Gemini will not draw you a dimensioned, manufacturing-ready part, and any prompt that promises otherwise is handing you a picture, not a drawing. But Gemini has one edge the other chatbots do not, and the smartest prompts play straight to it: its vision. It is genuinely sharp at reading a drawing you upload, decoding the symbols, reviewing it for holes and gaps, and then doing all the language and code around the geometry. Below are 18 prompts that lean on those strengths, grouped by the job they do, each ready to copy, paste and fill in.
How to prompt Gemini well
A prompt is only as good as what you put into it. Four habits make every prompt below work far better, and the first one is where Gemini differs from ChatGPT:
- Give it something to look at.Gemini's vision is its advantage, so upload the drawing, the photo or the sketch whenever you can. A prompt with an image attached beats the same prompt with a paragraph of description.
- Give it a role and a standard.“Act as a mechanical drafter working to ASME Y14.5” sets the register and the conventions before it answers.
- Be specific with real numbers. Replace every
[PART]with the actual part, dimensions, material and process. Vague in, vague out. - Verify before you build. Gemini reads and answers with confidence even when it is wrong, so treat every output, especially a dimension lifted off an image, as a first draft to check.
If you want to see how far Gemini actually gets on its own first, we put it through the read-versus-draw test in can Gemini make technical drawings. And if you live in ChatGPT instead, there is a parallel set of ChatGPT prompts for technical drawing.
Read & review a drawing you upload
This is where Gemini earns its keep and pulls ahead of the other chatbots. Its long-context vision is well suited to a dense, symbol-heavy drawing, so these five prompts all start by uploading something and asking it to read. Pair them with our guide on how to read a technical drawing when you want a human reference alongside.
Photo
Drawing1. Walk me through a drawing I uploaded
Turns a dense sheet into a plain-English explanation.
I've uploaded a technical drawing. Act as a mechanical drafter and walk me through it: identify each view and how they relate, list the main dimensions and what feature each one controls, explain every symbol and note, and summarise what the part is and how it would be made. Flag anything that looks ambiguous or contradictory.
2. Review an uploaded drawing for missing information
Surfaces the questions a machine shop would email back.
Review this drawing image the way a machine shop would before quoting it. List everything that would stop them making the part: missing or unreadable dimensions, untoleranced critical features, no material or finish, no projection symbol, ambiguous notes. Return each gap as a specific question the shop would have to ask.
3. Decode every symbol and abbreviation on the sheet
Clears up the shorthand on someone else's drawing.
From the uploaded drawing, extract every symbol, abbreviation and callout you can see (for example ⌀, R, THRU, CBORE, TYP, SF, the GD&T frames and the surface-finish marks). For each one, give the symbol, its meaning, and what it tells the machinist. Note any that differ between ISO and ASME conventions.
4. Pull a structured dimension and hole table from the image
Uses Gemini's OCR to turn callouts into a table you can check.
Read all the dimensions and hole callouts off this drawing image and return them as a table with columns Feature, Nominal, Tolerance, Notes. List holes separately with diameter, depth (or THRU) and quantity. At the end, flag any value you are unsure you read correctly so I can verify it against the original.
5. Compare two revisions and list what changed
Catches the one dimension that quietly moved between revs.
I've uploaded two revisions of the same drawing, [REV A] and [REV B]. Compare them and list every difference you can see: changed dimensions or tolerances, added or removed features, edited notes, and any change to material, finish or title block. Present it as a change log, and note anything that would affect fit or function.
Plan the drawing & write the spec
Before any geometry exists, Gemini can help you decide what the drawing needs to say, which is the same order we follow in how to make a technical drawing. Feed it a description or a photo of the part and it will plan the sheet for you.
Photo
Drawing6. Choose the right views
Stops you over- or under-drawing a part.
Act as a mechanical drafter working to ASME Y14.5. I need to draw this part: [DESCRIBE PART, or say "see the uploaded photo"]. List the minimum set of orthographic views that fully defines it, tell me which face should be the front view and why, and say whether I need any section or detail views.
7. Plan the dimensioning and datum scheme
Gives you a datum strategy before you place a single dimension.
For this part [DESCRIBE / LIST FEATURES / or reference the uploaded image], propose a dimensioning scheme: which edges or faces to use as datums, whether to use baseline or chain dimensioning for each feature group, and the order to dimension features so that every feature is defined exactly once and nothing is over-dimensioned.
8. Turn a description or photo into a drawing spec
Converts loose intent into a structured starting point.
