Skip to main content
rone39419531
rone39419531
Participant
July 14, 2025
Answered

Firefly is not understanding what I am typing.

  • July 14, 2025
  • 2 replies
  • 481 views

Trying out Firefly and for creating images to be used as an avatar it has done a great job. But today tried to use it to animate this photo of a Corvette I shot. Shared my original prompt with Chatgpt and it edited it a bit. First attempt the car slid sideways and turned into a cartoon and stayed in frame. This was the second attemopt after describing to Chat what happened the first time. The prompt for this second attempt was: Create a 5-second animation of a red sports car revving its engine on a shiny granite surface. The rear tires should spin rapidly, producing realistic smoke and tire burnout effects. The car should then accelerate straight forward in the direction it is facing and exit the screen bottom left, filling the room with tire smoke as it leaves. Keep the background as shown mostly white with only the dark shiny granite floor as a background element. I did something similar with a white SUV in AKool using a more complex background and it did a much better job. 

    Correct answer Kevin Stohlmeyer

    @rone39419531 AI prompting does have a learning curve and does take time to figure out. A couple of things I can point out. 1. The prompt is very long and the amount of "asks" or directives for a 5 second video clip is unreasonable. Trimming this down can help. 2. Redundant prompts can cause confusion. The clip is auotmatically 5 seconds, so you can remove that. Also did you upload a first frame image? You are noting "keep the background..." which wouldn't compute if there are not elements to refer to. 

    Here is my result using "A red sports car revving its engine on a shiny granite surface. Rear tires spinning, producing smoke and tire burnout effects. Car exits the screen leaving smoke. Background seamless white. "

    While not perfect, editing this starting prompt might get you further. 

     

     

    2 replies

    Kevin Stohlmeyer
    Community Expert
    Kevin StohlmeyerCommunity ExpertCorrect answer
    Community Expert
    July 14, 2025

    @rone39419531 AI prompting does have a learning curve and does take time to figure out. A couple of things I can point out. 1. The prompt is very long and the amount of "asks" or directives for a 5 second video clip is unreasonable. Trimming this down can help. 2. Redundant prompts can cause confusion. The clip is auotmatically 5 seconds, so you can remove that. Also did you upload a first frame image? You are noting "keep the background..." which wouldn't compute if there are not elements to refer to. 

    Here is my result using "A red sports car revving its engine on a shiny granite surface. Rear tires spinning, producing smoke and tire burnout effects. Car exits the screen leaving smoke. Background seamless white. "

    While not perfect, editing this starting prompt might get you further. 

     

     

    rone39419531
    rone39419531Author
    Participant
    July 17, 2025

    Thanks Kevin for your insight. I had started to use Chat GPT to refine my prompts thinking it would be able to help - what it does for the most part is make them longer. What I am finding is there are some things Firefly looks like it does not accept/understand (I had another animation where to flags were supposed to fly into each other from left/right sides of frame and roll into one and the sceen fades to white as an intro for a product announcement and no matter how many times I asked it to now have the flags on flag poles every render had the flags flapping about on flag poles).Using Chat and just myself stating 2 flags no flagpoles written every which way it never got it right. Even Chat could not understand why. 

     

    Then another instance a 3/4 single image of an object firefly turned it into a 360 spin flawlessly. 

     

    The image of the Corvette I shot and used it as a starting point same as I had done in akool only it with a much more complex background (the vette image has a dark marble floor with a blown out to mostly white window for the background) with a prompt not refined by GBT did a much better job. But since I am an Adobe user for decades I want to not have to use/subscribe to so many different programs. I only wanted a 5 second clip or even shorter as its supposed to used as an intro. 

    Kevin Stohlmeyer
    Community Expert
    Kevin StohlmeyerCommunity Expert
    Community Expert
    July 17, 2025

    @rone39419531 I have some good news for you - starting today (7/17) Firefly has been updated and now includes Prompt Suggestions. Click the icon on the bottom right to turn on. It will give you suggestions to improve your prompt for video and image generation!

     

     

    Kartika Rawat
    Community Manager
    Kartika RawatCommunity Manager
    Community Manager
    July 14, 2025

    Hi rone39419531!

    Thanks for sharing your experience and feedback while working with Firefly! We’re glad to hear it’s been working well for avatar image creation. We have noted your comparison and will share it with the product team.

     

    Let us know if you have any questions.

    ^KR

    rone39419531
    rone39419531Author
    Participant
    July 17, 2025

    Hi Kartika: Thank you for the comment. I think the challenge is finding the balance between too complex a prompt and too little info. I figured talking with chat gpt it would be able to clean up my prompt. what it did is make it more complex. The sample I showed from AKool the prompt was much more simple. 

     

    Terms & ConditionsAccessibility statement
    Using the community Terms of use Privacy Cookie preferences Do not sell or share my personal information ad choicesAdChoices