DEV Community: Asher Hu

10x quote gap: Class-A surfaces and T1 mold support vs render-only

Asher Hu — Tue, 10 Nov 2026 02:49:12 +0000

The same appearance brief can be quoted at five thousand and at fifty thousand, and both can be honest. They are not the same product. They differ along three lines, and if you do not align those lines, you are not comparing prices, you are comparing numbers that happen to be written in the same currency.

Delivery depth

Five thousand yuan buys one or two form directions, rendered images, and a basic 3D reference. It suits internal kickoff decks, packaging renders, and small modifications to an existing public-mold shell. Designers at this tier generally do not touch structure, and the quote does not include it. That is not corner-cutting, it is the product definition.

Projects go wrong when the buyer assumes five thousand contains fifty thousand of content. The renderings go to a tooling shop, and the shop asks where the undercuts are, where the parting line runs, who modelled the snap fits. Nobody can answer. Rescuing that situation after the fact typically costs 1.3 to 1.8 times the original design fee, because the missing work has to be reconstructed from a shape that was never built to be manufacturable.

Responsibility end point

A full-package quote is really three things: design hours, validation rounds, and a risk reserve. Two prototype rounds, two mold-trial visits, and two revision rounds written into the contract mean the project probably will not produce a surprise invoice. The risk reserve buys exactly that certainty.

Cheap-plus-metered contracts look different line by line. A revision round is 3,000. An extra prototype is 5,000. A mold-trial visit is 4,000. Each line is reasonable, and the completed project often costs more. What clients dislike is not being charged, it is being surprised, so the boundary should be drawn explicitly: in-scope changes are free, out-of-scope changes (reversing a direction, adding an SKU, altering an approved surface) have published prices.

The middle band

Most real orders land between fifteen and thirty thousand: appearance plus structural hand-off, manufacturable data, no long mold-trial support. Within that band the spread comes almost entirely from how far the structural hand-off goes. Some vendors produce stacking diagrams only. Some deliver full screw bosses and snap positions. Some include a design-for-manufacture report.

The practical move for a buyer is to send a three-line brief to every candidate: what deliverables, how far the responsibility runs, how changes are priced. The same brief to five vendors produces comparable quotes. A one-line request for "an appearance, how much" produces five numbers that cannot be compared at all.

5% more thinner dropped film from 38 to 26μm — and killed the soft-touch feel

Asher Hu — Mon, 09 Nov 2026 05:37:12 +0000

Two production lines were spraying the same batch of soft-touch paint onto the same part. Line A came out like silk. Line B came out like fine sandpaper. Both gloss meters read 4 to 5 GU at sixty degrees, and under the light booth the two panels were indistinguishable.

The paint was not the problem. The batch number was the same, drawn from one drum. What differed was everything around the paint: how much thinner went into the bucket, how thick the film ended up after two passes, and how hot the first zone of the curing oven ran that afternoon. Three readings, taken on the line in five minutes, would have told the whole story.

Soft touch is a coating category, not a gloss level. Matte and soft touch overlap in the 2 to 6 GU and 3 to 10 GU bands respectively, so a gloss reading alone cannot tell you what a finger will feel. The feel comes from the particles suspended in the film and from a low-surface-energy additive that has to migrate to the surface while the film is still wet.

Matte and soft touch are two different formulations

Conventional matte uses silica, three to eight microns, hard and angular. The hand feels the mechanical bite of those edges. Soft touch uses silicone or polyurethane microspheres, five to fifteen microns, round and slightly elastic, so the finger rolls over them instead of catching. The microspheres are paired with a low-surface-energy additive that pushes the water contact angle above one hundred degrees, which is why hand perspiration beads instead of spreading.

Both routes sit inside a similar gloss window, which is exactly why the two are confused in specifications. A matte part sold as soft touch will be returned the moment a customer touches it. Matte is also cheaper and far easier to spray: wider process window, fewer additives, fewer ways to fail.

Five percent more thinner cost twelve microns of film

On line A the paint measured eighteen seconds in a number four Ford cup. On line B it measured fifteen. Somebody had added roughly five percent extra thinner to keep the viscosity comfortable in the heat.

That drop in viscosity cuts solids content. Sprayed twice, the same way, line A built a thirty-eight micron film and line B built twenty-six. Twenty-eight microns is the line below which the microspheres start sinking into the resin and the hand drops a full grade. Line B was two microns under the threshold that makes soft touch worth paying for.

The normal soft-touch window is twenty-eight to forty microns. Above forty-five, sagging and material cost become the problem. A process card should leave at least five microns of margin above the floor; running at twenty-eight exactly means the next hot afternoon puts you under it.

The additive has to float up, and a hot oven traps it

The low-surface-energy additive migrates to the air-liquid interface because of a surface energy gradient. That migration only happens while the film is still wet. If the first oven zone runs fifteen degrees above the process card, the surface skins over early and the additive is locked below it.

The field test is a thirty-two dyne pen. Draw a line and wait two seconds. If the line contracts into beads, surface energy is low enough and the additive made it to the surface. If it spreads flat, the surface energy is too high and the additive never arrived. On line B the line spread flat.

There is a second trap: adding more booster does not simply make the surface slipperier. It does, but abrasion resistance falls with it, and a fingernail leaves a white mark. The conservative additive level plus correct film thickness and correct oven profile is more reliable than loading the paint.

How much difference a finger can actually feel

Ten microns of film thickness is perceptible. A thirty micron panel and a twenty micron panel are distinguishable by an ordinary hand. Five microns is not, which is why arguing over a two micron deviation in a measurement report is pointless while nobody has checked whether the whole part is under the floor.

Hand feel also depends on who is touching it. The only defensible review condition is twenty-three degrees Celsius and fifty percent relative humidity, hands wiped with a lint-free cloth. A workshop floor in summer, with damp hands, makes every sample feel slick. One program passed review on the line and was rejected by a customer in northern China in winter, where the air is dry and every hand is rougher.

Soft touch also ages. Silicone additives migrate away slowly and hand oils work at the surface, so a dry, grippy finish becomes slightly tacky over six months to a year. Gloss barely moves, which is why the change is easy to miss in a re-measurement. Keep an unused original alongside the approved sample.

FAQ

Can matte and soft touch be substituted for each other? No. The gloss ranges overlap but the formulations are different routes. Matte is a hard-particle mechanical effect, soft touch is microspheres plus a low-surface-energy additive. Substituting one for the other gets the part rejected at first touch.

What should a line check first when the feel is wrong? Film thickness, then a thirty-two dyne pen. Those two steps localise roughly eighty percent of feel problems. Only if both are in range does the paint itself become the suspect. Splitting one batch into two mixing buckets is the most common failure point in the whole process.

What film thickness should soft touch run? Twenty-eight to forty microns, with five microns of margin written into the process card. Above forty-five microns, sag and cost take over.

Does the feel change over time? Yes. Six months to a year, dry and grippy turns slightly tacky while gloss stays put. Keep an unused original sample for comparison.

Source: Hzdesign — industrial design, structural design and CMF for mass production, Dongguan.

Mold cost tiers: multi-cavity steel loses 3-8 RMB/part under 50k units

Asher Hu — Mon, 09 Nov 2026 00:36:21 +0000

Mold cost for an enclosure splits into three tiers, and choosing the wrong tier is not a one-off expense. It is a per-unit tax that follows the product for its whole life.

