A request in a conversation, a generation mode and a stored picture are three separate objects. Confusing them can lead to the wrong charge, an unwanted video request or a mistaken belief that removing a message also removed its media.
This comparison examines documented routes and the confirmation text on a public Candy character page. It does not report a paid chat test, a completed deletion or an image-quality contest.
Start with the action, not the adjective
Infatuated’s plan table lists different token entries for an image in chat, a standalone image-tool action, and their V2-character variants. The table shows 3 and 5 tokens for the two chat routes, compared with 5 and 6 for the corresponding tool entries on October 15, 2026.
That does not make the cheapest entry the best route for every job. A conversational reply may be intended to reflect the current scene. A standalone image may be intended for a composed visual with explicit setting and lighting instructions. Record the route before comparing the result or its cost.
Candy exposes a mode distinction
The public Emma Moore chat page contains action buttons and a warning that a video request requires video mode. It also includes confirmation text for turning media mode off: images and videos stop, with no media-mode token use while it remains disabled.
Those are useful controls to understand before asking for content. They are not evidence that every Candy character, account tier or future interface has identical pricing or behavior. Nor does disabling media imply that every other paid feature becomes free.
For a text-only scene, leave the intended session text-only. If the goal is a picture, identify the actual image action. If the goal is a clip, use the video route and read its quote. A phrase containing “video” should not be treated as interchangeable with a text description of a scene.
Budget a bounded request
Suppose the job is two still-image replies from an existing Infatuated character. At the listed ordinary chat-image rate, those image actions total 6 tokens; on the V2 chat-image entry they total 10. Those are conditional image subtotals, excluding any message charges, new-character creation or other actions.
The corresponding standalone entries would produce different table-based subtotals, but they would also be a different request route. A fair record therefore contains:
| Field | Example entry |
|---|---|
| Intended result | Two retained still-image replies |
| Character route | Ordinary or V2, as actually selected |
| Action | Image in chat |
| Generation attempts | Two, before any retries |
| Other charges | Record separately |
| Retained results | Count after review |
Do not convert Candy tokens and Infatuated tokens as if they were a shared currency. Candy’s subscription page distinguishes included text access from tokens, while the per-action quote must match the selected media request. A nominally smaller number cannot establish a smaller bill.
A deleted message can leave a gallery object
Candy’s same public page includes a deletion confirmation explaining that a response is removed with the message, but an image or video response can remain visible in the gallery. This is a concrete reason to separate conversation cleanup from media cleanup.
Infatuated’s creation page describes a private gallery. That label identifies an audience setting; it does not establish what happens to every stored image when a chat is removed. This research did not verify a matching deletion behavior there, so the Candy instruction must not be projected onto Infatuated.
If the purpose is simply to tidy the conversation, removing the message may be enough for that visual task. If the purpose is to remove a generated asset, the gallery object requires its own attention. Provider retention and backup rules remain a separate policy question.
Use a small action record
After a session, preserve only the operational details needed to understand it: mode, action, shown charge, result reference and whether the result is retained. Avoid storing private dialogue in a comparison spreadsheet when a simple action count is sufficient.
This record helps explain a result such as “two requests, one useful picture, one remaining gallery asset.” It is much more actionable than “the image feature was good.”
Which documentation is more useful for this question?
Infatuated’s table is useful for distinguishing several image-action units before budgeting. Candy’s public confirmation text is useful for understanding media-mode behavior and the difference between removing a message and removing its gallery media.
Neither source proves output quality or successful erasure. The practical gain is a clearer sequence: choose the mode, read the action quote, generate deliberately, and manage the resulting media as its own object.