Make a Product-Transformation Test: Morph the Geometry or Conceal a Swap?
Make a Product-Transformation Test: Morph the Geometry or Conceal a Swap?
Two frames can be identical and describe two completely different events. In one, a single object has passed through every shape between them. In the other, the first object left the frame and a second one arrived while nobody was looking. Show only the before and the after, and neither the client nor the editor can tell which treatment they're approving. The middle is where the claim lives.
So the working answer is this: choose the morph when the transformation itself has to stay visible and the geometry can actually support the states between the endpoints. Choose the concealed swap when the two ends of the product need to be independently excellent and continuity matters less than how each endpoint looks. Neither is automatically the more convincing result. They make different promises — one about continuity, one about arrival — and each promise carries a price you should be able to name before somebody else names it for you.
What follows is a design walkthrough of both routes built on one hypothetical product, using the standard weighted-morph workflow as the constructive reference. It is a paper comparison. No model was built, deformed, rendered, or exported for it, so read the failure modes as what the construction predicts rather than what a test showed.
The middle is the evidence
A still frame is a claim about a shape at an instant. It cannot tell you whether that shape arrived by movement or by substitution, because at that instant there is nothing to distinguish. The distinguishing frames are the ones nobody puts in the deck: the halfway point of the morph, and the two frames on either side of the swap's cut.
Each route fails in its own idiom. The morph fails visibly and early — a crease, a pinch, a surface that thins where two movements of the same vertices meet. The swap fails at one seam, and it fails quietly, because everything the audience learns about the identity of the object has to be supplied by them in the half-second after the reveal. One failure is in the model sheet. The other is in the edit, the match, and the lighting. Before you choose, decide which of those you'd rather own at three in the morning.
Decide what has to survive
Write the requirement before opening a single application, because a creative requirement and the appeal of a software button feel almost identical while you're at it.
Four things usually have to survive a product transformation: the silhouette at the moment the audience recognizes the object, the feature that identifies it as this product, whatever relationship the label has to the form, and the sense of what the thing is made of. The soft block in this walkthrough is a rounded cube of matte silicone with a debossed wordmark on the front face and a molded seam down each side. The handled form is a compact pouring shape from the same material family with a single handle taking off from the right side and a narrower neck. Same family, different object.
Now the question that splits the two routes. Does the audience need to witness continuous change, or only connect two endpoints? If the commercial's argument is that the material does something, the change has to be seen happening. If the argument is that one product stands in place of another, seeing the mechanics is optional and possibly a distraction.
Say the convention out loud either way. An animated transformation is a visual convention, not a demonstration of physical capability. If a soft block becomes a handled form on screen, that is a designed event, and the treatment should say so, in plain language, near the storyboard. Silently letting a morph imply that the real product deforms this way is a claim nobody signed off on.
The handle has to exist before the transformation starts
Here is the part that surprises people who haven't built one. In a target-based morph, the target shares the base's vertex structure. The mechanism moves points; it does not add them. So the handle cannot be created at frame 40. It has to be sitting in the block you approve, as a flat strip of extra faces folded against the right side, waiting for its weights to come up.
That has consequences that reach backward into the rest of the shoot. A zero-thickness strip pulled outward gives you a ribbon — a paper handle with an edge you could cut yourself on. To get a handle with volume, the thickness has to be in the base too: either the block is a little plumper along that side, or the handle's two faces sit coincident against the body and are driven apart as the weights rise. Both are legitimate. Both mean the "neutral" block is not neutral. If an earlier shot needs a plain squishy cube, the reserved strip has to hide — shaded as part of the side, turned away from camera, or covered by the hand that's about to pick it up.
The construction itself is not exotic. Maya's help documentation describes a route that duplicates a base mesh, edits the target components, creates a blend-shape deformer, and keys the target's influence; the same procedure appears in the 2017 pages and again in the 2026 page covering the workflow, with predictable deformation depending on compatible geometry and later topology changes handled separately. Target edits are stored as differences from the base, and isolating a target lets you see what one shape contributes on its own. That last operation is worth adopting as a habit, because the contribution of a single target in the middle of a stack is exactly where unexpected geometry comes from.