The soft tier, aluminium or resin tooling, runs 8,000 to 25,000 yuan and lasts 5,000 to 30,000 shots. Its life and accuracy ceiling are both low. Resin tooling starts wearing after 3,000 to 5,000 shots, and textured parts sooner. Aluminium reaches 30,000 to 50,000 shots, but sliding cores and lifters wear faster than steel, so batch consistency fails before the tool does.

The right way to use a soft tool is validation plus first launch: shoot T0 parts to check assembly, put three to five thousand units on the market, and cut the steel tool in parallel. Running a soft tool as if it were production tooling shows up around shot twenty thousand, when flash grows thick enough that the assembly station starts hand-trimming every part.

Single-cavity steel

Three to eighty thousand yuan, thirty-thousand-plus shots. This is the mainstream band for five thousand to fifty thousand units a year, and it is where most products should live.

The quoted price varies by a factor of two for the same nominal specification. The difference sits in four places: steel grade (P20 versus 718H, roughly thirty percent), mold base class, hot runner type (valve gate versus open, 8,000 to 20,000), and the least visible item of all, whether the tool shop includes a trial run and on-site support. Trial runs cost 1,500 to 3,000 per machine day plus a setup technician plus rework. A quote that lists trials separately looks cheaper and finishes more expensive.

Multi-cavity, and the cost people forget

A two-cavity tool is not simply two parts per cycle. Projected area doubles, so the machine goes up a size class (168 to 268 tonnes) and machine rates rise 35 to 50 percent. Cycle time grows 10 to 20 percent because of shot size and cooling. A part that cycles in 38 seconds single-cavity commonly runs 46 to 50 seconds two-cavity.

So the real multi-cavity arithmetic trades higher machine rate and longer cycle against reduced labour and mold amortisation. Above two hundred thousand units a year it wins clearly. Around one hundred thousand it has to be computed part by part, and between those two figures there is almost always a middle option worth tabulating: single cavity plus semi-automation.

3 cost blocks in a design quote: 50-60% design, $210-1,120 mockups

Asher Hu — Sun, 08 Nov 2026 08:49:32 +0000

An honest appearance-design quote breaks into three blocks: the design fee, the prototype fee, and the mold-trial support fee. When a quote shows one lump sum, either the risk is priced into the total, or those stages are simply not included. The second case is the dangerous one, because the money gets paid later, elsewhere, at a worse exchange rate.

Take a desktop smart speaker. A full appearance package runs 24,000 to 36,000 yuan: three directions, two rounds of refinement, A-class surface data, and a CMF specification. Prototypes add 6,000 to 12,000 for two rounds of CNC or 3D printing with surface treatment. Mold-trial support adds 4,000 to 8,000 for two site visits. The total lands between 35,000 and 55,000. A quote under 10,000 usually stops at rendered images, and neither the structural hand-off nor the production responsibility is in the price.

What the design fee actually buys

Not drawings. Rounds of argument. A 10,000 quote gives you three renderings. A 30,000 quote gives you competitor teardowns, brand-tone reasoning, three form directions, two rounds of client review and refinement, A-class surface data, a CMF document, and three rounds of collision with the structural team while the drawing is still soft.

The difference is not sheet count, it is how many times the question "why does this shape work" got answered. In our projects, appearance packages argued through two or more rounds went into tooling with an average of 1.2 mold revisions. Projects that went from rendering straight to tooling averaged 3.4. A mold revision costs 8,000 to 30,000 yuan and two to four weeks.

Why prototypes are a separate line

A prototype is where "looks right" becomes "feels right". The price spread, 1,500 to 8,000 per set, is almost entirely surface treatment: paint, plating, texture transfer. The first round exists to expose problems that no screen shows: button feel, the visual weight of a seam, interference when the hand grips the part.

Bundling prototypes into the design fee is possible, but ask how many rounds are included. The industry default is one. Everything after that is billed separately, and a comparison that ignores this is not a comparison.

The line everyone cuts

Mold-trial support is a designer standing at the tooling shop during T0 and T1: sink marks, weld-line routing, the appearance of the parting line. Those decisions take ten minutes on site and three days over photographs. Without this line in the quote, a trial problem leaves two outcomes: the client referees a fight between the molder and the design firm, or pays extra for an improvised fix. What the two visits cost is usually smaller than the mold revision they prevent.

Paint-free molding versus painting: what you save and what bites back

Asher Hu — Sat, 07 Nov 2026 02:56:35 +0000

Paint-free molding does not save paint. It saves the paint line: fixtures, paint and labour, rework risk, and the paperwork around VOC compliance. Unit cost typically drops 15 to 30 percent, yield risk turns into a single injection cycle, and the environmental file gets thin. That is the upside.

There are three hard boundaries, and they are physical, not negotiable.

Undercut or two-tone stacking depends on a split between textures. When the depth difference between the two leather-grain regions exceeds 0.02 mm, a white witness line appears at the seam. No press setting fixes it. The mold gets reworked or the parts are scrap.

High-gloss paint-free resins can reach 60-degree gloss above 85, close to a painted primer look. They fear flow marks: when the melt-front temperature spread exceeds about eight degrees Celsius, ripples form, and large parts are the worst offenders. That means zone-controlled mold temperature, which means water-line cost you should budget before the design is frozen.

Glass-filled paint-free grades hide fibre float, but not perfectly. In dark colours under direct sun, stipple highlights remain. If the acceptance criteria do not specify the inspection light angle, the argument happens at delivery.

When not to use it

Three product types get a straight no. Figurine-grade parts that need deep vacuum-metallised mirror depth cannot get it from paint-free pellets. Dark large-screen parts used outdoors under strong light will read as dirty under polarised lenses. Products that change colour often are also a poor fit, because the colour lives in the pellet and the mold: changing it means re-qualifying material.

Three types get a yes. Textured appliance housings, where the grain itself masks variation. White or light-grey office products, where fibre float is invisible. Products with EU environmental requirements, where the documentation burden is roughly halved.

The break-even

Paint-free pellets cost eight to twenty-five yuan per kilogram more than standard. Against that: a paint fixture costs two to six thousand yuan, paint and labour run three to twelve yuan per part, and paint defect rates of two to eight percent are normal. Above roughly twenty thousand units a month, paint-free usually wins on total cost. Below three thousand units a month, painting wins because fixture amortisation is small. The crossover sits somewhere between eight and twelve thousand units per month, and the only way to know for your part is to put all four line items into the per-unit number.

A snap-fit lives or dies on half a millimeter

Asher Hu — Thu, 05 Nov 2026 02:41:57 +0000

A snap-fit that feels loose, or one that cannot be pressed home, is almost never a design mistake in the shape sense. It is a tolerance-direction mistake. Three parameters decide the outcome: undercut depth, wall thickness of the beam, and insertion angle. Each has a working band, and the bands only work if the tolerances are assigned in the right direction.

Undercut depth sits between 0.5 and 1.5 mm. The important part is not the nominal value but the sign of the tolerance: only positive. A snap-fit is not a screw. If a screw comes loose you turn it another half-turn. If the undercut drifts negative in production, the interference is gone the moment the parts click together, and no downstream process can rescue it. Nominal values should therefore lean conservative, and the tolerance should be one-sided. A tight fit can be polished out of the mold. A loose one is scrap.

