When Your Card Stops Working: Why Templates Break and How to Fix Them (2026)
حكم سريع: A template card that worked last month and fails today is almost never a broken card — it is a changed input, a changed model, or a changed expectation. Cards break in five predictable ways, and four of them have a fix that takes less than two minutes. Learn to tell them apart before you spend credits re-running a card that was never going to work.

Contents:
- What “broken” actually means
- The five ways a card breaks
- The two-minute diagnosis
- Fix it yourself: the five repairs
- When the card is not broken — you are
- Build a card that does not break
- What a broken card costs, in credits
- Honest limitations
- الموافقة, stated plainly
- FAQ
1. What “broken” actually means
The phrase “stops working” covers at least four different failures. The card may refuse to run at all — an error, a blocked generation, a message about the image. It may run and produce something unusable — a mangled body, a wrong face, a background that makes no sense. It may run and produce something fine but different from what it used to produce. Or it may run, produce a good image, and still feel broken, because the result no longer matches what you wanted.
Each is a different problem with a different fix, and the most common mistake is to treat them as one problem called “the card is broken.” A card that errors is an input problem. A card that produces mangled bodies is a model problem. A card that produces a different result from last month is a model-change problem. A card that produces a good image you no longer like is a taste problem. The diagnosis in section 3 separates them, because the fix for one will not help the others, and the wrong fix costs credits.
The other thing to establish is whether the card ever worked. A card you have run twenty times has a history, and the history tells you what changed; a card you have run once has no history, and the problem may be that it never suited your photo. If you have never had a good result from a card, the card is not broken — it is the wrong card for your input.
2. The five ways a card breaks
The input changed. The most common cause, and the easiest to miss. You used a different photo — a different angle, a different light, a different crop — and the card, tuned for the last photo, does not work on this one. A card built on a front-facing, well-lit portrait will struggle with a side view in bad light. Compare the two photos: are they the same kind of photo?
The model changed. The generator updates its model, and the card’s prompt — written for the old model — now means something different. A word that meant one thing last month means another now. This is the hardest break to fix, because nothing is wrong with your input or the card: the ground moved underneath both. Re-test with the original reference image, and rewrite the prompt if the result has drifted.
The card’s reference image changed. Some cards depend on a reference image — a pose, a face, a composition — and if that reference was replaced or removed, the card’s output changes with it. If you built your own card, check the reference is still the one you chose. If you use a site’s card, check whether the preview image has changed: a new preview usually means a new reference.
Your settings changed. A different size, a different ratio, a different style setting — any of these can change the result out of all recognition, and all are easy to change by accident. A card built for a square image will behave differently at 16:9 — check the settings before you blame the card.
Your expectations changed. The card works exactly as it always did, and you no longer like the result. You have seen better images, your taste has moved on, or the card’s style — fresh six months ago — now looks dated. This is not a failure, and there is no repair: it is the normal end of a card’s life, and the fix is to outgrow it.
3. The two-minute diagnosis
Run these four checks in order, before you spend another credit. They take two minutes, and they separate the five breaks from section 2 — and when your card stops working, knowing which break it is matters more than any single fix.
| تحقق | How | What it tells you |
|---|---|---|
| 1. Run the card on its own preview image | Use the image the card shows, not your photo | If it works, your input is the problem; if it fails, the card or the model is |
| 2. Run your photo on a different card | Pick a card you know works | If it works, the first card is the problem; if it fails, your photo is |
| 3. Check the settings | Size, ratio, style — compare with the card’s preview | A mismatch here explains most “sudden” failures |
| 4. Run the same card twice | Same photo, same settings, two runs | If the two runs differ a lot, the problem is randomness, not breakage |
Read the results as a grid. If check 1 passes and check 2 fails, the problem is your photo. If check 1 fails, the problem is the card or the model — and whether the card worked last month tells you if the model changed. If check 3 finds a mismatch, fix the settings and re-run before anything else. If check 4 shows big differences between two identical runs, the card is not broken: the generator is random, and you need to run it twice before you judge it.
4. Fix it yourself: the five repairs
Repair 1: match the input to the card. Look at the card’s preview image and list what it has: front-facing or side, well-lit or moody, full body or close-up, plain background or busy room. Then look at your photo and list the same things. If two or more differ, the card is fighting your photo. Choose a photo that matches the preview, or a card whose preview matches your photo. Do not force a mismatch: a card will not learn a new pose from your photo, and the attempt costs 49 credits a time.
Repair 2: re-test with the original reference. If you suspect the model changed, run the card on the image it was built on. If the result has drifted — same card, same reference, different output — the model changed, and the prompt needs rewriting. Shorten the prompt first: a long prompt written for an old model is the most likely thing to have drifted. Then re-add details one at a time.
Repair 3: reset the settings. If check 3 found a mismatch, set the size and ratio to match the card’s preview, and clear any style setting you added. Run once. If it works, the card was never broken — the settings were. Write the working settings down.
Repair 4: run it twice before you judge. If check 4 showed big differences between identical runs, the card is fine and the generator is random. Run the card three times and keep the best. The alternative — declaring a good card broken on one bad roll — costs more, because you lose the card.
Repair 5: retire the card. If none of the above works, and the card worked last month, the model has moved on and the card has not. Retire it. Keep the prompt in your archive — it may work again on a future model — but stop spending credits on it. A card you built yourself, on a photo you chose, is the one you can repair.
5. When the card is not broken — you are
There is a failure mode no diagnosis will catch: the card is working, the photo is right, the settings are right, and the result is still bad. The problem is the expectation — you are asking the card to do something it was never built to do. A card built for a full-body portrait will not make a good close-up; a card built for a studio look will not make a good outdoor shot.
The test is simple: look at the card’s preview and ask whether your photo could have been taken in the same session, in the same room, by the same photographer. If the answer is no, the card is not broken — it is the wrong card. Find the card whose preview looks like your photo.
The other version is the opposite: you keep using the same card because it worked once, and every new photo goes through it, whatever the photo is. A card is a tool for one kind of image, not a style for all images. When it fails on the third different kind of photo in a row, the card has not broken three times — you have outgrown it.

