Advanced Prompting Techniques for AI Video Generation

The difference between a mediocre AI video and a stunning one almost always comes down to the prompt. The same model that produces generic, flat output from a vague prompt can generate cinematic, detailed footage when given the right instructions. This guide covers advanced prompting techniques that work across all models on Kensa, along with model-specific tips that help you get the best results from Sora 2, Veo 3.1, Wan 2.6, and Seedance 1.5.

The Anatomy of an Effective Prompt

Every strong video prompt contains a combination of these core elements. You do not need all of them in every prompt, but understanding each component gives you more control over the output.

1. Subject

What is the primary focus of the video? Be specific about appearance, position, and action.

• Weak: "a person walking"
• Strong: "a young woman in a tailored navy overcoat walking along a cobblestone street, carrying a leather messenger bag"

2. Setting and Environment

Where does the scene take place? Include physical location, time of day, weather, and atmospheric conditions.

• Weak: "city background"
• Strong: "a narrow European alley lined with terracotta buildings, clotheslines strung between windows, late afternoon shadows cutting across the pavement"

3. Lighting

How is the scene lit? Lighting is one of the most impactful elements you can specify.

• Weak: "good lighting"
• Strong: "warm golden-hour light streaming from the left, casting long soft shadows, slight lens flare where the sun catches the edge of the frame"

4. Camera and Framing

What is the camera doing? Specify the shot type, movement, and lens characteristics.

• Weak: "show the product"
• Strong: "slow tracking shot from left to right, medium close-up, shallow depth of field with f/1.4 bokeh in the background"

5. Style and Mood

What visual style or emotional tone should the video convey?

• Weak: "make it look nice"
• Strong: "moody and atmospheric, muted teal and orange color grade, film grain texture, reminiscent of a Nordic noir film"

6. Motion and Action

What is moving in the scene, and how?

• Weak: "things moving around"
• Strong: "steam rising slowly from a coffee cup, the surface of the coffee rippling gently, a hand reaching into frame from the right to lift the cup"

Camera Movement Vocabulary

AI video models, especially Sora 2, respond well to established cinematography terms. Here is a reference guide for the most useful camera movements.

Static Shots

Moving Shots

Vertical, Rotational, and Zoom

Prompt Example Using Camera Movement

Slow crane shot rising from a close-up of wildflowers to reveal a vast mountain valley at golden hour, mist settling in the lower elevations, warm directional sunlight from the right, cinematic anamorphic lens with horizontal flare.

Lighting Techniques

Lighting sets the mood of your video more than almost any other element. Here are the key lighting conditions to reference in your prompts.

Natural Light Conditions

• Golden hour: Warm, low-angle sunlight that creates long shadows and flattering skin tones. Ideal for lifestyle, fashion, and outdoor product content.
• Blue hour: Cool, diffused light just before sunrise or after sunset. Creates moody, contemplative atmospheres.
• Overcast: Soft, even lighting with minimal shadows. Good for product detail shots where you want uniform illumination.
• Harsh midday sun: Strong, directional light with deep shadows. Creates dramatic contrast but can be unflattering.
• Dappled light: Sunlight filtered through leaves or a lattice. Adds organic texture and visual interest.

Artificial and Stylized Lighting

• Neon lighting: Colored neon reflections on wet surfaces. Creates cyberpunk and urban nightlife aesthetics.
• Rim lighting: Light from behind the subject, creating a glowing outline. Adds depth and separation.
• Volumetric light: Visible light rays through fog, dust, or haze. Adds atmosphere and depth.
• Practical lights: Visible light sources within the scene (lamps, candles, screens). Adds realism.

Prompt Example Using Lighting

Close-up of a glass perfume bottle on a dark reflective surface, volumetric light streaming from behind through a thin haze, creating visible light rays. Rim lighting on the bottle edges produces a prismatic rainbow effect. The background is deep black with subtle blue-purple gradients.

Style References and Visual Language

Referencing established visual styles gives the model a rich context to draw from. Here are categories of style references that work well.