Beam ratio, not just thickness

The beam is a one-shot elastic element. PC and ABS allow roughly four to six percent strain. The longer and thinner the beam, the higher the strain at the root for the same undercut. The practical control is the ratio between root-to-load-point length and thickness: ten to one is safe, twelve to one is the edge, beyond that the part will whiten on the second disassembly and crack on the third.

Disassembly count belongs in the design input, not in a later conversation. A product that needs a serviceable battery compartment, designed with a single-use snap, cracked on the third inspection. The fix was not a thicker hook, which would have doubled the interference, but a two-stage snap: a long arm for travel and a short thick arm for locking. The mold only needed two inserts changed.

Insertion angle is a trade with the assembly line

A shallow insertion angle assembles easily but feels soft under a thumb. A steep one feels crisp but pushes assembly force up, sometimes past what a line operator or a fixture can deliver. The comfortable band is 25 to 35 degrees. Inside that band the choice is contextual: automated assembly takes the low end, user-serviceable parts take the high end.

A crisp click needs a defined impact point. Give the undercut face a 0.5 to 1 degree over-centre angle so the elastic return has something to snap against. On the mold this is a shut-off insert, so the machining cost barely changes, but it has to be called out on the drawing or it will not exist.

The five numbers

Undercut 0.5 to 1.5 mm, one-sided positive. Assembly interference 0.1 to 0.3 mm. Beam ratio at or under ten to one. Insertion angle 25 to 35 degrees. Draft at least one degree, including the inside of the arm. When all five hold, the snap is designed. When any one is missing, the failure shows up in the field.

Same leather, a crooked stitch line ruins the whole interior

Asher Hu — Wed, 04 Nov 2026 01:52:55 +0000

We have wrapped leather interiors for years. The look of a leather wrap depends 80% on flatness after stitching, not on the hide itself. Two workers can wrap the same imported hide and produce two price tiers. One leaves corners tight and stitch lines straight like a ruler. The other leaves a bubble on the seat center and a line off by half a millimeter with hard edges.

In our projects, stitching, tension, and wrinkling always fail together. Fix one parameter and the other two follow. We use concrete ranges from rework. Leather thickness of 1.1-1.3mm wraps most stable. Below 1.0mm the stitch tears. Above 1.5mm corners will not sit flat. Stitch pitch of 3.5-4.0mm and edge margin of 4.0-5.0mm give the best visual. Pitch below 3.0mm crowds holes on curves. We keep stretch under half the elongation rate. For leather at 40% elongation, we pull no more than 20%. Beyond that, the part wrinkles within two days of demold.

Stitching is not decoration, it is the force line that flattens the leather

Stitching first allocates tension. Good looks are a side effect. A thread passes through two holes. When pulled, it forms a crease channel on the leather. This channel draws surrounding material toward the center. Even draw yields a flat face. Uneven draw builds bulges on both sides of the channel. Pitch and margin are force parameters, not aesthetic ones.

We built a door panel. First we used 3.0mm pitch. The drawing looked dense and fine. The physical part exposed problems. At corners, holes clustered. The leather between holes became a narrow bridge. Pulling broke the bridge. Broken ends lifted small fuzz. We changed to 3.8mm pitch. The line immediately looked orderly. Corners needed no extra stitch reduction. Every 0.5mm tighter pitch drops the bridge strength one grade. This shows clearly on thin hides.

Edge margin acts the same. A 4mm margin seems far from the edge. It reserves material for folding and wrap. Push margin to 2.5mm and the stitch sits too close. During folding, the line shifts. The thread end gets pulled into the edge. The front shows a broken stitch. On a center console armrest project, we set 3mm margin. After installation, half the parts showed thread ends at corners.

How much tension to apply, and why it shows only after two days

Tension fails with delay. The part looks perfect on wrap day. After 48 hours it changes. Leather is elastic. We stretch it during wrap. The glue has not fully cured. Internal stress releases slowly. Two days later, the over-stretched section recoils. Places that cannot return build wrinkles.

We simplify the check to one line: no bulge when wrapped, no wrinkle after two days, no tight pull when installed. The second clause breaks most often. Our method sets stretch at half the elongation rate. We pre-stretch the hardest curved areas first. Large flat seat areas can take full tension. Corners and small radii must reduce tension. There the leather already extends to fit. More pull fights the material.

Another point is easy to miss: align stitch direction with main stretch. If the line runs perpendicular, it becomes a rigid constraint. It splits the hide into two tension zones. One side tight, one loose. Bulges appear along both sides of the stitch. Run the stitch along the main stretch. The leather then works as one stressed surface.

The location of wrinkles is determined at the drawing stage

Wrinkles are rarely wrapping mistakes. Most come from surfaces that forbid flat leather contact. Spherical and double-curved faces are typical. A flat hide on a sphere needs somewhere for excess material. It either wrinkles or stretches. Stretch cannot eat all of it. The rest piles up at some point.

We wrapped a seat side bolster. The shape had a continuous concave-to-convex section. Curvature changed fast. The first sample showed radial fine wrinkles at the inflection. Light revealed them even if subtle. The fix was not more glue or longer press. We enlarged the transition radius from R15 to R30. Curvature change slowed. The leather found its own level. The shape lost a bit of sharpness. That area looked far cleaner.

We follow one more rule: split-piece wrapping often beats whole-piece. Designers want one seamless hide. Whole wrapping forces every curvature change to be absorbed by one piece's stretch. Difficulty rises geometrically. Split into two parts and place the stitch at the boundary. The stitch becomes a design line. Many high-end interiors show a seemingly decorative stitch. It is actually a parting line.

Wrong leather choice means all later processes are remediation

Dimension	Top grain cowhide (1.2mm)	Microfiber leather (1.0mm)
Curvature fit	Good elongation, large radii fit naturally	Lower elongation, sharp corners need cutting or splicing
Stitch line strength	Dense fiber, stable at 3.5mm pitch	Use pitch above 4mm, tight pitch tears
Rebound wrinkle risk	Low, but uneven thickness may bulge locally	Higher, stress release after stretch more obvious
Visual texture	Natural grain, layered under light	Uniform texture, flat look up close

The "flat up close" row draws the most client feedback. Microfiber texture is pressed. It lacks the randomness of natural grain. Light reflects evenly. It looks clean but loses depth. For high-grade interiors we put top grain on main visual faces. We use microfiber on non-visible areas to control cost. Color difference between the two must be matched in the same batch sample. Otherwise the seam shows two colors.

We have shipped over 320 production items. Leather wrap rework never comes from hide quality. It comes from stitch density and tension not written in the drawing. The factory works by habit. Habit varies by worker. Batches then drift.

Three things to watch on trial fit day

First is lighting. Put the part under oblique light. Bulges and fine wrinkles invisible from front appear under side light. Our rule is a 30-degree side light pass. This angle matches interior car light conditions.

Second is edge feel. Slide fingers along both sides of the stitch. Feel whether groove depth is even. Depth difference signals uneven tension. Even if front looks fine, the shallow side will lift after months.

Third is batch comparison. Take two parts from same mold and same hide. Check color and grain direction. Leather has direction. If naps oppose, one light source shows two brightness levels. Joined, they look like different colored parts.

These checks need no instruments. Eyes and hands do it. But we must tell the factory the acceptance method before drawing finalization. Process stability is decided the moment the acceptance standard is written clearly.

FAQ

Does leather wrapping wrinkle because of too little glue?