Compatibility, though, is not the same as correspondence, and correspondence is the actual craft decision. Any two shapes can be said to correspond; a vertex here is called the same vertex as a point over there, by index, and the deformer interpolates between them. If the point on the right side of the block is designated the same point as the tip of the handle, the middle shows that point travelling — out of the body, through the side, into the handle. If it's designated the same point as part of the neck, the handle forms from material that was never near where the handle finally sits. The number of frames and the curve of the easing have almost nothing to do with whether the middle reads well. The pairing does.
So inspect those middles on purpose. A contact sheet at ten-percent steps, not five frames around the end. What you're looking for:
- A ridge where the reserved strip lifts off the body, formed by vertices that are halfway between flat and out and therefore bulge instead of branching cleanly.
- A waist at the junction, thinner than either endpoint, where the body's vertices have moved inward and the strip's have moved outward in the same frame.
- The wordmark. The texture is bound to the faces it was mapped onto, so the letters travel wherever those faces travel and compress or stretch as the front face narrows into the jug. You cannot key the decal's position independently of its geometry; depending on the mechanism, moving the label somewhere else on the finished form means authoring a separate UV target or a shader-driven projection, which is a second thing to build and a second thing to break.
- Light. A handle that appears in a frame where no highlight travelled along it looks pasted on, even when the geometry is correct.
If the strip's takeoff, the body's rounding and the label's shell are fighting each other in the same region of the mesh, the fix is upstream: reshape the target, move where the handle leaves the body, or accept a simpler proposed form. The morph's worst frame is not an embarrassment to be cropped out. It's the evidence that tells you whether the transformation is worth doing at all.
The swap, without pretending it's the same mechanism
The concealed replacement has no correspondence problem because it has no correspondence. Two models, built on their own terms, each as good as you can make it. No shared vertex structure, no reserved strip, no compromise in the block's rest pose. That freedom is real and worth having.
What it replaces is continuity, and continuity has to be purchased back somehow. The usual currency is an occluder: a matte board tracked past the lens, a crate, a doorway, a glove, a sleeve, a passing vehicle — something that covers the object completely for a few frames. Inside those frames you cut. In this walkthrough's hypothetical, the board is a flat card travelling left to right at about a hand's length from the lens, covering the block for six frames of a twenty-four-frames-per-second take. The cut lives inside that cover.
The match list is short and unforgiving. Screen position and path, so the object is going somewhere the frame after the cut as plausibly as the frame before. Camera movement, since a moving camera makes every other mismatch harder to see and easier to forgive. Key light direction and quality, so the shading reads as continuous across the cover. The specular pass on the occluder's leading edge, which is the trick that makes this work more often than it should — if the eye is already tracking the board, it will track the board rather than look for the seam. And sound, which is doing more of the work than anybody wants to admit.
Then check what the plan actually claims. The swap makes no statement about how the object changes. It makes a statement about identity, and the audience supplies it, but only if the two forms are positioned so the substitution is invisible. The cheapest way to fail is a mismatch that has nothing to do with the cut itself: the handled form's widest point sits somewhere the block's didn't, or its base is narrower, or its contact with the table lands a centimetre off. Perfect occlusion, perfect lighting, and the object still jumps.
There's also the question nobody asks during the pitch. The board entering frame is a second event in a commercial that was supposed to be about one object changing. If the obstruction feels motivated — a cart passing a shop window, a hand setting the product down — the audience files it under scenery. If it doesn't, they'll wonder what that thing was, and some fraction of them will decide the object changed because of it. That's not a failure of execution. It's a different idea than the one on the brief.
Put both middles beside the same requirement
Write the requirement at the top of the page in one sentence. Something like: the viewer must read a single object, identified unbroken, changing over four seconds; the wordmark is legible at the start; the handled form is the hero at the end. That sentence is not neutral between the two routes, and that's the point of writing it.
Now check four things.
Does the route produce a middle at all? The morph does, and that middle is where its weak frame lives. The swap has no such middle — no geometry in transit between the endpoints, so nothing to inspect at the halfway point. Its middle is the occlusion instead: the covered interval where the substitution happens. Which turns the question around. Not what does the middle look like, but what fills the viewer's attention across those covered frames — if the answer is "the board," the treatment is spending its best seconds on scenery.
