Guide
PixVerse V6 Max Text-to-Video Prompt Recipes: Copy-Paste Prompts With Real Outputs
A practical prompt cookbook for developers: shot structure, camera-motion wording, pacing, duration and aspect handling, and negative prompts
On this page
Why Prompt Recipes Beat One-Off Prompts
Every developer who has generated more than a handful of clips knows the pattern: you write a long, elaborate prompt, get something decent, then try to tweak it and the output drifts in a direction you didn't intend. The next iteration loses the lighting you liked. The one after that changes the camera movement. Twenty generations later you're no longer sure which wording produced which result.
A prompt recipe fixes this by treating the prompt as a structured artifact rather than a sentence. Instead of free-form prose, you compose from named blocks — subject, action, environment, camera, lighting, style — in a fixed order. That gives you three concrete advantages:
- Diffability. When every prompt follows the same block order, changing one block is a visible, reviewable change. You can log prompts in version control and compare runs like code.
- Reusability. A good camera block or lighting block carries over to dozens of scenes. You build a small library of fragments and assemble prompts from it.
- Reproducibility of intent. Vague adjectives like "cinematic" or "epic" mean different things on different runs. Structured, specific wording pins down more of the intent so that variance comes from the model, not from your prompt.
This matters more for video than for still images because a clip contains motion over time. A single ambiguous phrase can change the entire feel of a shot — a "slow push in" that becomes a "fast zoom" is a completely different edit. Recipes keep those variables explicit.
The rest of this guide builds that structure block by block, then gives complete copy-paste recipes you can run, verify against real outputs, and adapt. Treat every recipe as a baseline, not a final answer: the goal is a repeatable loop, not a perfect first prompt.
The Anatomy of a Copy-Paste Video Prompt
A reliable text-to-video prompt reads best when its information arrives in a predictable order. The model parses a scene more consistently when the subject is established before the camera, and the camera before the style. Use this six-block order as your default:
1. Subject
Who or what is on screen, described concretely. Prefer physical detail over abstract labels: "a woman in a charcoal wool coat" beats "a stylish woman." If the subject is an object, include material and scale — "a brushed steel kettle, roughly 20 cm tall."
2. Action
What the subject is doing, in the present tense, with a clear start and end state. "Walks from frame left toward frame center and stops" gives the model a trajectory. "Is walking" gives it almost nothing.
3. Environment
Where the action happens: location, foreground and background elements, weather, time of day. This block anchors depth cues, which in turn help the model decide how foreground and background should move relative to each other.
4. Camera
Shot size plus movement. "Medium close-up, static tripod shot" or "wide shot, slow dolly in." Keep camera instructions in their own sentence so they don't get absorbed into the action description.
5. Lighting
Direction, quality, and color temperature. "Low-angle late-afternoon sun, warm highlights, soft shadows" is specific enough to be repeatable. Avoid stacking three lighting metaphors in one phrase.
6. Style
The visual register: photographic realism, documentary, animation, product-video look. Put style last so it colors the whole scene rather than overriding earlier blocks.
Assembling the blocks
Join the blocks in order, one to two sentences each. A filled-in template looks like this:
Subject: a ceramic mug on a wooden desk, steam rising
Action: the steam curls upward and slowly dissipates
Environment: a quiet morning kitchen, blurred window in background
Camera: close-up, static shot, shallow depth of field
Lighting: soft window light from the left, neutral white balance
Style: clean product photography, photorealistic
You can paste the labeled version directly, or flatten it into prose in the same order. Both work; the labeled form is easier to diff when you iterate.
Camera Motion and Pacing Wording That Actually Works
Camera language is where vague prompts fail most visibly. "Dynamic camera" is not a shot; it is a wish. The model responds much more consistently to established motion vocabulary paired with a speed qualifier.
Movement vocabulary
- Dolly in / dolly out — the camera physically moves toward or away from the subject. Use for building or releasing intimacy.
- Pan left / pan right — rotation on a fixed point. Use to reveal environment.