Mostly no. The surface itself often forbids flat leather contact, leaving excess material nowhere to go. Adding glue only presses temporarily; stress release later builds bulge. Check wrinkle shape: radial means fast curvature change, local pile means over-tension, symmetric bulge along stitch means direction conflict. More glue fixes none of these.

How should stitch pitch and edge distance be set?

For interior wrap we use 3.5-4.0mm pitch and 4.0-5.0mm edge margin. This balances leather bridge strength and visual order. Thin hides and microfiber go to the upper limit; top grain can take the lower limit. Below 3.0mm pitch the bridge narrows and tears on curves; below 3mm margin the fold pulls the stitch off and exposes ends. Write both numbers in the drawing to avoid batch drift.

Why does it look fine on wrap day but wrinkle after two days?

Leather stretched during wrap holds internal stress. The glue is not fully cured, so the part looks good. About 48 hours later the stress releases. If stretch exceeded half the elongation rate, the material recoils and wrinkles. For 40% elongation hide, keep stretch under 20%. Corners and small radii are worst hit because they already use extension to fit.

Should we use whole-piece or split-piece wrapping, and how to choose?

Use whole-piece when curvature changes slowly and area is small; it looks cleaner. Use split-piece when the shape has continuous concave-convex turns or fast curvature change. The stitch falls on the boundary and becomes a design line. Insisting on seamless whole-piece forces one hide to absorb all curvature, multiplying difficulty and likely causing later splits. Decide the parting line at the modeling stage.

Sources: Compiled from Hezi Industrial Design's own project experience; no external citations.

Article by Hezi Industrial Design — Dongguan, China. Originally published at hezidesign.com.

Industrial Design Sketching: How Many Rounds Before a Mold Is Safe?

Asher Hu — Tue, 03 Nov 2026 04:03:56 +0000

Hezi Industrial Design — Dongguan, China. 320+ products shipped.

A client approves a beautiful sketch. Three weeks later the mold is cut, the prototype lands, and the thickness is wrong and the button can't be pressed. The sketch was the problem — not because it was ugly, but because it was drawn to impress instead of drawn to decide.

At our studio, product design sketching runs two to three rounds. Here is what each round is actually for, and the three checks a sketch must pass before we let it anywhere near a 3D model.

The round structure: each round has one job

Round	Output	Real purpose
Round 1	5–8 form directions	Elimination — confirm what the client does not want
Round 2	Narrowed to 2–3	Converge: one shared perspective and line language
Round 3	Detail fixes only	Buttons, ports, parting lines — no new ideas

Round one is not a beauty contest. The direction a client finally picks is rarely the one we liked most internally — it's the one closest to the vague picture in their head. Redrawing a sketch costs a fraction of rebuilding a 3D model, so the elimination work belongs here. Every direction must carry its main dimensions and its parting direction; a proposal missing either doesn't advance, because it will be redrawn anyway once it reaches engineering.

Paper or tablet? The structure team doesn't care

Structural engineers read three things: orthographic views, key dimensions, parting direction. The medium is irrelevant. What differs is cost and communication:

Paper: fastest for round-one divergence — seven or eight directions in one review. Scanned pages keep line weight and hesitation visible; remote clients can see which stroke was uncertain.
Tablet: better for round two — layers make color and overlay iteration cheap. But a clean digital sheet can falsely signal the design is already resolved.
Quick 3D rough model: slowest (half a day per form), but the only way to honestly judge volume. Use it after the form is basically settled.

For remote clients we'd rather send paper originals plus a voice note walking through each direction than a polished render. It produces better decisions.

The three checks before a sketch moves forward

Side profile at near 1:1. We control length-to-height proportion on a near-orthographic side view and refuse dramatic perspective. The harder the perspective, the faster the form collapses in 3D — client signs, model arrives, thickness is wrong, buttons don't fit, full rework.
The hand is in the drawing. For handheld products we sketch the grip holding it. Button positions are marked against an adult index fingertip — roughly 20 mm; a smaller contact area is hard to press accurately.
Parting direction marked in round one. A smooth sweeping surface with no draft angle or ejector-pin space is the most expensive mistake to fix later. This one is written from 320+ shipped products.

When the client rejects three rounds in a row

It's usually not the sketching. At least one of three inputs was never locked: target user, retail price band, or benchmark competitor. The missing one keeps resurfacing as "it just feels off." We once had two rounds rejected, asked about retail price, and the answer was "not calculated yet" — once the price band landed, most of the facet details were scrapped. That rework is free of charge but eats the schedule: a 7–15 working-day appearance phase becomes a month.

When to stop sketching

Not when the client nods. When a structural engineer can pull orthographic views straight from the sheet. If they can't, keep drawing — don't rush into 3D. Sketching is the cheapest round in any engagement tier (appearance 5,000–12,000 RMB; with structure 15,000–40,000; full case 30,000–80,000), so it deserves the extra two days.

Originally published on the Hezi Industrial Design website. Hezi is an industrial design studio in Dongguan specializing in product appearance, structural, and CMF design with 320+ products in mass production.

Three color-matching rounds: how a product color survives to mass production

Asher Hu — Tue, 03 Nov 2026 01:20:14 +0000

A color approved on a paper chip is not a color yet. It is an intention. Between the chip and the molded part, the light path changes, the carrier changes, and the surface changes, and each change pulls the hue somewhere else. We split color approval into three separate rounds: chip matching, prototype matching, and first-article matching. Each round has a different target and a different pass criterion. Skip one, and the outcome is predictable: the sample looks expensive, the shipment looks like a market stall.

The cost of skipping is not a conversation, it is a re-run. In our projects, almost every color rework traces back to the one round nobody scheduled. A wrong masterbatch means scrapping pellets. A wrong surface finish means re-texturing a mold. Both are measured in weeks.

Three rounds, three different sets of numbers

Round one looks at hue direction and the lightness band. We compare against a Pantone chip under standard light and hold the deviation to ΔE ≤ 1.5.

Round two looks at the substrate itself. Plastic pellets, anodized aluminum, and sprayed paint film render the same code very differently. One code can land 2 to 3 lightness steps apart on PC and on an aluminum part.

Round three looks at batch stability. After the first article is signed off, we keep sampling: ΔE ≤ 1.0 across cavities within one batch, relaxed to ΔE ≤ 2.0 across batches.

Those three numbers are our internal acceptance lines. If the customer contract does not state them, we raise them ourselves.

Round	Main risk	What we do
Chip matching	Carrier difference shifts hue	Shoot a plaque in the real production material, not on paper
Prototype matching	Surface finish and gloss do not match production	Align texture and gloss with the production plan before comparing
First-article matching	Batch drift and cavity-to-cavity variation	Sample multiple cavities, retain parts per batch for comparison

Why that black turns blue on a plastic part

A color chip is ink printed on paper. A plastic part is pigment dispersed in resin, and light passes through it before coming back to the eye. Paper is a reflective carrier: light hits the surface and bounces straight back. Plastic, especially in light or translucent grades, lets light in and out again, and that round trip drags the hue toward the cool side.

We ran into this on a small desktop appliance. The customer specified a deep gray-black close to Pantone 419C. The first prototype was molded in PC with masterbatch, and next to the chip it was visibly blue. The shop floor suspected too much masterbatch. We measured the ratio and it was correct. The problem was the material: PC has a higher refractive index than the coating on the chip paper, so short-wavelength blue scatters more, and the eye reads blue.

