How to read TRY AI FX test records
Understand evidence, example status and the limits of an accepted visual result.
The catalog combines editorial advice with records of generated examples. These serve different purposes. Advice explains how to choose or inspect an effect; a test record describes a named attempt. Reading one as the other can create an impression of reliability that the evidence does not support.
What an available example establishes
TESTED / EXAMPLE AVAILABLE means at least one acceptable real result is documented. It does not mean the recipe works for every subject, source or model. Compare the displayed input with your own and read any notes about changed geometry, text or inferred detail. A car example does not automatically validate the same effect on a face.
What verification requires
The published methodology describes a protocol of at least six recorded attempts across three representative inputs, with at least five acceptable results and rights-cleared evidence. The collection identifies XRAY, CHROME and LUXURY as VERIFIED. This editorial expansion does not add tests or upgrade any status.
Why the tool description matters
The recorded examples were produced using the Work image-generation tool, whose exact model version was not exposed in the records. The recipes target a photo-and-prompt workflow in ChatGPT, but that does not establish identical output or availability in every account. Dates and tool labels describe the evidence available, not a promise about future models.
Read failures as useful boundaries
EXPAND remains FAILED and outside the main approved-example grids. Some subject-specific comparisons are also pending or unavailable. These states are not repaired by writing a longer explanation. They tell you where a successful example is missing or where review did not establish the intended effect.
Keep observations bounded
A note that a silhouette survived one run does not prove exact geometry. A readable product label does not validate every line of packaging copy. Use the named observations together with your own before/after review. No new experiment, model benchmark or success rate is claimed by these guides.
Editorial guidance · 29 September 2026. Recipe pages retain their own recorded evidence and limitations.