AI Companion Media Claims Check: Images, Voice and Video
A page may display polished examples without explaining which plan, model, queue or credit amount produced them. Verify the path from the promise to the feature you can actually use.
Separate previews from user output
Look for labels that distinguish marketing artwork, community examples and generated output. A beautiful character card does not prove the same style is available in chat. Check whether examples state the feature, plan or generation mode used.
Find the real cost unit
Images may cost credits per attempt, while voice and video can use different balances. Check whether failed generations consume credits, whether regeneration is charged and whether a subscription includes a monthly allowance. Avoid estimating from one promotional example.
Check availability boundaries
Some features vary by device, region, character, queue or account age. Read current help pages and the final plan description on the day you pay. If a feature is essential, ask support to confirm it for your account context.
Protect consent and identity
Do not upload another person's private image or voice without permission. Avoid requests that impersonate real people or create deceptive intimate material. A service allowing a prompt does not make every use ethical or lawful.
Run a small first test
If you choose to pay, start with the smallest commitment that can test your main feature. Record the number of attempts and credits used. Your own low-stakes result is more useful than assuming a showcase represents every output.
Compare a promise with a testable outcome
Translate a broad promise such as premium video into one outcome you can observe: maximum length, queue time, resolution range, character consistency or credits per attempt. Find where the current service explains that outcome. If it does not, ask support before buying a large balance. During a small test, note the settings and credits used without publishing private output. One successful sample does not prove every generation will match, and one failed attempt does not prove the whole feature is broken. The useful evidence is whether the service explains limits, charges consistently and gives enough low-stakes access to judge the feature. Keep consent and identity boundaries in place regardless of technical capability.