There are two ways out. Change the material, for example PC to PC+ABS, which lowers the refractive index and pulls the hue back toward neutral. Or adjust the formula, adding a small warm compensation to the masterbatch to push the blue down. We chose the second, because the structural parts were already dimensioned around PC shrinkage, and a material change would have meant re-checking every dimension. The compensation window is narrow. At 0.3 percent warm masterbatch the parts went red. At 0.15 percent they matched. That figure does not transfer to another project, because masterbatch concentration varies by supplier.

So the correct first round is a plaque in the material you will actually run, not a chip against a chip. Injection molders can shoot an A4-sized plaque cheaply. Many projects skip it, tune the paint against the chip, and only discover the problem when the prototype arrives.

The prototype stage is where we fool ourselves

A prototype is a single piece, and its surface treatment is far from the production part. We have seen many projects where the painted prototype looked beautiful, the customer signed, and production switched to natural resin or in-mold decoration. The color was a different object entirely.

Three things matter at prototype stage. The first is surface process. A painted prototype and a textured molded prototype read half a step apart at the same code, because texture changes the diffuse reflection angle. The coarser the texture, the lighter the color looks. We require the prototype surface to match the production plan. If that is impossible, compare on a plaque with the same texture.

The second is gloss. Matte and gloss move color more than most people expect. The same can of paint reads grayer in matte and deeper in gloss. When a customer says they want a more premium matte black, the follow-up question is how matte: 3 gloss units or 5. That number goes into the CMF specification. The word matte does not.

The third is viewing condition, and it is the most overlooked. A prototype that looks correct under warm office light turns yellow by the window. Color must be judged in a standard light booth, or at least under 5000-6500K near-daylight. If the customer has no booth, we carry a portable D65 tube to the meeting. That habit came out of a rework: a product read off-white in the customer conference room and turned cream-yellow on a mall display stand, because the stand spotlights ran at 3000K.

First-article approval is a starting point, not a finish line

Injection molding color drifts. It is not constant. The same machine and the same lot of material can produce visibly different parts in the morning and the afternoon, because of barrel temperature drift, uneven masterbatch mixing, and changes in regrind ratio. On one job the first article was exactly what the customer wanted. By part 3,000 the color had gone noticeably darker. The cause was regrind, raised from 10 percent to 25 percent.

So first-article approval is the beginning. We ask the factory to retain samples by batch, 3 to 5 parts each, and to compare them side by side with the approved first article under standard light. For high-volume programs we also require colorimeter readings with the ΔE values logged. Those logs are worth a lot later, when a complaint arrives and we need to say which batch started to drift.

Cavity-to-cavity variation is the other thing that gets missed. In a multi-cavity mold, fill speed and shear heat differ between cavities, so color differs slightly. If two appearance parts sit next to each other and one comes from cavity 1 and the other from cavity 8, the seam shows the difference, most obviously on light colors. We pair appearance parts from the same cavity where possible, or we move the visible faces toward the center cavities at mold design stage.

What actually belongs in a color specification

Write what can be executed, not adjectives. A workable color specification carries five items: the color code reference, the carrier material grade, the surface process, the gloss value, and the acceptance ΔE range. Leave one out and the factory has room to interpret.

We have seen a specification that said only deep space gray, matte. The factory delivered something purple. The customer rejected it. The factory said, that is deep space gray. There is no productive argument at that point, because the specification had no measurable anchor. The correct version reads: code per Pantone or RAL, specific material grade, texture number or coating type, gloss as a number, and a ΔE ceiling.

Add boundary conditions too. State which direction the color will shift under indoor warm light and which direction under outdoor sun, so the customer knows in advance. This is not shifting blame, it is managing expectation. On projects where the customer approved a demo unit in the showroom and then disliked it in the real environment, having the light-source note in the original specification made the conversation much shorter.

Color work is really the management of three information losses: chip to material, material to prototype, prototype to production. Each loss has a physical cause, and none of them means someone was careless. Run all three rounds, write the measurable parameters into the specification, and the color holds. Across the 300-plus mass-produced products we have worked on, the projects with color problems almost always trace back to one round that was skipped.

FAQ

When should product color be fixed?

Fix the color direction once the form is settled and before structural detailing begins. Color drives the choice of surface process, surface process drives mold texture, and texture affects fit clearances on structural parts. Late changes cascade. The cheapest moment for plaque comparison is right after the form review. Once the drawings are done, a material or finish change pulls dimensions and tooling with it.

The customer gives us only one color chip. Can we judge feasibility from it?

Not directly. Shoot a plaque in the actual production material first. The chip is printed ink, and plastic, metal, and paint render the same code with different hue shifts. We ask the molder or the coater for an A4 plaque, compare it with the chip under standard light, and confirm the direction and size of the deviation. Then we decide between formula compensation and a material change. The plaque costs little and prevents a full-batch rework.

How much do matte and gloss change the perceived color?

More than most teams estimate. The same can of paint reads half a lightness step apart between matte and gloss. A matte surface scatters more diffusely and looks lighter and grayer. A gloss surface reflects more specularly and looks deeper and more saturated. Write a gloss number, 3 units or 5, into the specification. Match the prototype surface to production, or the comparison tells you nothing.

What is the minimum content of a color specification?

Five items: color code reference, carrier material grade, surface process type, gloss value, and acceptance ΔE range. Miss any one and the factory can interpret it their own way, which raises the odds of a dispute. Beyond the five, add a note on boundary conditions, such as how the color shifts under warm and cool light, so the customer has an expectation before the product reaches its real environment.

Sources: Compiled from Hezi Industrial Design's own project experience; no external citations.

Article by Hezi Industrial Design — Dongguan, China. Originally published at hezidesign.com.

Same 3D file, new resin, cracked corners: digest shrinkage into the drawing

Asher Hu — Mon, 02 Nov 2026 00:31:48 +0000

A customer sent us a 3D file from a mold that had already run. The only request was to swap PC/ABS for ABS. We released the model as it was. At T0, all four corners of the housing had cracked, and the snap fits broke on the first latch.

Wall thickness was not the problem. Ribs were not the problem. Every dimension in that file had been scaled for the old resin's shrinkage. Changing the resin without changing the shrinkage is walking a new road with an old map. The mold was cut wrong, the parts came out wrong, and the whole T0 cycle had to be paid for twice.

The fix is one action, applied everywhere: rescale every dimension on the 3D model to the new resin's shrink rate, and let the model carry that value out to the mold shop. It is not a line in the BOM.

Shrinkage is not a number in the corner of the drawing

Most structural engineers open a drawing, find the technical requirements box in the lower right, and change "shrinkage 1.005" to "1.006". That is not wrong. It is also not enough. The shrink rate is the basis for every cavity dimension in the tool. The cavity is cut oversized by that factor. Change the factor and the cavity changes. If the 3D model was never rescaled, the mold shop machines to the old geometry and the parts come out at the wrong size no matter what the box says.

The numbers look harmless until you put a length on them. PC/ABS runs 0.4% to 0.6%. ABS runs 0.4% to 0.7%. On paper the gap is 0.1%. On a 200 mm housing, 0.1% is 0.2 mm. That is enough to break assembly at snap clearance, at boss center distance, and at the shutoff fit.

