Sometimes a video generation won't go through in Visuals. Don't worry — your credits are automatically refunded whenever this happens. Here's why it might fail and how to fix it.

AI video models have built-in content policies. Your generation may be rejected if your prompt or reference image contains:

Occasionally the AI model itself encounters a technical issue. This isn't anything you did wrong — it just means the model couldn't complete the request. This is rare but does happen.

We know many of you are labels, managers, and artists creating content for real public figures — cover art, promotional videos, visualizers, and more. You can absolutely use celebrity and artist photos as reference images. The key is how you write your prompt.

When you upload a photo of an artist or public figure, do not use their name to describe what the person is doing in your prompt. Instead, refer to them as "the subject," "the person," or "the figure." The AI model already sees who's in the image — naming them in the action is what triggers content filters.

❌ Will likely fail: "Bad Bunny walking through a neon-lit city at night"

✅ Will likely work: "The subject walking through a neon-lit city at night"

If you need the artist's name or a song title to appear as text on the visual (for cover art, title cards, etc.), you can include that separately in your prompt — just don't tie the name to the person doing the action.

❌ Will likely fail: "Bruno Mars performing on stage with the song title I Just Might"

✅ Will likely work: "The subject performing on stage. Include the text 'Bruno Mars — I Just Might' as a title overlay"

Beyond the celebrity naming tips above, try removing any brand names or copyrighted references. Describe the style or vibe instead.

Before: "A rapper who looks like Drake performing on stage" After: "A male artist in a black jacket performing on a stadium stage at night"

If you uploaded a reference image, make sure it doesn't include:

Try swapping it for a different image or removing it entirely to see if that resolves the issue.

Not all models have the same content policies. If one model rejects your prompt, switch to a different model from the dropdown and try again. What gets flagged by one model may work fine on another.

Longer, more complex prompts can sometimes cause issues. Try shortening your description to the essentials and see if that helps.

Will I lose credits if a generation fails? No. Your credits are automatically refunded every time a generation fails.

Can I use celebrity or artist photos as reference images? Yes! Just don't use the celebrity's name to describe what the person is doing in your prompt. Upload the photo and refer to them as "the subject" or "the person." See the section above for detailed examples.

Can I include an artist's name or song title on the visual? Yes — mention it as a text or title overlay instruction. Just keep it separate from describing the person's actions. For example: "The subject singing into a microphone. Include the text 'Artist Name — Song Title' as a title."

Can I use song titles in my prompts? Generally yes, but pairing a song title with an artist name as an action description can trigger content filters. If you run into issues, try describing the song's vibe or theme instead of using the exact title and artist together.

How do I know if it was a content issue vs. a technical error? Right now, both show the same error message. A good test: if you simplify your prompt and remove any names and it works, it was likely a content policy issue. If it keeps failing regardless, it's probably a temporary model error — try again later or switch models.

Can I use my own photos as reference images? Absolutely — original photos you've taken yourself are the safest option.

I changed my prompt and it still fails. What now? Try these steps in order:

How do I contact support? Use the chat widget in the bottom-right corner of the app or email us directly. We're happy to help troubleshoot.