6. Build a card that does not break
A card breaks when it depends on too many specific things: وضعية واحدة, ضوء واحد, جسم واحد, one model. A card that does not break depends on as few as possible. Three rules make a card durable.
First, write the prompt for the idea, not the image. “A woman standing by a window in soft evening light” survives a model change better than “a woman in a red dress standing by the third window from the left.” The first is a scene; the second is a photograph, and a photograph is fragile.
Second, keep the prompt short. Every word is a point of failure: a word the new model interprets differently, a detail it renders badly, a phrase that meant one thing last year and another now. A ten-word prompt has ten chances to break; a thirty-word prompt has thirty.
Third, test the card on three different photos before you trust it. If it works on all three, it is a card. If it works on one, it is a lucky prompt, and it will break the first time you change the input.
7. What a broken card costs, in credits
A broken card costs credits twice: the generations that failed, and the generations you spend diagnosing it. The diagnosis in section 3 costs nothing if you use the card’s own preview image. The expensive mistake is re-running a broken card five times hoping it fixes itself.
| فعل | النقاط | $19.99 قطيع ($0.020/credit) | $59.99 قطيع ($0.012/credit) |
|---|---|---|---|
| One image template run (49 الأرصدة) | 49 | $0.98 | $0.59 |
| Three diagnostic runs (49 كل) | 147 | $2.94 | $1.76 |
| Five blind re-runs of a broken card | 245 | $4.90 | $2.94 |
| One video template, 5 ثوان (351 الأرصدة) | 351 | $7.02 | $4.21 |
| One video template, 8 ثوان (561 الأرصدة) | 561 | $11.22 | $6.73 |
| One undress (21 الأرصدة) | 21 | $0.42 | $0.25 |
For reference: $19.99 buys 1,000 credits and $59.99 buys 5,000 الأرصدة, both one-off payments. The annual packs are $119.88 $359.88, paid once, with credits issued monthly across twelve months. الأسعار تم التحقق منها 13 نوفمبر 2026. The cheapest repair in this guide is the two-minute diagnosis, which costs nothing — run it before you spend a single credit on a re-run.

8. Honest limitations
This guide assumes the card was working and then stopped. If a card never worked, none of these repairs apply — it was never right for your photo, and the fix is a different card. Be honest about which case you are in.
The second limitation is that the model can change without notice, and without a changelog. A card that worked yesterday can fail today because the generator updated overnight, and there is no announcement to read. The only evidence is the output, and the only fix is the re-test in section 4.
The third limitation is that some cards are broken by design: a prompt that only worked by luck, a reference image that was never quite right, a style that depended on a quirk of an old model. These cards cannot be repaired, only replaced. Retire them quickly.
9. الموافقة, stated plainly
Only generate and share images of yourself, or of fictional adults you created. Never generate or share an undress or a face swap of a real person who has not agreed — not a partner, not an ex, not a public figure, ليس غريبًا. Agreement means a clear yes, for this image, for this kind of sharing. Silence is not agreement, and a yes to generating is not automatically a yes to sharing.
A broken card does not change this. If a card fails on a real person without agreement, the failure is not the problem — the attempt is. If the person in the image is under 18, or might be, stop entirely. A working card is not a permission slip, and a broken one is not an excuse.
FAQ
How do I know if the model changed or my photo is the problem?
Run the card on its own preview image. If it works, your photo is the problem. If it fails on its own preview, the card or the model changed.
Should I re-run a broken card to see if it fixes itself?
لا. Run the diagnosis in section 3 first — it costs nothing. A blind re-run costs 49 credits for an image card and 351 for a five-second video.
Can I fix a card by adding more words to the prompt?
Usually the opposite. A long prompt written for an old model is the most likely thing to have drifted. Shorten the prompt first, then re-add details one at a time.
What if the card works on the preview but not on my photo?
Then the card is fine and the photo does not match it. Compare the two: angle, ضوء, crop, خلفية. Either choose a photo that matches the preview, or a card whose preview matches your photo.
How many times should I run a card before declaring it broken?
Three times, with the diagnosis in between. One bad run is randomness; three bad runs after a diagnosis is a broken card. Retire it and keep the prompt in your archive.
Is a video card more likely to break than an image card?
نعم, and it costs more when it does — 351 credits for five seconds, 561 for eight. Diagnose a video card with a five-second run, not an eight-second one, and only scale up once the short version works.
تابع القراءة: Build Your Own Card: Turning a Winning Prompt Into a Reusable Template · Outgrowing the Card: When to Stop Using Templates and Generate From Scratch · Your First 10 Template Cards: A Starter Set for Every Mood