Material	Shrinkage	What it does to the drawing
PC/ABS	0.4% to 0.6%	Baseline for the original file
ABS	0.4% to 0.7%	Wider range, batch spread can reach 0.1%
ABS + 10% glass fiber	0.2% to 0.4%	Strong anisotropy between flow and cross-flow
PA, POM	1.5% to 2.5%	PA keeps shrinking after ejection, 0.1% to 0.3%

There is a second layer. Different resins distribute shrinkage differently through the wall and along length versus width. ABS is more anisotropic than PC/ABS, so the stress concentration points at wall thickness steps move as well. A drawing scaled uniformly for the old resin puts those points in the wrong place.

We had a project where the customer moved from PC/ABS to ABS with 10% glass fiber. Glass-filled shrinkage is 0.2% to 0.4%, roughly half of plain ABS. The drawing was never revised. The housing came out 0.5 mm short, the snap positions no longer lined up with the front cover, and the entire mid-frame was scrap. That 0.5 mm was not a tolerance issue. Every dimension was small in the same direction. Systematic offset, not random variation.

The drawing shows the finished part, not the mold

A 3D file describes the shape and size of the molded part. It does not describe the cavity. The mold shop takes the file, multiplies each dimension by the shrink rate, and cuts the steel. Give them the wrong rate and they apply the wrong multiplier. The logic is simple, yet most material changes stop at renaming the resin in the BOM.

The correct move is to rescale the whole part inside the 3D software. If the original file carries 1.005 and the new resin needs 1.007, scale the model by 1.007 / 1.005, about 1.002. This is not editing a text field. It means resetting the scale factor inside the modeling feature tree so the exported model itself carries the right value. Different CAD packages put the command in different menus. The principle does not change.

We got this wrong once. We updated the technical requirements box and left the model alone. The mold shop built the cavity to the box, but their programmer read dimensions straight off the 3D model and never opened the note. The cavity was cut at the old shrinkage. The part came out 0.3 mm small, and the snap clearance opened from 0.1 mm to 0.4 mm. The latch would not hold.

Three questions before you change material

When a material change request lands, do not start editing drawings. Ask three questions first.

What is the actual shrinkage? Get measured data from the supplier, not the range on the datasheet. ABS is listed at 0.4% to 0.7%, and batch-to-batch variation alone can reach 0.1%. The same grade from a different lot shrinks differently. Ask for the measured value for the lot, or shoot a sample from the pellets and measure it.

How anisotropic is it? Glass-filled grades shrink far less in the flow direction than across it. On a rectangular housing with the long edge along flow, the long and short edges need different compensation. A single global scale factor only covers one direction. Anisotropy has to be compensated by the mold shop during cavity machining, and that instruction has to be on the drawing.

How much post-shrinkage? Some resins keep shrinking after ejection. PA can move another 0.1% to 0.3% and only stabilizes after 24 hours. On a precision fit, that movement opens the gap. Leave allowance on the drawing or build the correction into the tool.

Shutoffs and snap fits are the most sensitive features

The places where shrinkage difference bites hardest are the shutoff and the snap fit. Shutoff clearance is typically 0.05 mm to 0.15 mm. A 0.1% shrink difference across a 100 mm shutoff length is 0.1 mm, which can consume the whole clearance. A straight shutoff can still be saved by reworking the tool. A tapered shutoff gives you very little rework range.

Snap fits are worse. The deflection available on a snap is usually only 0.3 mm to 0.5 mm. A 0.2% shrink difference across a 50 mm snap arm is 0.1 mm. The arm has almost no margin to absorb that. We have seen a material change make a snap too tight, and the latch snapped off during assembly. Teardown showed the arm had come out 0.15 mm shorter than expected, which pushed engagement depth up by 0.15 mm and drove snap stress past its limit.

So after any material change, check shutoff and snap dimensions separately. If the shrink rate moves by more than 0.1%, recalculate the fit clearance and adjust arm length or shutoff depth if needed.

A material change is not a renaming exercise. Every dimension in the drawing was designed around one specific shrink rate. Change the resin and the model has to be scaled again. Do the work up front. Fixing cracks after T0 costs double in tooling and in schedule.

FAQ

Where do shrinkage differences actually show up?

Three places: the shrink value itself, anisotropy, and post-shrinkage. Crystalline resins such as PA and POM shrink heavily, around 1.5% to 2.5%. Amorphous resins such as ABS and PC shrink little, around 0.4% to 0.7%. Glass-filled grades add strong anisotropy, with flow and cross-flow shrinkage sometimes differing by a factor of two. Matching the headline number is not enough. Match the material class too. Moving from ABS to PC/ABS keeps the number close, but PC post-shrinks more than ABS. On precision fits, mark the drawing "scale to measured shrinkage" and attach the supplier's measured data.

I changed the shrinkage in the 3D file. Can the mold shop still get it wrong?

Yes, if only the technical requirements box was edited and the model itself was not rescaled. Programmers usually read dimensions directly from the 3D model. If the model still carries the old shrink rate, they program the old size. A written note can be missed. The reliable route is to reset the scale factor in the CAD features so the exported model carries the correct value, then print the shrink rate in a visible spot next to the dimension and tolerance table with the note "this shrinkage is already applied in the 3D model". Confirm the number with the mold shop in writing when you release the files.

Which features fail first after a material change?

Shutoff fits, snap fits, boss center distance, and the corners of long housings. Shutoff and snap clearances are typically 0.05 mm to 0.15 mm, so a 0.1% shrink difference can erase the gap or double it. Corners on long housings concentrate stress, and a shrinkage mismatch raises local internal stress until cracks appear. Boss center distance affects screw assembly: two bosses 100 mm apart shift 0.2 mm at a 0.2% shrink difference, and a screw will skew in a blind hole. Run a separate dimension chain calculation on long dimensions, mating faces, and stress concentration points to confirm every clearance still sits inside tolerance.

How do I get accurate shrinkage data?

Ask the supplier for the measured shrinkage of the specific lot, or mold a sample from that lot and measure it yourself. Datasheet values are ranges, and real batch variation reaches 0.1%. For tight-tolerance parts, measured data beats the datasheet every time. To measure, injection mold a simple bar 200 mm long, 20 mm wide, 2 mm thick, measure its length, and calculate shrinkage as (cavity length minus bar length) divided by cavity length, times 100%. Measure 24 hours after ejection so post-shrinkage has fully released.

Sources: Compiled from Hezi Industrial Design's own project experience; no external citations.

Article by Hezi Industrial Design — Dongguan, China. Originally published at hezidesign.com.

Your prototype feels wrong: measure three ratios before you touch the shape

Asher Hu — Sun, 01 Nov 2026 02:42:23 +0000

A hand sample lands on the desk and the room goes quiet. It feels clumsy, it feels cheap, it looks nothing like the render that everyone approved six weeks earlier. In our experience nine out of ten of those meetings are about the wrong layer. The surfacing is fine. The proportions collapsed before anyone started modelling. When we measure, the body length-to-width ratio is off by 0.3, or the head takes 5 percent more of the volume than the sketch said, or the waistline sits outside the visual centre of gravity. You can redraw the fillets ten times. If the ratios are not back in place, the tenth version will still read as the same product with the same wrongness.

Measure three ratios first, then talk about surfaces

Our triage order is fixed. Measure the body length-width-height ratio, the volume share of the head against the body, and the waistline position as a percentage of total height. Then, and only then, discuss curvature and chamfers.