Film and Cinema References

• "In the style of a Wes Anderson film -- symmetrical composition, pastel color palette, centered framing"
• "Christopher Nolan aesthetic -- desaturated colors, IMAX-scale wide shots, dramatic contrast"
• "Wong Kar-wai visual style -- neon-soaked nights, motion blur, saturated greens and reds"

Photography and Art Styles

• "Editorial fashion photography -- clean studio lighting, neutral backdrop, strong posing"
• "Street photography aesthetic -- candid moment, urban environment, black and white with high contrast"
• "Minimalist Scandinavian design -- clean lines, neutral tones, lots of negative space"
• "Japanese wabi-sabi -- embracing imperfection, natural materials, muted earth tones"

Prompt Example Using Style References

A ceramic teapot pouring green tea into a handleless cup, wabi-sabi aesthetic with natural imperfections in the pottery, warm diffused light from a paper screen, muted earth tones, shallow depth of field, slow pace.

Negative Prompting

While not all models support a dedicated negative prompt field, you can include exclusion language directly in your prompts to steer output away from unwanted elements.

Direct exclusion: "No text or watermarks in the frame," "Without any people visible," "Clean background with no clutter."

Quality guards: "Sharp focus throughout, no motion blur," "Stable camera with no jitter," "Consistent lighting with no flickering."

Style boundaries: "Photorealistic, not stylized or cartoonish," "Natural colors, no heavy color grading."

Prompt Example Using Negative Guidance

A minimalist workspace with a laptop, coffee cup, and a small plant on a white desk. Scandinavian style, natural window light from the left, shallow depth of field. No people visible. No text or logos on screen. Photorealistic, not stylized.

Model-Specific Prompting Tips

Each model on Kensa has its own strengths and tendencies. Tailoring your prompts to the specific model significantly improves results.

Sora 2 Tips

Sora 2 is the most cinematically-minded model. It interprets prompts with a filmmaker's sensibility and adds production value to your descriptions.

What works well:

• Cinematography terminology (dolly, crane, tracking, anamorphic)
• Emotional and atmospheric descriptions ("melancholic," "triumphant," "serene")
• Complex scenes with multiple elements interacting
• References to film styles and directors
• Long, detailed prompts with layered descriptions

What to watch for:

• Sora 2 may add creative interpretation you did not request. If you want precise control, be very explicit about what should and should not appear.
• Text rendering is unreliable. Avoid prompts that require legible text on signs, screens, or labels.

Optimal prompt length: 50 to 150 words. Sora 2 benefits from detailed prompts more than most models.

Example:

Slow push-in on a vintage record player in a dimly lit room, warm tungsten lamp light casting amber tones across a wooden desk, vinyl record spinning with visible grooves catching the light, thin trail of cigarette smoke curling upward in the background, shallow depth of field, grain texture reminiscent of 1970s film stock, moody and nostalgic atmosphere.

Veo 3.1 Tips

Veo 3.1 excels at dense visual detail within its shorter time window. It tends to be more literal in its interpretation.

What works well:

• Precise, specific descriptions of materials and textures
• Clear spatial relationships ("on the left," "in the foreground")
• Simple, focused scenes with one clear subject
• Technical photography terms (aperture, focal length, exposure)
• Short, punchy prompts that get to the point

What to watch for:

• Overly complex scenes with too many elements may lose focus in the short duration
• Abstract or poetic prompts may produce less predictable results compared to Sora 2

Optimal prompt length: 30 to 80 words. Veo 3.1 performs well with concise, focused prompts.

Example:

Extreme close-up of water droplets hitting a bright red rose petal in slow motion, each drop creating tiny concentric ripples, studio lighting from above, pure black background, crystal-clear detail on every water bead, f/2.8 macro lens perspective.

Wan 2.6 Tips

Wan 2.6 offers the most flexibility in terms of aspect ratios and duration, making it the workhorse for multi-format content.

What works well:

• Straightforward scene descriptions
• Clear subject-action-setting structure
• Consistent results across different aspect ratios
• Good response to color palette descriptions
• Effective for batch generation with template-style prompts

What to watch for:

• Very artistic or abstract prompts may produce more generic results compared to Sora 2
• For maximum quality, keep scenes focused rather than overly complex

Optimal prompt length: 30 to 100 words. Wan 2.6 handles mid-length prompts well.

Example:

A cup of matcha latte on a light oak table, café setting with blurred background, steam rising gently from the surface, natural daylight from a large window on the right, green-and-cream color palette, overhead camera slowly descending toward the cup, clean and minimal aesthetic.

