How to Troubleshoot AI Companion Image Generation and Credits in 2026

What should you record before retrying?

Record the account state, prompt type, input type, displayed balance and exact error before sending another request.

Take notes rather than repeated screenshots of private content. Write the date, feature name, whether the request used text or an upload, the visible credit balance and the result. Keep the test prompt neutral and short. If the product shows a request ID or error code, save that identifier without sharing the full prompt publicly.

1. Check whether you are signed into the intended account and plan.
2. Record the displayed credits, tokens or generation balance.
3. Note the exact error, loading state or missing output.
4. Check whether the attempt appears in history or billing activity.
5. Confirm that a refresh or retry will not submit the same paid request again.

Expected result: you should have a baseline that lets you tell whether the next attempt changed the system. If the balance is hidden, the billing event is delayed or the error is vague, treat that uncertainty as a product limitation.

How do you run a safe control test?

Use a short text-only prompt with no personal names, faces, sexual content, brand claims or copyrighted source material you do not control.

Try a neutral request such as “a simple fictional landscape with two colors and soft lighting.” The purpose is not to judge artistic quality. It is to learn whether the feature can complete one ordinary request on the current account. If the control succeeds, the next problem is likely connected to the original prompt, upload or requested style. If it fails, do not keep spending credits; inspect the service or account path.

1. Use the same neutral prompt once, without an upload.
2. Wait for the provider's stated processing state or a clear error.
3. Compare the ending balance with the baseline.
4. Record whether a usable output and history entry appear.
5. Stop if the request is charged but the result is missing or unclear.

Expected result: you should know whether the basic generation path works and whether a control request consumes a credit. Do not treat a successful test as proof that every subject, style or upload will work.

How should you change a failed prompt?

Change one variable per attempt: subject, composition, style, length, image input or sensitivity level.

Start from the shortest version of the intended request. Remove unnecessary adjectives, multiple characters, exact real-person likenesses, private details and instructions that combine many actions. Then change only one element and compare the outcome with the baseline. This creates a useful troubleshooting trail instead of a sequence of guesses.

1. Keep the subject fictional and describe it in plain language.
2. Specify one composition or output goal.
3. Change only the style or aspect instruction on the next attempt.
4. Add one harmless detail if the result is too generic.
5. Stop if the request is blocked; do not try to evade a safety rule.

Expected result: you should learn which input causes the change. If the output remains inconsistent after a few controlled tests, record the limitation instead of supplying more personal context or buying a higher plan.

What should you do when an upload is involved?

Use a neutral test image only after checking file handling, retention, training, sharing and deletion terms.

Inspect the image before upload. Remove faces, documents, addresses, reflections, workplace clues and other people's likenesses. Consider whether metadata or the background identifies you. Do not upload an intimate image, an ID document or a private photo of someone else merely to improve a character image.

1. Read the provider's current upload and privacy terms.
2. Check whether the image may be stored, reviewed, used for improvement or shared.
3. Use a generic image or a crop that removes identifying context.
4. Record the upload name and delete it if the test is complete.
5. Verify what remains in history, media storage or trash after deletion.

Expected result: you should know whether the feature needs an upload and what happens afterward. The FTC explains that websites and apps can collect activity through permissions, identifiers and tracking, so minimize the data before troubleshooting the creative result.

How do you check credits and unexpected charges?

Compare the balance before and after one controlled request, then match any change to the provider's current credit definition and billing record.

Do not assume that one click equals one image or one credit. A product may distinguish generations, edits, upscales, variations, input processing, failed attempts or add-ons. Use the current plan and help pages to identify the unit, but treat the provider's actual account history as the source for your transaction.

1. Record the balance before a single test.
2. Submit only one request and wait for its final state.
3. Record the balance, history entry and any receipt afterward.
4. Check whether a failed or duplicate attempt was charged.
5. Pause and contact official support with the request time and ID if the records disagree.

Expected result: you should be able to estimate the cost of one successful output and identify whether a failure consumed a credit. If the provider cannot explain the charge, do not make more attempts to investigate.

What should you verify before paying?

Verify feature access, credit expiry, add-ons, renewal, cancellation, image deletion and support before purchasing more credits.

Complete the basic control test first. Then write down the feature you need and check whether payment unlocks it, increases a limit or only adds credits. Review the current price, billing channel, renewal terms, refund language, cancellation route and data controls. Check whether unused credits expire and whether paid status changes storage or media retention.

Use AIDatingGuide's methodology and reviews to compare media, chat, customization, privacy, price and cancellation. Scores are editorial summaries, not guarantees. Affiliate compensation does not determine the score, and you should verify current provider terms before paying.

When should you stop troubleshooting?

Stop when the error is unexplained, retries consume credits, privacy terms are unclear or the feature requires data you are not comfortable uploading.

Use an offline image editor or a different category of tool when you only need cropping, color changes or a simple visual mockup. Stop immediately if the provider asks you to bypass content controls, upload another person's intimate image or provide identity and payment data without a clear reason.

For a charge dispute, preserve the transaction details and use the official support or payment-channel process. Do not publish private prompts, receipts or personal images in a public forum while seeking help. A useful service should give you enough information to understand failure, cost and deletion.

Summary: a low-waste image troubleshooting workflow

Record the baseline, run one neutral control test, change one variable at a time, track credits and protect uploads before deciding whether to pay.

The objective is not to force every request through. It is to learn whether the feature is reliable, understandable and appropriate for your data. If you cannot answer what failed, what was charged or what happens to an upload, stop and choose a tool with clearer controls.