5 Things I Learned From Bad AI Video Generations
AI video prompts can look completely reasonable and still produce a result that is not what you expected. I ran into this while testing a skiing video. The skier moved downhill correctly, the snow reacted to the skis, and the overall scene looked good. The problem was the camera. I wanted it to move in behind the skier, come closer, pass near the subject, and then pull away. Instead, it barely…
1. **Action-Oriented Language Matters**
The key takeaway was to move beyond vague descriptors. A prompt like "A skier moving through the snow" lacks precision. Instead, explicitly describe the action: "The skier accelerates downhill from left to right, leans into a sharp turn, and cuts across the slope." This gives the AI clearer direction about direction, sequence, and timing, resulting in a more dynamic video output.
2. **Specify Camera Movement When It Matters**
Initially, I wrote "dynamic tracking shot" expecting the AI to interpret that as a specific camera movement. However, the AI often filled in the gaps with inadequate or irrelevant camera behavior. After testing, I realized that specifying the camera's exact path significantly improved results. For instance, instead of "dynamic tracking shot," instruct the camera to "approach from behind the skier, move alongside them, close the distance, pass near the subject, then separate as snow bursts across the frame." This level of detail translates into the intended cinematic effect.
3. **Consider Environmental Reactions**
Sometimes, the subject’s movement is technically correct, yet the video feels static. In such cases, the environment’s reaction to the action can make a significant difference. For example, instead of just saying "The skier makes a sharp turn," include the environmental detail: "The skier makes a sharp turn. The skis cut into the snow and throw a burst of powder outward as the skier passes." This extra layer of description brings the scene to life by showing how the action impacts the surroundings.
4. **Leverage Reference Images for Identity**
When creating recurring characters, relying solely on detailed textual prompts can lead to inconsistencies. A reference image can serve as a visual anchor. Initially, I described Carrie’s appearance in exhaustive detail within the prompt, but this often led to variations between generations. Changing my approach, I now use a reference image to establish her identity (front-facing, profile, full-body views) and let the prompt focus on actions like "Carrie walks toward the table, pulls out the chair, sits down, and looks toward the window."
This separation ensures that the character’s visual traits remain consistent while the prompt efficiently conveys the intended actions.
5. **Plan the Shot Sequence in Stages**
Generating a full video as one continuous block can sometimes lead to confusion about the intended shot order and transitions. In one experiment, I attempted to describe a 15-second bookstore sequence all in one paragraph. The result was a lack of clarity in shot transitions and timing. Instead, breaking down the sequence into individual shots—such as "a character enters a bookstore, walks between the shelves, finds a book, opens it, and ends in a quieter final shot"—allowed for clearer direction on both action and shot transitions.
Even if the AI produces the entire sequence in one go, planning the sequence in shots helps identify and correct any logical inconsistencies.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
