Finding the model strings you already send
A retirement schedule is only actionable if somebody can say which of their own systems uses the retiring version. Most teams send a string written months ago and never revisited, which makes the schedule a list of unfamiliar names rather than a plan. As of 2026-09-12.
| Where to look | Why it hides there |
|---|---|
| Scripts and pipeline configuration | Written once, during setup, by somebody who left |
| Saved presets inside a product | Chosen in a hurry and never reopened |
| A tier or plan selection | No string at all, so nothing to grep for |
| A convenience alias | Deliberately not a version, and it moves |
| A third tool in the chain | It calls a model and does not say which |
Inclusion rule. Places a model name ends up in a production pipeline, with the reason each one is easy to miss. Order. From the easiest place to search to the hardest.
1An old string is the real default
Where a platform lets a caller name a version, the platform chooses nothing and almost every caller has a string it uses without thinking. That is a default in every practical sense and it belongs to the customer.
It is better than an unnamed tier, because nothing changes underneath without an edit. It is also a liability nobody is reminded to audit.
2An alias is worse than an old string
A convenience name resolves to whatever the vendor currently points it at, and its whole purpose is to change. It appears on no deprecations table, so there is nothing to watch and nothing to match.
For exploration that is convenient. For a run of episodes it is the invisible version of exactly the event a retirement schedule exists to warn about.
3Some layers cannot be audited at all
Where a production reaches a model through an application, the version string belongs to the application and is often not published. The audit then stops at a tier name, which cannot be matched to any dated row.
What is left is the record kept per delivered file, plus two dated readings of the application's own page. That is weaker and it is what most of this market offers.
4Do it while the model still answers
An audit run after a retirement is an incident response. Run before one, it is an afternoon: list the strings, note where each is sent from, and check each against whatever lifecycle page the vendor publishes.
The output is a short list somebody can act on, which is the point. Most teams have never produced it and would find something surprising on the first pass.
5Where each vendor's own wording lives
None of the above is a date. Announced end dates, and the apps still carrying the models they apply to, are kept on the calendar with the announcement each one came from.
- Model identifier — why schedules are written about strings rather than products
- Version alias — a name that moves on purpose, and appears on no table
Guidance rather than a dated entry. No line here is a vendor statement, and none should be read as an announcement. The sourced material is on the calendar. Related: Finishing a season, Telling people downstream.
6Production that outlasts a model
Model end dates touch “Auditing your own strings”; SceneMixer's guides describe how its pipeline treats the models it runs:
16:9 landscape (SceneMixer's default) and 9:16 vertical (the usual short-drama frame), chosen per project.
SceneMixer, Seedance 2.0 in a Real Short-Drama Pipeline — “Which aspect ratios does the pipeline generate?” (read 2026-10-13)
Recommended is a badge on the tier the site suggests for most finished segments. You stay on the default, 480P, until you pick another tier.
SceneMixer, Wan 3.0 · 720P: the recommended tier — “Does Recommended mean 720P is the default?” (read 2026-10-13)