Convert this part (uploaded photo, or the description below) into a structured drawing specification with sections for Views, Key dimensions, Tolerances, Material, Surface finish and General notes. Ask me up to 5 clarifying questions first if anything critical is missing. Description: [PASTE, or "use the image"].
9. Draft the general-notes and title-block content
Produces the boilerplate text you drop straight onto the sheet.
Write a concise, shop-ready general-notes block and the title-block fields for a machined [MATERIAL, e.g. 6061-T6 aluminium] part called [NAME], part number [NUMBER], scale [1:2], units [mm], projection [third-angle], finish [FINISH], revision [A]. Cover default general tolerance, break-all-sharp-edges, deburring, coating and inspection. Use numbered notes like a real production drawing.
Check whatever it drafts against our manufacturing-ready checklist before the sheet goes out.
Standards, GD&T & symbols
Anything that is fundamentally language, symbols, standards and definitions, is squarely in Gemini's wheelhouse. Pair these with our engineering drawing symbols and GD&T guide when you want a human-written reference too.
Photo
Drawing10. Explain or choose a GD&T callout
Decodes a feature control frame, or picks the right one.
Two things. First, explain this GD&T feature control frame in plain English: [e.g. position 0.2 (M) | A | B(M) | C], covering the characteristic, the tolerance zone, the material-condition modifiers and the datum order. Second, I need to control that [REQUIREMENT, e.g. a bore stays perpendicular to the mounting face within 0.1 mm]: tell me which characteristic to use, what datums I need, and write the frame to put on the drawing.
11. Recommend a tolerance strategy
Separates the features that need precision from the ones that do not.
Recommend a tolerancing strategy for this machined part [DESCRIBE or reference the image]. Suggest a sensible ISO 2768 general-tolerance class for the bulk of the dimensions, then list which specific features (mating diameters, hole positions, sealing faces) should get tighter individual tolerances, with a suggested value and the reason. Flag where I should consider GD&T instead of plus/minus.
12. Check first-angle vs third-angle projection
Avoids a part being machined back to front.
Explain how to tell first-angle from third-angle projection on a drawing, how the projection symbol looks for each, and exactly how a part would come out wrong if I read a third-angle drawing as first-angle. Then, from the uploaded drawing, tell me which convention it uses and how you can tell.
Geometry as code & driving CAD
This is the closest Gemini gets to actually “drawing”, and it is also where its agent side shows up. Because SVG and DXF are text formats it can write them, code like OpenSCAD builds real geometry, and through function calling it can operate a CAD tool that exposes an API. All of it is reliable only for simple shapes, and it cannot judge scale, so always import and check.
This short demo shows the idea end to end: handing Gemini a 2D drawing and having it drive FreeCAD toward a 3D model, exactly the read-plus-code workflow prompts 13 to 15 set up.
Photo
Drawing13. Generate a minimal DXF or SVG to import
A starting cut file you refine in CAD or our converter.
Generate a minimal ASCII DXF (AutoCAD R12) for this geometry: [e.g. a 100 x 60 mm plate with a Ø20 mm hole centred]. Use only LINE, CIRCLE and LWPOLYLINE entities, set units to millimetres, keep all profiles closed, and output only the DXF text so I can save it as a .dxf file. Then give me the same shape as clean, minimal SVG in millimetre units.
14. Build a parametric model in OpenSCAD
Lets you tweak dimensions and re-export.
Write OpenSCAD code for a parametric [PART, e.g. spacer / bracket] with named variables for every dimension [LIST DIMS]. Comment each variable, keep it printable and exportable, and structure it so I can change one number and regenerate. Note how to export it to STL and to a 2D DXF profile.
15. Drive a CAD tool as an agent
Uses Gemini's function calling to operate a modeller, not just talk about it.
I'm connecting you to [CAD TOOL / API, e.g. a FreeCAD or Onshape scripting endpoint] through function calling. Plan the sequence of tool calls to build this part [DESCRIBE with dimensions]: create the sketch, apply constraints, extrude, add holes, then export a DXF of the [VIEW]. Describe each step and the exact call before you make it, and pause for my confirmation between sketch and features.
Reverse engineer & the photo hand-off
These help when you are working from a physical part or a photo rather than a model, and they draw the honest line around what a chatbot can plan versus what it can actually deliver.
Photo
Drawing16. Turn caliper measurements into a clean spec
Converts raw readings into likely nominal sizes and fits.