Keep the worst frame of each and compare the failures as equals. The morph's midpoint is probably a pinched junction with a stretched logo. The swap's is the frame after the reveal, where the silhouette either continues or stutters.
Then cost the fix. The fold is a geometry problem: adjust the target, add support where the strip meets the body, move the takeoff, simplify the form. The stutter is a match problem: reposition, rescale, retime the travel, relight. Both are solvable. One leaves you with a model that has to survive every future revision; the other leaves you with an edit and a lighting setup that have to survive every future cut. Neither is free, and the one that's free now is usually the one that costs later.
Finally, ask what each route says about the product. Deformation through corresponding geometry says the material becomes the other thing — the material is the subject. A concealed replacement says these are two things and one takes the other's place — the object is the subject. If the copy line is about what the product does, a swap is a different sentence than the one the voiceover is reading.
What this comparison doesn't establish
The workflow cited here is one stable-topology route in one application, described in version-specific documentation: the 2017 help pages for creating blend shapes from multiple targets and editing target shapes, and the 2026 page covering the same procedure. Those describe steps. They do not promise that a given transformation looks good, stays attractive at the midpoint, or survives export. The documentation's own handling of later topology changes is a separate matter, and it's worth knowing that "compatible geometry" is a condition of this mechanism rather than a law of animation.
The glTF 2.0 specification's morph-target section describes weighted attribute deltas with matching accessor counts — a data relationship about how such shapes are stored, not Maya's interface, not a walkthrough, and not a guarantee that a round trip through a different pipeline preserves what you made. Different systems make different demands. Keep the correspondence requirement attached to the mechanism you're actually using.
And keep the physical claims out of it. A soft block becoming a handled form says nothing about how the product behaves in a hand, how it wears, or what it's made of. Every attribute you're showing symbolically should be labelled as symbolic somewhere the client can see it, because "the 3D version does it" is a sentence that will come back.
Choose, and name what you gave up
Take the morph when the change is the idea and the geometry can be bent to serve it. You will give up the freedom to model each end on its own terms. The block pays for the handle in its rest pose, and the wordmark pays at the halfway frame.
Take the concealed swap when the two endpoints are the point and the transformation is merely how you get between them. You will give up the claim that the object changed. In exchange you get two models that can each be the best version of themselves, and an identity the audience assembles for you out of position, light and timing — which works beautifully, as long as you never make them look for the seam.
Frequently asked questions
When should a morph be chosen over a concealed swap?
Choose the morph when the transformation itself has to stay visible and the geometry can support the states between the endpoints. Choose the concealed swap when the two ends of the product need to be independently excellent and continuity matters less than how each endpoint looks. The article says neither is automatically more convincing; they make different promises about continuity versus arrival, and each carries a price you should name.
What does the middle of a morph reveal that the before and after cannot?
A still frame cannot tell whether a shape arrived by movement or substitution. In a morph, the halfway point is where the claim lives and where the failure is visible and early, such as a crease, pinch, or surface thinning. In a swap, the failure is at one seam and is quiet, because the audience has to supply the object's identity after the reveal.
Why can't a handle be created at frame 40 in a target-based morph?
In a target-based morph, the target shares the base's vertex structure. The mechanism moves points; it does not add them. So the handle cannot be created mid-transformation. It has to exist in the approved block as a flat strip of extra faces folded against the side, waiting for its weights to rise. That means the neutral block is not neutral and may need to hide the reserved strip in earlier shots.
What does a concealed swap require to work?
It replaces continuity, which has to be purchased back, usually with an occluder covering the object completely for a few frames. Inside those frames you cut. The match list includes screen position and path, camera movement, key light direction and quality, the specular pass on the occluder's leading edge, and sound. A cheap failure is a mismatch unrelated to the cut itself, such as a different widest point, narrower base, or table contact landing off.
What does the comparison explicitly not establish?
It is a paper comparison. No model was built, deformed, rendered, or exported for it. The workflow cited is one stable-topology route in one application, described in version-specific documentation from 2017 and 2026, and the glTF 2.0 morph-target section describes a data relationship, not Maya's interface or a guarantee across pipelines. The documentation describes steps; it does not promise the transformation looks good, stays attractive at the midpoint, or survives export.