For handheld products we normally lock length-to-width between 2.2 and 2.6. Push past 2.8 and the product reads as long and thin, and the palm sits empty when you grip it. The waistline usually sits between 45 percent and 55 percent of overall height. Step outside that band and the visual weight drifts up or down, and the hand sample immediately looks top-heavy or bottom-heavy. These three numbers belong on the sketch sheet, not in a conversation held after the sample comes back from the shop.

Why the render looks good and the hand sample looks wrong

The difference is perspective. A render speaks in 35mm to 50mm lens language. A hand sample is judged by the human eye at roughly 40cm. The two do not feel the same machine the same way. A long lens stretches slender objects; a sample viewed at a 30 degree downward angle on a desk gets its length-to-width ratio flattened.

We built a desktop fan where the render pleased everyone. The first T0 sample came back and the first reaction was: why is it so stocky? Calipers said the body was 148 by 92mm, a ratio of 1.61. In the render, perspective made it read at about 1.75. That 0.14 gap was the whole difference between light and heavy.

We now have a rule. Every render release ships with an orthographic three-view built from the same revision of the CAD. Reviewers look at both, side by side. If the three-view looks worse than the render, you have been fooled by perspective, and the proportions get corrected against the three-view. This single rule has saved more rework than any rendering trick. Across the 200-plus clients we have worked with, misjudged proportions from render perspective are a common cause of a second round of samples.

One more point that gets missed: handheld products must be judged at real grip distance. A phone at 30cm from the eye and the same phone at 80cm on a desk give completely different readings of screen-to-body proportion. For handheld work we put a 30cm ruler on the review table and ask reviewers to stand at that distance. It simulates the real use angle instead of the presentation angle.

What body ratio keeps a product from looking cheap

There is no universal number, but there are bands and a way to choose between them. Cheapness usually comes from a ratio that sits nowhere: not full enough to feel solid, not slender enough to feel precise, stuck in the middle like a size someone nudged by accident.

We set the category character first, then the narrow edge. Tool-like products run 1.5 to 1.8. Tech-leaning products run 1.9 to 2.3. Grip-driven products run 2.2 to 2.6. Once the ratio is chosen, calculate the absolute narrow edge. On small devices where the narrow edge falls below 18mm, keep the ratio under 2.4. Otherwise the side view reads like a blade and the hand gets no sense of volume.

Last month we worked on a desktop charging dock. The client asked for weight. The original design was 2.1 with a 42mm narrow edge, and it looked light. We did not add a ballast block. We widened the narrow edge to 48mm, pulled the ratio down to 1.85, and tightened the bottom chamfer from R3 to R1.5. The client picked up the sample and said: now it sits. Perceived mass comes from the narrow edge, not from grams of metal. That is a form decision, not a tooling decision.

One counterintuitive rule: once the ratio is locked, do not let it drift during local tweaks. We learned this the hard way. Late in a project we widened the body by 1.2mm to fit a button. The ratio moved from 2.25 to 2.19, and the whole product read as slightly fat. We rebuilt the sample. The ratio is a global parameter. Every local change has to be checked back against it.

Why a 3mm waistline shift makes a product look top-heavy

The waistline decides where visual weight lands. People read an object's centre of gravity largely from the position of the split line across its middle. Move the line up and the upper half carries the mass while the lower half looks empty. Move it down and the product reads as heavy and dull.

We measured waistline positions across a batch of handheld devices. The 45 to 55 percent band held up as the most stable. That is not aesthetics, it is testing. We taped waistlines at 40, 50 and 60 percent on identical samples, lined them up, and asked people to pick the one that looked most stable without knowing the numbers. The 50 percent sample won almost every time.

A fascia gun is a concrete case. The first version had its waistline at 62 percent. In hand it visibly floated upward, the grip zone looked narrower than it was, and users instinctively did not dare to hold it firmly. The second version dropped the line to 48 percent and added a 0.6mm inward transition on the parting line below it. The product settled immediately. The silhouette never changed. Only that one line moved.

The waistline also ties into the parting line. A clean line on the outside has a cost on the inside, because the mould has to open somewhere. We mark the parting position on the render during the appearance review so the client can see what that line will become later. It is cheaper than arguing about it once the structural drawing exists.

Which revision should make you stop and doubt the proportions

Our internal rule: when the same area reaches its third revision and still feels wrong, stop and measure the ratios. The third version is a psychological threshold. By then the designer is committed and will look for problems in the details, while the problem has been one level up the whole time.

Three tests tell you whether proportions are the culprit. Shrink the product to finger size. If the overall character still falls apart, the problem is proportion, not detail. Then paint the shell pure white in a single colour and look again. Strip out the CMF. If the outline no longer states a clear shape intent, the proportions were never set. The third test is the most brutal and the most useful: trace the outline as a single line and look at the silhouette alone. If the silhouette does not look like something that could stand up on its own, no amount of detail work will save it.

We were five versions deep into a smart speaker and still felt it lacked something. We overlaid all five side silhouettes on one sheet. Every version had changed its top arc, while the bottom kept growing thicker. The two ends had never been aligned. We went back to the ratio framework from version one, adjusted three key dimensions, and version six passed.

Symptom	Layer it points to	What to do
Still falls apart at finger size	Global proportion	Reset the aspect ratio and waistline
Outline reads nothing in pure white	Silhouette and hierarchy	Tighten the body, raise contrast
One local area looks off	Detail layer	Change chamfers, transitions, CMF

Once the ratios are back in place, half the details you wanted to change stop needing changes. That is why we run the ratio check before any shape revision. Reset first, then edit. It is several times faster than editing details while guessing where the real problem sits. We have taken more than 320 products to production, and this is the lesson that came out of redoing a lot of hand samples.

FAQ

At what stage should the ratios be locked?
At the sketch stage, and as three numbers: body aspect ratio, head volume share, waistline height as a percentage of total height. Changes cost almost nothing on paper. Once modelling starts, every surface, chamfer and button position attached to the body has to be recalculated. We write the three numbers next to the concept and check the numbers before looking at the drawing.

How do we tell a proportion problem from a detail problem?
Shrink the product to finger size, or paint it pure white and read only the silhouette. If the character still falls apart at small scale, the fault is in the global proportion. Overlaying side outlines from several versions is faster still. If each version keeps changing a different area, the top-level framework was never fixed and debating fillet radii is pointless.

Why does a good render turn into a clumsy prototype?
Render perspective flatters the product. The camera stretches long, slender objects, while the eye judging a sample on a desk is much closer to an orthographic view, which flattens the length-to-width ratio and increases perceived bulk. The fix is mandatory: every render ships with an orthographic three-view from the same CAD revision, and the proportions are corrected against the three-view.

Is there a recommended aspect ratio?
No universal value, only ranges chosen by category character. Tool-like products run 1.5 to 1.8, tech-leaning products 1.9 to 2.3, grip-driven products 2.2 to 2.6. Judge the narrow edge at the same time. Below an 18mm narrow edge on a small device, keep the ratio under 2.4 or the side view reads like a blade. Widen the narrow edge and the same ratio will look more slender, so the two parameters are calculated together.

Sources: Compiled from Hezi Industrial Design's own project experience; no external citations.

Article by Hezi Industrial Design — Dongguan, China. Originally published at hezidesign.com.

