic, /leather, /chrome, /gold, /titanium |
| Environment | 33 | Time of day, weather conditions, scene settings | /night, /day, /goldenhour, /sunset, /studio, /rain, /fog, /snow, /desert, /forest, /city, /space |
| Actions | 25 | Targeted editing, object addition/removal, staging | /add, /remove, /replace, /color, /declutter, /renovate, /cleanup, /roomstage, /weathered, /wet |
| Transforms | 23 | Internal structure, X-ray, cutaways, technical views | /xray, /explodedview, /cutaway, /wireframe, /crosssection, /chassis, /internals, /ghostview, /blueprint |
| Output & Ratio | 23 | Aspect ratios, formats, target export destinations | /16:9, /9:16, /1:1, /4:5, /3:4, /wallpaper, /mockup, /banner, /avatar, /thumbnail |
| Lighting | 22 | Studio lighting, rim lighting, atmospheric illumination | /studio, /volumetric, /rimlight, /softbox, /dramatic, /neonlight, /backlight, /highkey, /lowkey |
| Layout | 17 | Multi-panel presentation formats, comparison sheets | /beforeafter, /contactsheet, /storyboard, /materialboard, /colorways, /grid4, /grid9, /diptych, /triptych |
| Optics & Lens | 13 | Depth of field, shutter speed, lens distortion behavior | /bokeh, /shallowdof, /macro, /fisheye, /tiltshift, /longexposure, /motionblur, /deepfocus |
| Fidelity | 11 | Source preservation constraints and enhancement | /preserve, /sameangle, /samebg, /samecolors, /sameidentity, /enhance, /restore, /clean, /upscale |
| Composition | 10 | Framing rules, focal placement, balance | /center, /centeredhero, /ruleofthirds, /symmetry, /flatlay, /hero, /negative-space, /pedestal |
| Background | 8 | Background isolation, seamless backdrops, replacement | /blackbg, /whitebg, /transparentbg, /graybg, /gradientbg, /removebg, /changebg, /bg |
📖 Complete Reference: Explore the full list of all 322 commands in
skills/visual-commands/references/COMMANDS.mdor inspect the canonical machine-readable registry inskills/visual-commands/references/commands.json.
[attach image]
/xray
Preserves the subject identity and perspective while revealing plausible internal structures through a transparent outer shell.
[attach image]
/removebg
Isolates the primary subject with clean edges on a transparent or neutral background.
[attach car photo]
/xray /cinematic /night
Preserves car identity and viewpoint, renders internal mechanical components, relights the scene to night, and applies cinematic color grading.
[attach watch image]
/360view /views 8 /productshot /blackbg
Produces an 8-angle commercial product turnaround on a seamless black background.
[attach shoe concept]
/explodedview /technical /whitebg /16:9
Generates an exploded technical assembly view on a seamless white background in 16:9 widescreen.
[attach room photo]
/remove chair /add walnut desk /goldenhour /sameangle
Removes the specified chair, inserts a walnut desk, shifts lighting to golden hour, and preserves the original camera angle.
[attach product photo]
/beforeafter /explodedview /technical /whitebg
Creates a side-by-side presentation with the original-style product on the left and an exploded technical view on the right.
Visual Commands is an Agent Skill that provides host agents with a deterministic, compact interpretation layer for visual instructions.
- Source Preservation by Default: Visual Commands maintains subject identity, proportions, geometry, camera angle, and background unless explicitly instructed to modify them.
- Deterministic Conflict Resolution: Compatible modifiers are merged; mutually exclusive values resolve cleanly with last-command-wins.
- Natural Language Coexistence: Slash commands blend seamlessly with natural language instructions:
Commands establish reusable visual baselines, while natural language provides prompt-specific context.
/cinematic /night /16:9 Turn this daytime street into a rainy Tokyo scene while preserving the original camera angle.
Several commands accept natural-language arguments that extend until the next slash command:
/remove <object>— Remove specific items from the scene/add <object>— Insert new elements into the scene/replace <from> -> <to>— Swap one element for another/color <target?> <color>— Recolors the subject or specific part/materialswap <from> -> <to>— Swap specific surface materials/changebg <description>— Replace the background with a described scene/bg <description>— Set a custom background description/weather <description>— Apply custom weather conditions/time <description>— Apply a custom time of day/season <season>— Transform season (spring,summer,autumn,winter)/style <description>— Apply an arbitrary visual style/views <number>— Specify the number of multi-view turnaround angles
Visual Commands is packaged as a standard Agent Skill in skills/visual-commands. Install it into your environment using the appropriate method:
In your Codex interface, run:
$skill-installer install https://github.com/natan-webs/visual-commands/tree/main/skills/visual-commands
Or clone manually into your Codex skills directory:
git clone https://github.com/natan-webs/visual-commands.git
mkdir -p ~/.codex/skills
cp -r visual-commands/skills/visual-commands ~/.codex/skills/Clone or copy skills/visual-commands into your personal Claude Code skills directory (~/.claude/skills/) or project-level skills directory (.claude/skills/):
git clone https://github.com/natan-webs/visual-commands.git
mkdir -p ~/.claude/skills
cp -r visual-commands/skills/visual-commands ~/.claude/skills/Clone or copy skills/visual-commands into your Antigravity global skills directory (~/.gemini/config/skills/) or workspace skills directory (.agents/skills/):
Global installation:
git clone https://github.com/natan-webs/visual-commands.git
mkdir -p ~/.gemini/config/skills
cp -r visual-commands/skills/visual-commands ~/.gemini/config/skills/Workspace-level installation:
git clone https://github.com/natan-webs/visual-commands.git
mkdir -p .agents/skills
cp -r visual-commands/skills/visual-commands .agents/skills/For any tool or agent supporting the Agent Skills format (SKILL.md), clone this repository and configure your agent to load skills/visual-commands/SKILL.md.
visual-commands/
├── README.md # Project documentation
├── CONTRIBUTING.md # Contribution guidelines
├── LICENSE # MIT License
├── tests/
│ └── test_compiler.py # Parser and resolution test suite
└── skills/
└── visual-commands/
├── SKILL.md # Core agent skill instructions
├── LICENSE.txt # Skill license
├── agents/
│ └── openai.yaml # OpenAI/Codex agent metadata
├── references/
│ ├── commands.json # Canonical 322-command registry
│ ├── COMMANDS.md # Human-readable command catalog
│ ├── composition.md # Composition & conflict resolution rules
│ └── examples.md # Detailed usage patterns
└── scripts/
└── compile_commands.py # Deterministic syntax parser helper
- Agent Support: Visual Commands is an Agent Skill format compatible with OpenAI Codex, Claude Code, Google Antigravity, and environments that support standard
SKILL.mdskill discovery. - Image Tooling: Visual generation and editing behavior depends on the image tools available in the host environment.
- Python (Optional): Python 3.8+ is only required if running the local compiler parser (
compile_commands.py) or test suite (test_compiler.py).
Contributions are welcome! Whether adding new visual commands, refining composition rules, expanding documentation, or adding test cases:
- Review CONTRIBUTING.md for guidelines on command structure and fidelity rules.
- Add new commands to
skills/visual-commands/references/commands.jsonandskills/visual-commands/references/COMMANDS.md. - Ensure existing tests pass (
python tests/test_compiler.py). - Submit a pull request.
This project is licensed under the MIT License.