Mastering Text-to-Video: 5 Tips for Better Prompts
Writing a prompt shouldn't feel like coding. It’s more like setting a stage. If you want your AI videos to look professional and intentional, follow these five simple "golden rules."
1. The PPA Rule: Person, Place, Action
The biggest mistake I see (and one I made for a long time) is being way too brief. If you just type "a man talking," the AI is basically guessing the rest, and it usually guesses wrong. I always try to hit these three specific points to anchor the scene:
- The Subject: Who are they? Instead of "a man," try "a weathered Viking with a braided beard" or "a futuristic DJ with neon hair."
- The Setting: Where is this happening? Is it a "rain-slicked back alley" or a "library at sunset"?
- The Vibe: What’s the mood? Are they "whispering a dark secret" or "laughing uncontrollably"?
2. Talk Like a Camera Operator
AI models like Hedra actually speak "cinematographer." If you want that high-end movie look, you have to use the right terminology. Don't just describe the person; describe the light and the lens.
- Lighting: Mention things like "golden hour," "moody shadows," or even "soft candlelight."
- The Angle: Try specifying a "close-up" for raw emotion or a "low angle" to make a character look powerful.
- The "Glass" Using terms like "bokeh" or an "85mm lens" tells the AI to blur the background and focus on the face, which instantly makes it look like a pro shot.
3. Give Your Characters Some "Texture"
One of the biggest giveaways of AI video is that "plastic" or overly smooth look. To fix this, you have to get specific about the physical surfaces in your scene. Instead of just saying "a person," talk to the AI about the grit and the materials. This is what helps models like Character-3 move past that artificial sheen and into something that feels grounded.
- For the Skin: Don’t let it be perfect. Mention things like subtle pores, weathered skin, or even a few freckles. These tiny "imperfections" are actually what make it look real.
- For the Wardrobe: Instead of just "clothes," name the fabric. Is it shimmering silk, tattered wool, or heavy velvet?
- The Fine Details: While adding "8k resolution" can help a bit, you’ll get much better results by describing the reflection in the eyes or the way stray hairs catch the light. It’s those small, physical details that really sell the realism.
4. Direct the Emotion (Don’t Just Hand Over a Script)
The real magic happens in the facial expressions. If you just give the AI a line of dialogue, you’re likely to get a "blank stare" that feels hollow. You have to tell the prompt exactly how the character is feeling while they speak.
- Examples: Instead of just "talking," try prompting for a "sinister smirk," "wide-eyed excitement," or someone who looks "visibly exhausted and sighing."
- Why it works: This prevents the "blank stare" and makes the character feel alive.
5. Use "Negative Prompting" (What You DON'T Want)
Sometimes the best way to get a good result is to tell the AI what to stay away from. In your mind, filter out the clutter.
- If you want a clean look, add: "No blur, no distorted limbs, no messy background, no cartoonish features."
- If you want realism, specify: "No 3D render look, no plastic skin."
| # | Development Phase | Implementation Details |
|---|---|---|
| 01 | Prompt Engineering | Crafting descriptive natural language prompts to guide the AI's understanding of scene composition and lighting. Semantic Mapping |
| 02 | Temporal Consistency | Utilizing Wan 2.2's flow-matching to ensure objects maintain their shape and identity across every frame. Motion Logic |
| 03 | Resolution Scaling | Native synthesis at 720p followed by 1080p latent upscaling to achieve cinematic premiere quality. 4K Refinement |
| 04 | Negative Prompting | Excluding unwanted artifacts like "morphing," "blur," or "unnatural limbs" to refine the final visual output. |
| 05 | Frame Distillation | Applying higher sampling steps to reduce noise and increase the overall sharpness of moving subjects. |
6. The "Golden Rule" of Details: Show, Don't Just Tell
Instead of using boring adjectives like "good" or "pretty," describe the textures. The AI loves to know what things feel like because that tells it how to reflect light.
- Instead of: "A man in a suit."
- Try: "A man in a sharp charcoal wool suit with a silk tie."
