Credit Physics: The Second Meter Nobody Puts in the Headline
Subscription copy hides burn rate. Chat may sit on the plan. Media often sits on credits. Map that split with this darlink aipricing walkthrough.
The hard test
Count weekly images times published cost, plus the plan. That ugly number is the honest price.
Verify this inside the product. If the claim cannot be observed, treat it as marketing.
How to run it
Failed jobs matter. If retries spend credits, the advertised rate is incomplete. Write the receipt.
Keep one character and written notes. Excitement belongs to onboarding. Evidence belongs to the return session after a break.
How it fails
Annual billing is a commitment tax. Test on a short term. Being wrong for twelve months is the expensive part. Watch renewal paths before the first cycle repeats.
Cost and controls
Read the plan and the credit meter on the same day. Find deletion and cancellation before the first paid cycle. A product that hides exit is telling you the ceiling of trust.
What to record
Date, forgotten details, image drift, credits used, and whether support answered with a number or an adjective.
When to pay
Pay after a 48-hour recall test and a small media batch. If those hold and the bill is readable, a short plan is rational.
When to leave
Leave if you reconstruct the character every evening, if credits vanish without a receipt, or if deletion is theatre.
Closing view
Continuity, cost, controls, and exit beat feature count. A first chat is a demo. A week of ordinary use is the product.
If two tools feel similar after one evening, keep both only until the first recall test. Most ties break there. The rest break when the credit balance moves.
A useful trial stays narrow on purpose. One character. One payment method. Two or three actions that can be repeated after a break. Wide tours feel productive and produce no comparable notes.
Write the priority in one line before you open the app. Memory, visual lock-in, voice continuity, or cost control. If you change the standard halfway, every product looks both good and bad.
Homepage galleries and first-message fluency are advertising. The account screen, the credit receipt, and the return chat after forty-eight hours are the product. Believe the second set when the two disagree.
Privacy work is part of quality. Dedicated email, two-factor if offered, deletion tested on a throwaway note, cancellation found before attachment. Context accumulates faster than people expect.
Stay if repair work shrinks and the bill matches the forecast. Leave if you are still pasting a biography, regenerating a face, or hunting a buried setting. Switching then is cheaper than another hopeful week.
Companion software should be boring in the best way: the same person comes back, the same face is close enough, the same price appears on the statement. Drama belongs in the chat, not in the settings and the invoice.
If you publish a comparison later, use dates and observed failures. Adjectives without a test protocol are indistinguishable from ads. The protocol is the article.
Do not treat a companion app as a substitute for people or professional help. It is software for conversation and generated media. That framing keeps the evaluation adult without turning the page into a sermon.
Revisit the same checks after thirty days if you stay. Plans, memory behavior, and credit rates move. A review from last quarter is a map. The live account is the territory.