AI guide
ChatGPT /explode /sketch /visualize — Which Slash Image Commands Are Real
Those three slashes are all over X and Reddit. Here is the honest split: they are not secret ChatGPT APIs. They are style labels. The picture still comes from the sentences you write after them — and those sentences are below, ready to paste.
By Pixelandpdf · Updated 18 August 2026 · 11 min read
If you searched ChatGPT /explode /sketch /visualize, you probably saw a screenshot of someone typing one word with a slash and getting a perfect exploded diagram, a pencil sketch, or a labelled infographic. The caption usually says “hidden command” or “unlock this.” That is the hook. It is also the part that is mostly wrong.
OpenAI does document real slash commands — they appear when you type `/` in the composer, and they change what the product does next (Canvas, Study, image tools when your account has them). `/explode`, `/sketch` and `/visualize` do not show up in that official list. They work, when they work, because ChatGPT Images reads them as English: explode the parts, draw it as a sketch, make a diagram. The slash is a bookmark, not a cheat code.
This guide stays on those three commands. For `/image`, product shots and thumbnails, use our separate ChatGPT image generator prompts with slash. Here you get a 30-second test to tell a real UI command from a costume, then copy-paste briefs for exploded views, pencil sketches and visual explainers — plus what to do when the PNG is too heavy to email.
Quick steps
- In ChatGPT, type / once and see whether your account lists a real image action. Note what is missing: /explode, /sketch and /visualize will almost never appear as menu items.
- Start image generation the normal way: pick the image tool if you have it, or write “Create an image of…”.
- Paste one of the three briefs below. Keep on-image labels short and in quotation marks.
- If the first render is close, change one thing: “separate the battery further”, “lighter pencil”, “fewer labels”.
- Download the file, then compress it if you need it under an email or CMS size cap.
Are ChatGPT /explode /sketch /visualize real slash commands?
A real ChatGPT slash command is an interface action. You type `/`, a menu appears, you pick a row, and the product switches mode. OpenAI’s own slash-command reference is a list of those actions, and it changes by app (chat vs Codex-style tools). Image generation, when it is offered, is usually `/image` or a tools button — not a secret verb for exploded views.
ChatGPT /explode /sketch /visualize fail that test. Type `/explode` in a fresh chat. If no official suggestion appears, you are not invoking a hidden API. You are sending a short English instruction. The model may still draw an exploded mouse, a graphite sketch or a labelled diagram, because those words are strong art-direction. They are not more official than writing the same request as a sentence.
That distinction matters when a post promises “unlocks.” Nothing is unlocked. If your plan cannot generate images, `/explode` will not grant it. If image generation is on, a plain sentence (“exploded technical illustration of a mechanical keyboard, parts separated and labelled”) will do the same job, and usually a better one, because you can name the parts.
- Real command: it appears in the / menu for your account.
- Costume command: /explode, /sketch, /visualize — style labels the model reads as English.
- If / does not list it, write the picture you want in ordinary sentences.
The 30-second test before you trust a viral slash
Open ChatGPT on the device you actually use. Tap the composer and type only `/`. Scroll the suggestions. Screenshot them if you like — that list is your ground truth for today, on this plan, in this app. Compare it to the viral list. Anything that is not in the menu is a prompt, including ChatGPT /explode /sketch /visualize.
Second check: paste the slash alone, with no subject. `/explode` with nothing after it often produces a generic “exploded something” or a confused question. A real tool-opener still opens the tool. A costume needs a noun. Always give it a noun: which object, which parts, which labels.
Third check: rewrite without the slash. If “Create an exploded technical illustration of a wireless mouse, parts floating in order, short labels” matches `/explode wireless mouse`, you have proof the slash was flavour. Keep the slash in your notes if it helps you find the prompt later. Do not treat it as a licence key.
What /explode is actually asking for
In industrial drawing, an exploded view shows every part pulled apart along an axis so you can see how the object is assembled. Repair manuals, product keynotes and IKEA-style diagrams all use it. When people type `/explode`, they want that grammar: pieces in a line, slight gaps, maybe callouts. ChatGPT Images can do a convincing version if you specify the object, the parts that must appear, and whether labels are allowed.
Where it fails: unnamed internals. “/explode iPhone” will invent a fantasy teardown and may smear logos you should not request. “/explode a generic wireless mouse: top shell, bottom shell, scroll wheel, battery, two screws, small board” is a job a model can finish. Name materials (matte plastic, brushed aluminium) so the parts do not all look like grey clay.
Labels should be short. Ask for three to six callouts in a small sans-serif, each one or two words, placed beside the part — not a paragraph of instructions inside the frame. If spelling breaks, reply: “Keep the same exploded layout. Fix only the labels to exactly BATTERY, BOARD, WHEEL.” Freeze the picture; edit the text.
What /sketch and /visualize are actually asking for
`/sketch` is a media request: graphite or ink on paper, visible strokes, maybe a spiral notebook edge. It is useful when you want an idea, not a product photo. Say pencil hardness and paper colour if you care (“2B on cream paper, light construction lines still visible”). If you only type `/sketch`, you often get a generic portrait. Name the subject and the viewpoint (three-quarter, overhead, orthographic).
