Revit Family Creation Services: How High-Quality BIM Families Should Be Built
A Revit family is reusable content with geometry, parameters and behaviour. Poor families slow models and break schedules. This guide explains hosting, connectors, shared parameters, performance and how to brief family creation services.
A Revit family is reusable content: geometry, parameters, connectors, and behaviour that instances inherit in a project. High-quality families schedule, host correctly, connect to Mechanical, Electrical and Plumbing (MEP) systems, and do not drag the project file to a halt. Low-quality families are downloaded meshes, wrongly categorised, over-voided, and full of unused parameters. Revit family creation services exist so that a project’s library matches its purpose instead of matching a random internet search.
Category and host: a door is a door, hosted in a wall, not a generic model sitting on a floor. A light fixture hosts to a ceiling if that is how it is installed. A switchboard sits on a floor or a pad with a clear origin. Wrong category breaks schedules, visibility, and IFC export.
Parameters: instance versus type, shared versus family, with names that match the project shared-parameter file if one exists. Do not invent a parallel “Width1” if the project already uses “Width.” Do not pack fifty unused manufacturer parameters “for later” if later is never.
Geometry: as simple as the Level of Development requires. A coordination family does not need every screw. A marketing family might, and it should not be the coordination family. Separate them or use detail levels honestly.
Connectors for MEP: ducts, pipes, electrical, and control as required, with system classification and flow direction that a project systems person can use. A chiller without connectors is a statue.
Origin and insertion: the insertion point should match how users think (centre of a manhole, back of a switchboard, face of a grille). A family that inserts 400 mm away from the cursor will be placed wrong a hundred times.
Hosting, nesting and the performance budget
Nested families can be elegant or catastrophic. Nested shared families can schedule hardware; nested unshared families can hide geometry you cannot control. Voids used to cut hosts should be justified; too many voids slow regeneration. Imported SAT or mesh geometry should be a last resort for truly irregular products, not the default for a rectangular air-handling unit.
Reference planes should be named and used for constraints. Over-constrained families break when a user types a size. Under-constrained families explode when flexed. Testing means flexing types through the size range in a sample project, not only looking at one type in the family editor.
Work planes, always vertical versus always horizontal, and face-based versus unhosted choices should follow installation. A face-based fire damper that users must place on a duct face is a different product from a level-hosted box.
Shared parameters, schedules and IFC
If the project schedules fire rating, flow, kW, or asset codes, those parameters must be shared and in the family. If the project will export IFC, categories and some parameters affect the result. Singapore IFC+SG mapping, where BIM submission applies, may need specific property sets — follow official mapping files, do not invent a parallel SGPset folklore. See BIM in Singapore.
COBie-style fields, if requested, are a handover brief, not a default. Empty required fields are worse than omitted optional fields because they look complete.
Manufacturer content: reuse with a cleanup milestone
Manufacturer downloads can be a starting point. They are often a problem: wrong units, huge meshes, logos as 3D, no connectors, American categories on a metric project, or a promotional model at a Level of Detail that the project cannot afford. A family creation service should include a cleanup path: rebuild the parametric skeleton, keep a simplified geometry, store the pretty mesh only if a 3D view needs it at fine detail.
Do not let a modeller insert raw manufacturer files into a live hospital model as a shortcut. That is how models die on a Friday.
Testing protocol you should pay for
A family is not done when it looks right in the editor. Test in a sample project: flex types, schedule, tag, section, coarse/medium/fine, print a 2D symbol if 2D is required, connect MEP, copy-monitor if relevant, export IFC if relevant, and check file size. Write the test sheet. Milestone payment against the test sheet, not against a screenshot of the 3D.
Version control: name the family, the type catalogue if used, and the date. A library without versions will fork.
Briefing family creation on a marketplace
List each family: category, host, types, parameter list, connector list, 2D symbol yes/no, max file size if you have a standard, and sample dimensions. Attach a sketch or a cut sheet. Attach the shared-parameter file. State the Revit version. State whether a type catalogue is required. Exclude marketing geometry unless you buy it as a second family.
Price is per unique family and per type complexity, not per building area. See cost. A parametric louver with many sizes is not a one-type trash bin.
Quality checklist
Correct category and host
Insertion origin sensible
Parameters named to the project file
Connectors correct
Flexes without breaking
2D/3D detail levels sensible
File size reasonable
No unexplained imported meshes
Tested in a project, not only in the editor
Version noted
Type catalogues, lookup tables and the user who will actually place the family
A family with forty types in a type catalogue is usable if the names match how a specifier speaks. A family with forty types named Type 1 through Type 40 is a trap. Catalogue files must be versioned with the family. Lookup tables for MEP sizes must be tested at the extremes, not only at the middle size that looks good in a screenshot.
