Troubleshooting generations

Most generations just work. When one doesn't, here's how to read what happened and what to do next.

First, tell which kind of problem you have — it decides whether retrying helps:

• Transient — an upstream hiccup, a timeout, or heavy load. Retrying usually works, often on the first try.
• Input or settings — a blocked prompt, a refused upload, or a value the model doesn't support. Retrying the same thing fails the same way; change the input first.

The message tells you which: "give it another try" or "took too long" is transient; anything that names your prompt, your image, or a setting means fix that first.

A generation failed or errored

If a render fails, the credits for it are refunded automatically — you're only charged for results you actually get. Failures are usually transient (an upstream model hiccup or heavy load), so the first move is simply to try again.

If it keeps failing:

• Simplify the request — a shorter clip, a lower resolution, fewer reference images.
• Switch to a different model for the same prompt; see Choosing a model.
• Check your connection and reload before re-rolling repeatedly.

It's slow or stuck "in queue"

Most generations finish in seconds. Longer clips, higher resolutions, and audio-native models take more time, and heavy load can add a short queue. That's normal — leave the tab open and it'll arrive. If you generate a lot, a Pro tier gets priority in the queue. A job that hangs far longer than usual is safe to cancel and retry; cancelled jobs are refunded.

"Too many tasks running"

You can have up to 10 generations running at once on your account — images, video, and audio combined. Start an 11th and you'll be asked to wait for one to finish. This is a per-account limit that clears as your own jobs complete, so it's different from a slow queue: nothing's wrong, you've just filled your own slots. Let one finish, or cancel one you don't need, and you're unblocked.

"Insufficient credits"

This means the generation costs more than your current balance. You can:

• Wait for your free daily credits to refresh.
• Upgrade for a larger allowance — see Credits and plans and Pricing.
• Lower the cost: shorter video, lower resolution, or a faster model tier all reduce the price shown before you press Generate.

A prompt or result was blocked

Content can be blocked before it runs (your prompt or a reference image) or after it finishes (the result itself), by the content policy or a model's own safety filter. Common triggers:

• Sexual, explicit, or violent content — reword to describe the scene without the blocked element.
• A real person's likeness, used in a misleading way — describe a generic character instead of naming a real individual.
• Recognizable copyrighted characters, logos, or styles — some video models (such as Seedance) block a *result* that looks like a known character or brand. Reword to avoid them.
• Music that names an artist or song — the music model rejects prompts that reference a specific artist, track, or copyrighted lyrics. Describe the *style and mood* instead — see Music & sound design.

If you're unsure what's allowed, review the Terms. A blocked prompt or result doesn't cost credits — blocked and failed generations are refunded automatically.

A reference photo of a real person was rejected

Some video models — notably Seedance, the default on the image-to-video tool — run a privacy filter that can reject a reference image it reads as a real person, with a message like "Photos of real people aren't supported." That's the model's rule, not a problem with your photo. To animate a real face:

• Switch to a model that handles real portraits — Kling 3.0 (also best for lip-sync) or Veo 3.1.
• Or stylize the photo first in Muse or the image generator, then animate the result.

My upload won't work

If a reference image or source clip is refused:

• Check it's a standard format (JPG/PNG for images, common video formats for clips).
• For face-based features like Muse or lip-sync, the face must be clear and unobstructed — uploads are validated for a usable face.
• Very large files can time out; try a smaller or shorter source.

"Your prompt is too long"

Each model caps how long a prompt can be, and PonPon doesn't trim it for you — an over-long prompt fails instead of running. Front-load the essentials (subject, style, the few details that matter most) and cut the rest. A focused prompt usually beats an exhaustive one anyway — see Prompting for images.

"Not available in your region"

A few models — some powered by OpenAI — aren't available in every country. If you see this, switch to another model for the same job; see Choosing a model.

The result is poor — but it didn't fail

That's a craft problem, not an error, and it's the most fixable:

Still stuck? The FAQ covers the quick questions, and each guide above goes deeper.