View Transitions at the CSSWG: navigation types, nesting, and a contested capture-mode change
View Transitions have moved from a first public draft in October 2022 to shipping in browsers and into production use — including on Airbnb — while its cross-document support means page-to-page transitions no longer require an SPA with a routing layer. The August 21 CSSWG telecon, working from a long agenda, spent most of its time on edge cases in the still-new API.
The navigation descriptor and forward compatibility
The first item was a small Chrome bug: the spec types the navigation descriptor — the opt-in used by every cross-document transition — as an enum, but Blink treats it as a CSSOMString. The GitHub discussion that followed argued CSSOM is generally not fully type-safe this way, and noted that an unknown value in a strict enum would make a rule invalid rather than simply yielding an empty navigation. The working group considered that distinction important for forward compatibility: a value added by one browser but unknown to another must be read as unknown, not confused with none.
noamr: No, the unknown value is not read for purpose of nav. It’s a vt role without navigation descriptor and no initial value Similar to having invalid rule
The semantics settled as follows: none supersedes auto and deliberately cancels a transition coming from the previous document, whereas an invalid value is a no-op.
RESOLVED: navigation is a CSSOMString, it returns an empty string when navigation descriptor is missing or invalid
Nesting and the future view-transition-group property
Today the pseudo-element tree view transitions generate is flat. Nesting groups inside one another — ::view-transition-group within ::view-transition-group — would allow more complex transitions, which is where the view-transition-group property comes in. Its projected syntax defines four kinds of behavior: normal (the current behavior, contained by the root ::view-transition), an <ident> (contained by the element with the matching view-transition-name), nearest (contained by the nearest ancestor with a view-transition-name), and contain (contain all descendants without changing the element’s own position in the tree).
The values can conflict. Given nested elements where B wants to contain C but C explicitly names A as its containing group, the group concluded that the ident is the more specific answer.
vmpstr: … noam and I agree that ident should probably win, seems more specific.
A stronger containment keyword was floated — suggestions included contain-idents and “contain-all” — but was left aside pending a use case.
RESOLVED: idents take precedence over contain in view-transition-group
Snapshot versus captured style
The main agenda item concerned how styles are captured. View transitions currently snapshot the old view and transition to the new one, but properties that define how an element draws when embedded — mix-blend-mode is the spec's example — cannot be applied to its image; they are instead applied to the element's ::view-transition-group() pseudo-element, which generates a box equivalent to the element. These properties therefore land on ::view-transition-group rather than ::view-transition-image-pair(). With groups nesting inside groups, that choice becomes far more consequential, since tree effects such as opacity, mask, clip-path, filters and perspective apply to the whole tree, as do borders, border-radius, overflow, and shadows that paint after backgrounds. Unc captured, the results look broken.
Three options were on the table: change capture for everything by default and add more captured properties instead of a flat snapshot; add a new property such as view-transition-style or view-transition-capture-mode; or add that property with an auto value that switches to the new mode only when a group contains other groups, with an escape hatch back to the old crossfade behavior (which authors can also get through regular DOM nesting). Bramus objected to the name cross-fade for a value, since authors can change the animation to something like scale-up/scale-down, and suggested view-transition-capture-mode: flat | layered instead.
noamr: Some of them are captured and not painted, while others are painted. The ones specifically would all be specified
Although some sentiment favored the first, breaking option, the group preferred not to act on it unseen, and the prototyping of nested groups is expected to surface edge cases that will come back to the working group.
RESOLVED: Change the capture mode for all view-transitions and specify how each property is affected by this capture mode change
RESOLVED: Describe categorization of properties in the Module Interactions sections of each spec
RESOLVED: Blink will experiment and come back with changes needed if there are compat concerns
The full minutes are available on W3C.org.