The user might be a junior on a deadline. Insertion should not require a ritual of hidden parameters. Tooltips and type comments help. A thirty-page family manual does not help if the family breaks when flexed. Prefer a family that flexes over a manual that apologises.
Materials and subcategories affect hidden-line drawings and filters. Assign them. Do not leave everything as Default. Filters in shop drawings depend on this boring work.
MEP connectors, flow and the equipment that is really a system interface
Connectors need correct classification, diameter or width/height, and flow direction. Dummy connectors that do not connect are worse than no connectors because they look done. Accessory families (valves, dampers) should connect without forcing the user to build a spaghetti of short ducts. Equipment that requires a maintenance envelope should include a clearly named, easily hidden solid so clash detection can see it. If the client did not buy access solids, the family description should say they are absent.
Electrical connectors and control connectors are often omitted, then added in a panic. If the project will ever systemise lighting or controls, add them at creation. Retrofitting connectors into a nested mess is more expensive than doing it once.
2D symbols, coarse detail and drawings that still plot
Families that look superb in Fine 3D and plot as a black blob in Coarse plan are not finished. Coarse and Medium should be drawing-ready if drawings are a purpose. Symbolic lines, masks, and annotation visibility are part of family creation services. A marketplace package that only delivers a 3D mesh has not delivered a BIM family for a drawing office.
Annotation families (tags) should read the shared parameters you actually use. A tag that reads the wrong width parameter will dimension lies. Test tags with the family in a sample sheet.
Library governance after the freelance package
Someone must own the library: naming, versions, deprecation, and a ban on random downloads. The freelance package should end with a register: family name, category, version, date, test status. Put the register in the BEP folder. If the next modeller cannot find it, they will download a chair.
Intellectual property: if the client paid for a rebuild of a manufacturer product, who may reuse it on the next job? Write it. If the specialist reuses a skeleton they brought, write that too. Silence here is how two companies later argue about a door.
Performance budgets
Agree a target file size per family category if you have a standard. Measure after purge in the family editor. Nested arrays, imported meshes, and void-heavy cutting are the usual offenders. A beautiful grille that adds seconds to every regen across a facade is a project cost, not a content triumph. Ask for a regen test in a sample facade bay, not only in the family editor’s empty space.
Shared parameter strategy for a project library
One shared-parameter file per project (or per office standard) is enough. Families created for the project should load from it. A freelancer who brings a fourth shared-parameter file with similar names will double parameters in schedules. Send the file with the brief. If you do not have one and you need schedules for assets or submission properties, creating it is Milestone 0.
Parameter discipline: type parameters for things that define the product (width of a standard door leaf), instance parameters for things that vary per place (mark, room). Mixing them randomly is how tags fail. Boolean parameters for “has closer” can be useful; nested geometry that always exists and is hidden is a performance cost. Choose.
Units in parameters must match the project: millimetres, litres per second, watts. A family built in imperial and dropped into a metric project is a defect even if the 3D “looks about right.”
Line-based, face-based, and adaptive families
Line-based families help for some linear products (certain gutters, simple rails) and become painful when they need to turn corners the tool cannot. Face-based families are right for many hosted devices and wrong for objects that must sit on a level regardless of a sloped face. Adaptive families can solve pattern-based facades and can also produce a file that only one expert can edit. Brief the pattern. Do not ask for adaptive if a curtain panel family would do. Complexity should follow the geometry, not the specialist’s favourite tutorial.
Testing in a dirty project, not a blank one
Families that work in a blank template can fail in a real project with view templates, phasing, and worksharing. The test project should include: a linked architecture, a section, a schedule, a sheet, a filter, and a phase. MEP families should be tested by connecting to a small system and seeing flow. Doors should be tested in a curved wall if the project has curved walls. If the family cannot sit in a curved wall and the project is full of them, fail it now.
Print a PDF of the 2D symbol at the scale you use most. If it is a black square, keep working.
Versioning and deprecation
When a family changes connectors, old instances in a model may break. Communicate breaking changes. Use a version in the family name or in a type comment. Deprecate rather than silently overwrite if the old types are already placed. The library register should mark deprecated items. Marketplace packages that deliver “final.rfa” twenty times with no version are not libraries.
Content for Singapore submission mapping
If IFC+SG property sets will be required, do not wait until export night to discover the family category cannot carry them. Test one family through export. Official mapping files win. This is not a reason to invent extra local parameters “just in case.” See BIM in Singapore.