`/visualize` is the vaguest of the three, which is why viral posts love it. People use it to mean “turn this explanation into a diagram.” The model cannot guess your diagram type. Say infographic, flowchart, labelled cross-section, or simple isometric. Put the only on-image words in quotes. Long sentences inside the picture still fail in 2026 — keep the chart labels to single words.
Do not stack all three slashes in one prompt (`/explode /sketch /visualize the mouse`). You are asking for an exploded photo, a pencil drawing and a diagram at once. Pick one job per generate, then ask for a second version in another style if you need both a teardown and a sketch of the same object.
Copy-paste: ChatGPT /explode /sketch /visualize briefs
Paste the whole block after you have started an image turn. Swap the object and the quoted labels. These are not official commands. They are briefs that survive if you delete the slash.
- /explode — Create an exploded technical illustration of a matte charcoal wireless mouse. Separate these parts in a straight line, small gaps: top shell, bottom shell, scroll wheel, AA battery, tiny circuit board, two screws. Soft studio light on a pale grey sweep. Short labels exactly "SHELL", "WHEEL", "BATTERY", "BOARD". No brand logos, no watermark, no extra tools.
- /explode — Create an exploded view of a metal pour-over kettle: handle, lid, body, spout, base. Parts hover in assembly order, slight shadows on a white table, photoreal product photography, 16:9. Labels exactly "LID", "BODY", "SPOUT". No logo, no watermark.
- /sketch — Create a pencil sketch of the same wireless mouse, three-quarter view, 2B graphite on cream paper, light construction lines, coiled cable suggested in two strokes. No colour, no photoreal render, no text, no watermark.
- /sketch — Create an ink line sketch of a city balcony garden, fountain pen, cross-hatching for shadow, white paper, slightly messy artist notes in the margin that are not readable words. No colour, no watermark.
- /visualize — Create a simple three-panel diagram on white: how a mouse click works. Panels labelled exactly "PRESS", "SWITCH", "SIGNAL" in bold sans-serif. Flat technical illustration, generous margins, arrows between panels. No paragraph text, no watermark.
- /visualize — Create a labelled cross-section of a rechargeable head torch, side view, cutaway showing battery, LED, reflector. Labels exactly "LED", "CELL", "SWITCH". Clean manual style, pale background, no brand, no watermark.
When the slash “fails” (and what to type instead)
If ChatGPT replies with text instead of a picture, you never entered image mode. Say “Generate an image now, do not only describe it,” or pick the image tool from + / attachments. `/explode` cannot force a picture on an account that has generation turned off or quota exhausted.
If you get a pretty object that is not exploded, the model treated `/explode` as decoration. Repeat the mechanical instruction in the next message: “Same object. Pull every part apart along a horizontal axis. Do not leave it assembled.” One constraint per reply beats rewriting the whole prompt.
If `/sketch` still looks like a photo, forbid the photoreal path: “No camera, no lighting setup, no CGI. Graphite only, paper tooth visible.” If `/visualize` turns into a movie poster, specify the diagram type and a white background. Costume slashes are weak; constraints are strong.
After you download the PNG
Exploded views and sketches from ChatGPT are often huge PNGs. Fine for a folder; painful for Slack, a CMS, or a form with a 200 KB cap. Compress the image in the browser, use JPEG or WebP for anything that is not a sharp diagram with tiny type, and cap the long edge if you only need a blog width.
Diagrams with labels sometimes survive better as PNG. Photos and sketches usually shrink well as JPEG. None of that needs an upload to a random converter — PixelAndPDF’s compress and format tools run in the tab. If you are budgeting long prompt chains on the API side, the token calculator is a separate, local estimate; ChatGPT image quotas still follow your OpenAI plan.
- Compress Image — optional KB target for email and forms.
- JPG to WebP — lighter heroes for a site.
- PNG to JPG — when a sketch does not need a transparent background.
Frequently asked questions
Almost never as official UI commands. Type / and check your menu. If they are missing, they are style labels ChatGPT reads as English — exploded view, pencil sketch, or diagram — not a hidden API.
It can, if image generation is on and you name the object and parts. It does not unlock a secret teardown engine. A full sentence (“exploded technical illustration of…”) is more reliable than the slash alone.
It asks for a drawing look — graphite or ink, paper, visible strokes. Add the subject and viewpoint. Without a noun, you often get a generic portrait sketch.
People use it to mean “make a diagram.” Say the diagram type (cross-section, three-panel, flowchart) and put short labels in quotes. /visualize alone is too vague.
No. Hidden-command posts are mixing real / menu actions with costume prefixes. /explode is in the second group. OpenAI’s slash list does not document it as an image mode.
Describing the view is better. Name the parts, the axis they separate on, and any labels. Keep the /explode tag only if it helps you file the prompt.
The slash was treated as flavour. Reply with hard constraints: graphite only, no camera, paper texture, no CGI. Change one instruction per edit.
You can type them, but you are mixing three jobs. Generate once as exploded, once as sketch, once as diagram. Ask for the same object if you need a set.
You need some way to start image generation — /image if your menu has it, or “Create an image of…”. /explode does not start image mode by itself.
Download, then compress with a browser tool (JPEG or WebP, optional max edge or KB target). PixelAndPDF’s compressor stays on your device.