- Why it works: Hedra’s engine will render the subtle fuzz of the wool differently than the shiny reflection of the silk. This makes the video look "expensive" and high-end rather than a flat cartoon.
7. Control the "Vibe" with Lighting
In film, lighting is 90% of the mood. You don't need to be a photographer to get this right. Just think about the source of the light:
- For a Hero: Use "Strong side-lighting" or "Cinematic rim lighting" (this puts a glowing outline around the character).
- For a Mystery: Use "High contrast, deep shadows" or "Flickering candlelight."
- For a Product/Ad: Use "Soft studio lighting" or "Bright morning sun through a window."
- Human Tip: If your video looks "flat" or boring, just add "Golden hour glow" to your prompt. It fixes almost everything.
8. Give Your Character a "Personality" (The Expression Secret)
Hedra is famous for its Character-3 model, which is incredibly expressive. If you just say "speaking," the face might look a bit stiff. You need to give the character an attitude.
- The Smirk: "Speaking with a confident, knowing smirk."
- The Concern: "Brows slightly furrowed, speaking with a worried whisper."
- The Hype: "Eyes wide with excitement, speaking quickly and gesturing."
- Human Tip: Think about how you would look saying the lines. If the script is sad, tell the prompt the character has "watery eyes."
9. Don't Forget the "Negative Space" (The Background)
A common mistake is focusing 100% on the person and 0% on where they are. This leads to messy, blurred backgrounds.
- Be Specific: Instead of "in a room," try "in a minimalist marble lobby" or "in a cluttered 1980s garage."
- Depth of Field: If you want the character to "pop," add the phrase "Blurred background" or "Deep bokeh. " This tells the AI to focus only on the face, making it look like it was shot on a $5,000 camera.
10. The "Iterative" Secret: Don't Settle for the First Try
Pro creators rarely use their first generation. AI is a bit like a slot machine—sometimes you get a "near-miss."
- The Tweak: If the hair looks weird, keep the prompt the same but add "Slicked back hair."
- The Reroll: If you like the character but the lighting is off, just hit generate again. Small changes in the AI's "seed" can lead to a much better result.
- Human Tip: Use the "Preview" feature if available. It saves you time by showing you a thumbnail of the character before you commit to the full video generation.
Mastering Text-to-Video
Unlock the full potential of Wan 2.2 with advanced prompting and control techniques.
While the model can handle long descriptions, the most effective prompts are between 40 and 75 words. This provides enough detail for lighting and texture without overwhelming the model's primary focus on the main action.
Use technical lighting terminology at the end of your prompt. Terms like "Rim lighting," "Volumetric fog," "Cyberpunk neon palette," or "High-key studio lighting" help the 3D VAE calculate how light should bounce off surfaces in the scene.
Yes. To achieve slow motion, use keywords like "Cinematic slow-motion" or "Time-lapse" for the opposite effect. Adding "stable" or "fluid" helps the model maintain 60fps-like smoothness.
The best way is to use a "Character Anchor". Give your subject a unique physical trait (e.g., "A man with a silver mechanical left eye") and repeat that exact phrase in every prompt to keep the neural network focused.
Use Negative Prompting or simplify your action verbs. Instead of "a cat flying while eating," try focusing on one primary movement like "a cat floating gracefully" to keep the physics stable.
The Motion Bucket value (1-255) controls movement intensity. A low value (60) is for subtle portraits, while a high value (180+) is necessary for explosive cinematic action.
Absolutely. Using "85mm lens" will result in beautiful bokeh for portraits, while "14mm wide-angle" will create an expansive perspective ideal for landscapes.
Include descriptors like "OLED black," "high-contrast lighting," and "Chiaroscuro style." This tells the AI to prioritize deep shadow fidelity.
Use words like "Iridescent," "Matte," "Translucent," or "Reflective Chrome." These prompts allow the AI to calculate more realistic surface shaders.
Add "floating dust motes," "volumetric haze," or "misty atmosphere" to your prompt. This gives the camera a sense of physical space and 3D depth.
Ready to try Hedra?
Transform your ideas into cinematic video in seconds.