What a cordless vacuum teardown shows about CMF specifications

Asher Hu — Sat, 31 Oct 2026 02:44:08 +0000

A CMF document is a factory instruction, not a colour choice. We took apart a cordless vacuum on the bench and counted: 27 parts, 19 of them appearance parts, produced by at least 5 different processes. That ratio is what our CMF work looks like every day.

Every one of those 19 parts needs its own written record. Leave one column blank and the injection moulder and the paint shop will each fill it with their own assumption. The result is a tray of parts that are individually correct and collectively wrong. The vacuum looks cheap without anyone being able to say which part is at fault.

One appearance part, five fields

A typical finished product carries 10 to 20 appearance parts. For each one, our specification sheet writes down at least five fields: substrate grade, colour code, surface process, texture number, and the colour-matching condition with its tolerance. Miss any field and you have handed the decision to the supplier.

Colour tolerance is measured as ΔE in CIE L*a*b*. Parts from the same process in the same batch are held to ΔE ≤ 1.0. Adjacent parts from different processes, a painted shell sitting next to a textured one for example, are usually relaxed to ΔE ≤ 1.5 to 2.0. Different processes colour by different mechanisms, so forcing one number on all of them makes the target unbuildable.

Anodised dyed aluminium is the hardest class to control on a whole machine. Shade variation between dye tanks is easy to see by eye under D65. The industry norm is ΔE ≤ 1.5 with a same-batch, same-tank requirement written into the spec.

Texture is specified by VDI 3400 number or by the texturing shop's own sample number. Always pair it with a gloss value in GU. A texture number alone lets two texturing shops deliver two completely different hand feels.

Three origins, three ways to write the spec

The parts inside this machine fall into three families, and each family needs a different CMF entry.

Appearance part	Substrate and process	Result	Consistency risk
Metal wand	Aluminium extrusion plus anodised dye	Real metal sheen, cool and hard	Shade variation between tanks in one batch
Dust cup	Clear PC plus hard coat	Transparent, the most expensive-looking part	Hard coat hardness fights haze
Main housing	PC+ABS textured in-mould, or primer plus colour plus clear	Matte, fine, resists fingerprints	Tolerance bands differ by process

The metal wand starts as an extruded aluminium profile, then gets anodised and dyed. The oxide layer is clear. Dye enters the pores and is sealed in. The advantage is that the metal look is genuine: light reflects with direction. No paint on plastic reproduces it. The limit is film thickness. Too thin and the colour is pale with poor wear resistance. Too thick and the fine extrusion lines get magnified, so a crisp highlight band turns muddy. Consumer electronics usually sit in the low tens of microns. Push past that and depth and uniformity both get hard to hold.

The dust cup is a clear part with a hard coat. The body is clear PC or a copolyester, and a hard layer goes on the outside to resist scratching. It does one job for the design: it makes the internals visible, which reads as precision, and it is a big part of why the machine looks expensive. The cost is that clear parts demand one grade more discipline from the tool and the room. Resin drying, mould temperature and airborne dust all become visible defects. Hardness and haze also trade against each other. Coat it harder and it hazes, and the transparency is gone.

The main housing is plastic with surface treatment, on one of two routes. Route one is texture-in-mould with no paint: the pattern comes from the tool and the colour comes from masterbatch at injection. Route two is painting: mould a gloss part, then primer, colour coat and clear coat. The difference in feel is obvious. A textured part has a fine grain under the finger and reads as one piece. A painted part can achieve a more even matte or gloss, but it adds three processes and three more yield gates.

A colour chip becomes three problems in production

When a client points at a chip and says "that black", the sentence splits into three questions on the factory floor.

First, one colour code does not produce one appearance across processes. Paint can be tuned accurately. A textured part carries masterbatch inside the plastic and its colour is further shifted by the texture surface. Anodising puts dye into pores. Three mechanisms, three results. We hit this on a small appliance: we sent the same Pantone code to a paint shop and an IMD supplier, and the T1 samples were visibly two different colours. Nobody had done it wrong. We had written one tolerance number where we needed three. Since then our specs give tolerances by process band, tight within a process, wider across processes with a signed-off sample.

Second, a spec with no viewing condition is not a spec. The same part can read as three colours under D65, TL84 and illuminant A, especially metallic sparkle finishes and sensitive dark greys. The document must state the light source, its power and the viewing angle. Otherwise the sample looks one way in the client's office and another under the yellow lamps in the workshop.

Third, decide where part boundaries go. On this machine the top cover and the handle shell are two parts in two processes, but the eye reads one colour zone. That is where colour error shows first. Our practice is to pull visually continuous areas into one part or one process. If that is impossible, put a step or a colour break on the split line, so the mismatch sits behind a designed edge instead of spread across a flat face.

The texture column is the one most often under-written

Texture looks like "just a pattern", so it gets written carelessly. It actually sets three things at once: gloss, hand feel, and how dirty the machine looks after six months.

Texture depth and draft angle are locked together. Deeper texture needs a larger draft. A larger draft puts a sloped wall in the pull direction, and the highlight band deforms across it. On our first version we drew side walls at gloss-part angles. The texturing shop came back and said that depth was not achievable at that angle, so either the depth came down or the angle went up. We split the difference, and the matte level came out slightly flatter than the render. The lesson: confirm the texture sample before the tool is cut. Depth and draft are decided together, and later changes mean cutting steel.

Gloss is the field that should be there and often is not. The same texture number, polished brighter at the shop, comes out semi-matte rather than fully matte. We specify a GU range and require the supplier to shoot a sample inside it. The visual difference is immediate. Drop one step in matte and the surface goes from a soft scatter to a thin sheen. The plastic look comes back, and the clean highlight line is eaten by a hazy reflection.

Why the whole list matters

Fill every column and the specification holds. The teardown gave us a plain conclusion. This machine does not look expensive because one part uses an exotic material. It looks expensive because metal, clear and matte plastic each carry a distinct role, the boundaries between them are clean, and the colours do not fight. That is the job: define each appearance part so it can be produced, repeated, and matched to its neighbours, so the machine on the line is the machine in the render.

FAQ

Is CMF colour work the same as graphic design colour work?
No. CMF has to answer whether a colour can be produced reliably on a real material with a real process. Colour is one field among five. The same code reads differently on paint, on texture-in-mould and on anodising, and the CMF job is to price those differences into the spec before tooling starts.

What is the minimum content of a CMF specification?
Five fields per appearance part: substrate grade, colour code, surface process, texture number with a gloss range, and viewing source with colour tolerance. Tolerances should be banded by process, tight within a process and wider between adjacent processes, with a signed sample as the final reference.

Texture-in-mould or painting, how do we choose?
Decide by the surface you want and the volume you can commit. Texture-in-mould removes three processes and three yield gates and reads as one integrated part. Painting gives a more even matte or gloss, and you pay for it in process cost and appearance-defect risk. Check where the split lines fall first, because texture depth and draft angle are set together.

What do we do when metal and plastic will not match?
Solve it in the part split first. Pull visually continuous zones into one process where you can. Where you cannot, hide the difference behind a colour break or a step, then close it out with a matching sample. Anodised parts always carry tank-to-tank shade variation, so write a same-batch, same-tank requirement and inspect under the specified source.

Sources: Compiled from Hezi Industrial Design's own project experience; no external citations.

Article by Hezi Industrial Design — Dongguan, China. Originally published at hezidesign.com.