Paying for quality versus paying for a catalogue count
Forty poor families are more expensive than twelve good ones with types. Buyers should price unique behaviours (connectors, nested shared hardware, parametric arrays), not a headline count. Sellers should show a type catalogue in the sample. BIMProSG services that say “100 families” without a test sheet are not comparable to services that say “these twenty, tested, shared parameters attached.”
Family editor hygiene
Delete unused types, unused materials, and unused reference planes that were experiments. Purge. Unpin where users will need to move instances. Lock only what must stay locked. A family that cannot be moved without exploding is a defect for anything that is not welded to a host.
Category-specific behaviour (door swing, window embedded sills, plumbing fixtures connecting to pipes) should be tested. Using the wrong category to “make the 3D easier” will break those behaviours and IFC class mapping.
Arrays, formulas and the breaking point
Formulas that divide by zero when a user types a legal size are defects. Arrays that go negative are defects. Test zero, one, and many for array counts. Test the smallest and largest catalogue size. Write the legal range in the type catalogue. Users will type outside it; the family should fail gracefully, not corrupt the project.
Delivery format
Deliver RFA, type catalogue if any, shared-parameter references, a sample project with a schedule and a sheet, and the test sheet filled. A render is optional. The sample project is not. BIMProSG milestones should open the sample project first.
Doors, windows and the host wall
Doors that do not cut the wall correctly, windows that ignore stacked walls, and openings that leave a gap in a fire-rated wall are family defects with code consequences. Test in the wall types the project actually uses, including compound walls. A family that only works in Generic 200 is not done.
Lighting and photometric extras
If the purpose is coordination, a light does not need a full photometric web. If the purpose is lighting analysis, it might. Do not pay for IES data on a clash family. Do not use a clash family for an analysis you then trust. Split the content or the purpose.
Connector direction mistakes
A connector with reversed flow will break systems and schedules. Test by creating a tiny system. If the system cannot form, fail the family. This test is five minutes and saves a week of mysterious MEP behaviour.
Nested shared hardware
If you need to schedule ironmongery or dampers separately, nested shared families can be the right tool. If you do not, nested shared families will clutter schedules. Brief the schedule. Test the schedule. A family that schedules fifty unnamed nested screws is not a gift.
Delivery checklist reminder
Category, host, origin, parameters, connectors, flex, coarse/medium/fine, file size, sample sheet, version. If any item is missing, the family package is not complete even if the 3D is attractive.
Content is infrastructure
A good family is infrastructure: boring, tested, versioned, and reused. A bad family is a tax on every future click. Pay for the test sheet. Demand the sample project. Refuse mystery meshes. Then put the register in the BEP folder so the next modeller does not download a chair. That is the whole library strategy, whether you buy ten families or two hundred.
If a family cannot be scheduled, tagged, and flexed in a sample project, it is not a BIM family yet. It is a sculpture. Pay for the sculpture only if you asked for a sculpture. Coordination libraries are not sculptures.
Test in a dirty project, not only in the family editor. Connect the MEP connectors. Flex the smallest and largest types. Plot the coarse symbol. If those four checks pass, the family can enter the library. If they fail, keep the sculpture off the live model.
Should every product be a new family? No. Types inside a well-built family are better than forty near-duplicate families.
Component versus in-place? In-place is not a library. Use it only for unique project geometry.
Can AI generate families? It may draft geometry or parameters. Testing and connectors remain human. See AI and BIM 2026.
Who owns the family IP? Write it. Many clients need to reuse content. Many sellers reuse skeletons. Be explicit in the order.
Do we need a content manager? If the library is large, yes, even part-time. Otherwise every modeller will download a new chair.
What about annotation families and tags? They are families too. Brief them if your template depends on them.
IFC identity? Category and IfcExportAs mapping may be required. Test one family before you build eighty.
Sources and reference notes
Autodesk Revit help on the family editor, connectors, shared parameters and nested families is the primary product reference. buildingSMART IFC mapping depends on export setup. BIMForum LOD examples sometimes illustrate element geometry; they are not a family-editor manual. Official Singapore IFC+SG mapping files, where applicable, come from CORENET X / BCA resources, not from marketplace blogs.
Image brief
Family editor with named reference planes and a type catalogue, plus the same family scheduled in a project. Do not show a mesh-heavy manufacturer download as the hero “quality” image.
A buyer’s guide to Revit modeling services: architectural, structural and MEP scopes, file hygiene, family work, deliverables, pricing factors, and a hiring checklist for specialist freelancers.