Skip to main content
Participant
July 21, 2026
Answered

Wasted Credits

  • July 21, 2026
  • 2 replies
  • 19 views

AI misunderstood all of my video and image prompts, and my credits were wasted. At this rate, with 4,000 credits, I can make at most 5 videos per month, which means I won't be getting my money's worth. I'm going to cancel my subscription.

    Correct answer Hajar Toutsi

    Hi Utku Fırat,
    I completely understand your frustration. It is incredibly disappointing to spend your credits and not get the results you're aiming for.

    Before you cancel your subscription, I'd like to share a few specific prompting strategies that might help Firefly better understand your vision so you can stop wasting those credits. The way the model interprets instructions can sometimes be a bit counterintuitive.

    Here are a few adjustments that can make a significant difference:

    • Be explicitly literal: Firefly doesn't make assumptions about what to preserve in an image or video. You have to spell out exactly what needs to stay intact. For example, explicitly adding a phrase like, "Keep the exact faces, identities, and all elements of the people in the image intact while adding..." gives the model strict boundaries.

    • Focus on what you want, not what you don't want: The model currently struggles with negative prompting. If you tell it "don't change the faces," it often fixates on the word "faces" and alters them anyway. Keep your prompts focused 100% on describing exactly what should be in the final result.

    • Use the highest resolution possible: If you are working with reference images, resolution is critical. Lower-resolution photos give the model too much room to guess, which frequently leads to unintended changes or weirdly substituted faces. Always provide the highest quality image available.

    • Choose the right model for your project: Not all models process constraints and visual effects the same way. If the current model keeps missing the mark, try switching to a different version in the settings. Learn how to determine the appropriate model for your project.

    • Avoid generating text: Text generation within AI models is a known limitation, and Firefly is no exception. Adding text or symbols directly during the generation step can sometimes be unreliable, even when quoted. The best way is to add text separately.


    Prompting definitely has a learning curve. If you want to dive a bit deeper into how to structure your requests to get exactly what you need, these official resources are highly recommended:


    I hope this helps you get more value out of your credits. If you want to share a specific prompt that isn't working for you, feel free to post it here and we can try to help you troubleshoot it!

    Hajar

    2 replies

    Participant
    July 22, 2026
    Thank you. Despite all this, the credit you are offering is still very low. I might reconsider if that changes. Goodbye.
    Hajar Toutsi
    Community Manager
    Hajar ToutsiCommunity ManagerCorrect answer
    Community Manager
    July 22, 2026

    Hi Utku Fırat,
    I completely understand your frustration. It is incredibly disappointing to spend your credits and not get the results you're aiming for.

    Before you cancel your subscription, I'd like to share a few specific prompting strategies that might help Firefly better understand your vision so you can stop wasting those credits. The way the model interprets instructions can sometimes be a bit counterintuitive.

    Here are a few adjustments that can make a significant difference:

    • Be explicitly literal: Firefly doesn't make assumptions about what to preserve in an image or video. You have to spell out exactly what needs to stay intact. For example, explicitly adding a phrase like, "Keep the exact faces, identities, and all elements of the people in the image intact while adding..." gives the model strict boundaries.

    • Focus on what you want, not what you don't want: The model currently struggles with negative prompting. If you tell it "don't change the faces," it often fixates on the word "faces" and alters them anyway. Keep your prompts focused 100% on describing exactly what should be in the final result.

    • Use the highest resolution possible: If you are working with reference images, resolution is critical. Lower-resolution photos give the model too much room to guess, which frequently leads to unintended changes or weirdly substituted faces. Always provide the highest quality image available.

    • Choose the right model for your project: Not all models process constraints and visual effects the same way. If the current model keeps missing the mark, try switching to a different version in the settings. Learn how to determine the appropriate model for your project.

    • Avoid generating text: Text generation within AI models is a known limitation, and Firefly is no exception. Adding text or symbols directly during the generation step can sometimes be unreliable, even when quoted. The best way is to add text separately.


    Prompting definitely has a learning curve. If you want to dive a bit deeper into how to structure your requests to get exactly what you need, these official resources are highly recommended:


    I hope this helps you get more value out of your credits. If you want to share a specific prompt that isn't working for you, feel free to post it here and we can try to help you troubleshoot it!

    Hajar