I measured these features off a physical part with calipers: [LIST measurements]. Turn them into a clean dimensioned specification: suggest the likely nominal sizes (a 9.98 mm shaft is probably 10 mm), identify probable standard fits, propose sensible tolerances, and list what else I still need to measure to fully define the part.
17. Plan the photo-to-drawing workflow
Maps the whole route when all you have is a snapshot.
I have only a photo of a part and need a manufacturing drawing and a DXF from it. Lay out the exact step-by-step workflow to get there: how to establish real scale from the photo, what to measure, which steps a person or a dedicated tool must do versus what you can help with, and how to verify the result before I send it to a shop. Be honest about what you cannot do yourself.
18. Sanity-check a finished drawing before you cut
A last, tireless second pair of eyes on the sheet.
Here is my finished drawing (uploaded). Do a final sanity check before I release it: is every feature dimensioned exactly once, do the views agree with each other, are the tolerances realistic for the process, is the title block complete, and would the part actually assemble? List anything you would fix, ranked by how likely it is to cause a scrapped part.
The drawings paired with the photos throughout this post are generated illustrations, not inspection-grade output. The point is the workflow, not the exact values. Always verify dimensions against the real part.
Where Gemini stops and a real tool starts
Every prompt here works because it plays to what Gemini is good at: reading an image, language, structure, code and reasoning. The one thing it cannot do is produce accurate, scaled geometry of your actual part. It cannot measure, it cannot hold a tolerance, and it cannot recover true scale from a single photo. Its image model will happily paint something that looks like a blueprint, but the numbers on it are decorative and contradict each other. That is not a prompt you can write your way around; it is the boundary of the tool.
So the workflow that actually ships parts is a hand-off. Use Gemini for the reading, the reviewing and the text. Use a purpose-built tool for the geometry: our photo to manufacturing drawing workflow and image to DXF converter turn a real part or image into editable, dimensionable output, and TechDraw AI adds the measurement step a chatbot is missing. For the wider landscape of what each tool is for, see the best AI technical drawing tools, and for a physical part in hand, our guide to reverse engineering a part.
Get the division of labour right and Gemini stops being a novelty that paints wobbly blueprints and becomes what it should be: the best drawing reader you have ever had, sitting next to a tool that does the part it cannot.
Frequently asked questions
Can Gemini create a technical drawing from a prompt?
Not a real, to-scale, dimensioned one. Gemini cannot produce accurate CAD geometry, so no prompt will make it hand you a drawing a shop can cut from directly. What it does exceptionally well is the reading and the thinking around a drawing: parsing a drawing you upload, planning views, decoding GD&T, writing notes and title-block content, and generating simple geometry as SVG or DXF code you import and correct. The image model can paint something that looks like a blueprint, but the numbers on it are invented.
What is the best prompt to get a drawing out of Gemini?
There is no single magic prompt. The best results come from giving Gemini a role, an actual part with real dimensions or an uploaded image, the manufacturing process, and the standard you work to, then asking for one thing at a time. Gemini's strongest prompts lean on its vision: upload a drawing or a photo and ask it to read, review or plan, rather than asking it to draw from nothing.
Can Gemini read or interpret an engineering drawing I upload?
Yes, and this is its real strength. Gemini's long-context vision parses a 2D drawing, runs OCR on the callouts, counts holes and features, identifies symbols, and flags things that look wrong against a rule set. Treat it as an excellent reader and reviewer. Just verify any exact dimension it lifts off an image before you quote or cut, because it reads with confidence even when it is wrong.
Can Gemini output a DXF or CAD file?
It can write the text of a simple DXF or SVG, since both are text-based formats, and you can save and import that into CAD. It is only reliable for basic geometry such as rectangles, circles and simple polylines, and it cannot judge scale or fit. For anything past a simple shape, generate the geometry with a purpose-built tool and use Gemini for the reading, planning and notes around it.
Is Gemini or ChatGPT better for technical drawing prompts?
For reading and reviewing a drawing you upload, Gemini's vision is usually the stronger of the two, so the reading prompts here pay off more on Gemini. For planning, notes and code, the two are close. Neither can produce a manufacturing-ready drawing on its own. If you work mostly in ChatGPT, we have a parallel set of prompts for it too.
Sources
- Google AI: Gemini API vision and document understanding (image input, OCR)
- Google AI: Gemini API function calling (driving external tools)
- Google AI: Gemini image generation (raster output)
- ASME Y14.5: Dimensioning and Tolerancing
- ISO 2768: General tolerances for linear and angular dimensions
- Autodesk: About the DXF format