Generate video ignores my prompt and does the opposite

Hi Ben22851644n6m2,
Thanks for bringing this to our attention.

Firefly can definitely feel unpredictable at times, and it’s not you; the AI does make mistakes. Yet, there are a few things that often trip people up, let me explain how to work around them:

• Negative prompting: Firefly doesn’t support it at all. So if you say “no facial features” or “don’t add text,” it will usually ignore that. Instead, flip the phrasing into what you do want. For example: “stylized characters with simplified shapes and clean lines” works much better than “no facial features.”

• Vague descriptors: Words like “beautiful,” “cool,” or “nice” don’t give the model much to latch onto. Try to be concrete, specify colors, shapes, moods, or styles.

• Conflicting style cues: If you ask for “realistic watercolor digital sketch,” the model gets confused because those styles contradict each other. Stick to one clear style direction.

• Model choice: One thing that makes a big difference is picking the right model. For example, if you’re working on vector illustrations, Text to Vector models tends to handle simplified, clean requests better than the Text to image model.

The golden rule with Firefly is: tell it exactly what you want, not what you don’t want. That shift alone usually makes prompts much more reliable.
 

If you wish, you can share the exact prompt you’ve been using with us, and we’ll test it on our side and show you a tweaked version that’s more likely to give the results you’re after. 

Let us know.

Hajar
[Moved to questions]

You could have firefly actually follow the prompts, and not add in unwanted effects that the user never asked for. Just an idea.
Doing what the prompt asks, not adding in other random effects and changes.  That would really help.