The Problem With Long Prompts
Most failed generations come from prompts that try to do too much. A 200-word paragraph describing the lighting, the mood, the camera, the action, the wardrobe, and the soundtrack leaves the model guessing about what matters most. The result is usually muddy: nothing is wrong, but nothing is right either.
The fix is structure. Replace paragraphs with a small, repeatable template.
The 4-Line Prompt Template
- Subject: Who or what is in the frame. One line. Be specific (a woman in a red jacket, not "a person").
- Action: What they are doing. One verb or a short phrase.
- Setting: Where this happens. Time of day, weather, location.
- Style: The visual feel. Cinematic, handheld, documentary, anime — whatever the reference conveys.
Example
Reference clip: a chef plating a dish in a busy kitchen.
- Subject: a young chef with rolled sleeves
- Action: carefully plates a piece of seared fish
- Setting: open kitchen at dinner rush, warm lighting
- Style: cinematic handheld, shallow depth of field
Combine into one prompt: "A young chef with rolled sleeves carefully plates a piece of seared fish in an open kitchen at dinner rush, warm lighting, cinematic handheld, shallow depth of field."
Why This Works
- The model gets a clear hierarchy: who, what, where, how it looks.
- Each line is short enough that nothing competes for attention.
- It maps cleanly onto negative prompts (just negate each line).
- You can A/B test by changing one line at a time.
Iterating Without Rewriting Everything
When the result is wrong, change one line, not the whole prompt. If the subject looks right but the lighting is off, only edit the setting line. This isolates the variable and tells you what is actually driving the model.
Pair It With a Negative Prompt
Once the four lines are stable, layer a short negative prompt on top. Defaults from the app cover the basics; add only what is specific to your scene (no watermark, no subtitles, no extras in the background).
Save Your Wins
When a prompt produces a clip you would actually post, save the exact text. Over time you build a small library of "known good" prompts for recurring subjects. That library becomes your competitive advantage: every new generation starts from a winning baseline.