Video questions: credits, memory and image quality

Reviewed November 2, 2026. These answers connect questions raised by public video reviews with the library's existing evidence. Creator captions and identity-unverified comments are leads, not native test results or a verified audience trend. No new OD account action was performed.

Record your own observation · Plan a comparison · Read this review's evidence register

Does having credits unlock every prompt field?

A credit balance and permission to use a feature are different things. The public Generate form read November 1 labels Custom prompt Premium and displays an upgrade gate. That observation does not establish every signed-in account's access, and having dreamcoins does not by itself establish entitlement. If the scene field is locked, pause that exercise or use the controls your account offers. Keep scene instructions out of permanent appearance fields when they do not belong there. See Generate: inspect the active inputs.

A request for free credits appeared in the comment sample, but it does not establish an offer. No free-credit offer or method of unlocking Premium was verified by this research. Read the current account and offer terms through the platform's actual controls before relying on a promotion.

Is the balance in a review the monthly allowance or the cost of an image?

Keep three values separate:

One review associates a coin balance with a monthly purchase and later uses a different currency term for an image charge. Its captions do not establish a unit conversion or a current allowance. Another review advises checking current prices. This library therefore retains current-control checks instead of copying the narrated numbers. See account costs and support and voice/call costs.

Record the cost before a manual action, then keep the actual charge or refund evidence separately. Unknown amounts remain unknown. The local recorder cannot inspect your balance or enforce your spending ceiling.

Does a character bringing up an event prove its memory works?

First decide what you are testing. Prompted recall asks about an established fact. An unsolicited callback asks whether the character brings that fact up without a direct reminder. The second outcome also depends on initiative, timing and intervening context. A scheduled notification is a separate feature and was not established by this review.

An October 23 short proposes a future-event callback test but explicitly leaves the result pending. It is a test idea, not a memory success report. The memory lesson now separates those criteria. Absence of a callback alone cannot isolate a memory failure.

Use a saved accepted fact and record the model, relevant context layers and exact intervening messages. Inspect the saved summary separately from the actual reply: a correct summary does not guarantee a correct next reply. Keep prompted recall and callback conditions in separate records. Leave them not_run until actual outputs exist.

The dated guide's approximate memory-update cadence and field allowances describe documentation, not a guarantee of perfect recall. Earlier native pilot results retain their own limited cohorts and outcomes; the new captions do not change them.

Why do two reviews disagree about image quality?

An overall rating can combine different models, routes, source images, presets and expectations. The sampled comments included contrasting quality reports, but they lacked comparable author identity and controlled settings. Neither establishes a platform-wide result. A favorable video narrator also acknowledges defects; captions alone do not let this review inspect those images.

For a useful manual check, retain the exact input, actual model/route, active presets, batch size and every returned output. Judge identity, clothing, scene, framing and extra people separately. Keep visual polish separate from whether the request was followed. See Generate and compare.

For an edit, select the actual accepted source and retain its identifier. Copying its words into Generate makes a new generation; it does not select the original pixels. Compare the requested change and the retained features side by side. A mask or preservation instruction is a target, not a guarantee. See the editing lesson and test records.

What this video review covered

vidIQ discovery returned 24 rows across three bounded searches, representing 18 distinct results. Five relevant review videos from separate channels were selected; three caption tracks were read. Captions describe narration, not verified on-screen controls or a visual inspection. Commercial review or offer links appeared in the source descriptions; those links were not followed.

Each selected video received one relevance-ordered comment request capped at 30 threads. The returned sample contained eight threads and five replies, dated August 4–October 24. One creator promotion was excluded. The other 12 comment items lacked comparable author-channel URLs, even after two owner handles were verified; they are not counted as verified viewer evidence and are paraphrased without identities. Repeated complaint text, chatter and off-platform promotion do not supply independent corroboration.

A separate niche-wide sample drifted into other AI tools and general production topics. None of its 11 returned threads was retained for OD findings. Empty comment responses, sparse samples and metadata/count differences are coverage limits. This review does not establish a recurring audience pattern, a current price, a plan entitlement, a model winner or a new prompt-efficacy result.

The answers above were checked against the existing dated lessons, retained official-source evidence and separately scoped native records. Those sources were not freshly fetched at OD. Automated OD collection and native testing remain on hold under the policy review described in the research workflow.