I select an image of a pitbull with its owner on the couch and asked for a video generation, the video showed a golden retriever and a completely different woman... rip off of my credits for the month.

Hey ​@caring_mindset0099 
To add to what ​@Trevor.Dennis mentioned

I like to download the image I like. In the video generation model, if the image I choose isn’t the present I can at least remove the unwanted image; and, then upload the image I have downloaded.

When I use a reference image and want to keep the reference I tend to use one these prompts at the beginning of the Prompt.

Prompt suggestion 1
Keep the reference image…

Prompt suggestion 2
Use the reference image…

Help us. Help you.
Could you create a full screen video of your workflow showing us what is happening? This would help considerably.

>>Make sure to hide your personal information: email account and anything other personal details.<<

Cheers

Nate

Starting out with Firefly video can be frustrating.  Especially if your plan does not include a lot of credits.

What was your prompt, and what were you hoping to see in the video?

First frames are a good way to keep some control of the output.  I use Gemini 3.1 Nano Banana in Photoshop to generate the first frame.  Often with longer prompts than I use in the video prompt because a lot of the work is already done.

All AI models will go off at a tangent unless you tie them down.  That means telling them what you want in different ways.  Just as important is to tell them what you don’t want.