Seedance 1.5 Tips

Seedance 1.5 is particularly strong with motion-heavy content and offers multi-resolution output from 480p to 1080p.

What works well:

• Motion-focused prompts (dancing, sports, flowing materials)
• Clear descriptions of the type and quality of movement
• Specifying resolution when you have a specific output target
• Action sequences with dynamic camera work

What to watch for:

• Static scenes with minimal motion may not leverage the model's strengths
• For the highest quality, specify 1080p when resolution matters

Optimal prompt length: 30 to 100 words. Focus on describing movement clearly.

Example:

A silk ribbon dancer performing in a spotlight, deep black background, the fabric catching light as it spirals through the air in fluid arcs, slow motion at moments of peak extension, camera orbiting slowly around the dancer, dramatic side lighting creating sharp highlights on the fabric folds.

Prompt Templates

Here are ready-to-use prompt templates for common use cases. Replace the bracketed sections with your specific details.

Product Showcase Template

[Camera movement] of [product] on [surface/background], [lighting description], [product action or motion], [atmosphere/mood], [aspect ratio context if relevant]. No text or logos visible. Sharp focus on product details.

Filled example: Slow orbit around a matte black wireless speaker on a polished concrete surface, soft directional studio light from the upper right creating a subtle shadow, the speaker gently rotating to reveal all sides, clean minimalist aesthetic, premium feel.

Lifestyle Scene Template

[Subject description] in [setting], [time of day] light, [specific actions], [emotional tone], [camera movement], [visual style reference]. [Negative guidance].

Filled example: A young man reading a hardcover book in a cozy window seat, afternoon golden-hour light through sheer curtains, calm and contemplative mood, slow push-in to close-up on hands and book. No direct eye contact with camera.

Social Media Ad Template

[Product/subject] in [eye-catching setting], [bold visual element], [dynamic action], [brand-appropriate style]. Clean composition, no clutter.

Filled example: A bright orange running shoe landing on a rain-soaked city street creating a dramatic splash, water droplets suspended mid-air in slow motion, neon city lights reflecting in puddles, high-energy athletic aesthetic.

Advanced Techniques

Layered Prompting

Build your prompt in layers, adding one element at a time. This helps you isolate which descriptions have the most impact on the output. Start with just the subject ("A glass sphere on a reflective surface"), then add lighting, then atmosphere, then camera movement. Each layer gives you more control. If a generation goes wrong, you can identify which layer caused the issue and adjust it specifically.

Temporal Descriptions

For longer clips (especially with Sora 2 at 10 to 15 seconds), you can describe what happens over time.

The scene begins with a tight close-up of closed eyes. After a moment, the eyes open slowly, revealing bright green irises. The camera gradually pulls back to reveal the full face, then continues to widen to show the subject standing on a cliff overlooking the ocean at sunset.

This technique gives the model a narrative arc, resulting in more intentional and dynamic output.

Aspect Ratio-Aware Prompting

Different aspect ratios suit different compositions. Adjust your prompt to match.

For 16:9 (landscape): Emphasize wide scenes, horizontal motion, panoramic views, and side-by-side compositions.

For 9:16 (portrait): Emphasize vertical elements (tall buildings, standing figures, waterfalls), top-to-bottom or bottom-to-top camera movements, and vertically-centered subjects.

Consistency Across a Series

When generating multiple videos that need a cohesive look, create a "style anchor" paragraph that you append to every prompt. For example: "Shot on 35mm film with warm tungsten color grading, shallow depth of field at f/1.8, slight natural grain, clean compositions, premium aesthetic." This maintains visual consistency across an entire campaign.

Common Pitfalls and How to Fix Them

Next Steps

Prompting is a skill that improves with every generation. Start with the templates above, experiment with different techniques, and build a personal library of prompts that work for your specific use cases.

Ready to put these techniques into practice? Open the text-to-video tool or the image-to-video tool on Kensa and start generating. For model-specific guidance, visit the individual model pages for Sora 2, Veo 3.1, Wan 2.6, and Seedance 1.5.

For more context on choosing the right model for your project, read our Veo 3.1 vs Sora 2 comparison and our complete guide to AI video generators in 2026.