- Tilt up / tilt down — vertical rotation. Use to reveal height or scale.
- Orbit — the camera circles the subject. Specify direction and arc, e.g. "slow orbit clockwise, roughly 45 degrees."
- Tracking shot — the camera follows a moving subject. State whether it leads, trails, or moves alongside.
- Zoom — change of focal length without physical movement. Keep zoom and dolly distinct in your wording; mixing them in one phrase produces mush.
Speed and pacing words
Attach exactly one pacing qualifier per camera instruction: slow, steady, gentle, brisk, or fast. Avoid piling up adverbs — "very slow, smooth, gradual push" is three ways of saying one thing and tends to flatten the motion.
Pacing also applies to the subject. Phrases like "over the course of the shot," "in the first half," and "by the final frame" give the model a sense of timing inside the clip. Contrast these two actions:
Weak: the person turns around
Better: the person slowly turns from the window toward the camera,
completing the turn before the clip ends
The second version tells the model where the motion starts and where it should be when the clip finishes. That is the single biggest lever for making short clips feel intentional rather than truncated.
Mixing moves
One primary move per shot is the safe default. If you need two, make one clearly dominant: "primary slow dolly in, with a slight pan right." Two equally weighted moves usually read as camera shake.
Recipes: Shot-Structure Prompts to Copy and Run
Each recipe below follows the six-block order. Paste it as-is, then check the listed signals against your generated clip.
Recipe 1 — Static product close-up
Subject: a matte black wireless earbud case on a stone surface
Action: a soft reflection shifts across the lid as the light changes
Environment: minimal studio set, seamless grey backdrop
Camera: close-up, static tripod shot, shallow depth of field
Lighting: large softbox from the upper left, gentle falloff
Style: clean commercial product video, photorealistic
Look for: whether the case stays the sole sharp subject and whether the reflection moves smoothly rather than flickering.
Recipe 2 — Slow dolly-in reveal
Subject: a figure in a hooded jacket standing at the edge of a rooftop
Action: the figure stays still, jacket fabric moving slightly in the wind
Environment: city skyline at dusk, distant windows beginning to glow
Camera: wide shot, slow dolly in toward the figure
Lighting: cool blue ambient light with warm highlights from the city below
Style: cinematic realism, muted color grade
Look for: whether the dolly reads as physical movement rather than a zoom, and whether the skyline parallax matches the foreground motion.
Recipe 3 — Orbit around a subject
Subject: a vintage motorcycle parked on a gravel path
Action: none, the motorcycle is stationary
Environment: open countryside, low hills in the far background
Camera: medium shot, slow orbit clockwise around the motorcycle, about 45 degrees
Lighting: golden-hour sun, long shadows across the ground
Style: documentary photography, natural color
Look for: consistent geometry on the motorcycle during the arc. Orbits are where object distortion shows up first.
Recipe 4 — Tracking shot with a walker
Subject: a person in a light rain jacket walking a dog on a leash
Action: both walk from frame right toward frame center at a steady pace
Environment: wet city sidewalk, parked bicycles, reflective pavement
Camera: medium-wide tracking shot, moving alongside the subject
Lighting: overcast daylight, soft and even
Style: documentary street photography, slight film grain
Look for: whether the camera keeps pace with the subject and whether the leash stays connected across frames.
Recipe 5 — Handheld-style reaction shot
Subject: a woman in a charcoal wool coat, close on her face
Action: she glances off-camera, then turns her eyes back toward the lens
Environment: a busy station concourse, blurred figures behind her
Camera: close-up, subtle handheld movement
Lighting: warm overhead lamps, soft shadows under the brow
Style: narrative film look, shallow focus
Look for: the timing of the glance. If the turn completes before the clip ends, the pacing wording did its job.
Treat these as starting points and vary exactly one block at a time in your next run.
Handling Duration and Aspect Ratio in Prompts
Duration and aspect ratio are constraints, not style choices, and the model handles them best when they are stated plainly alongside the scene rather than buried in prose.
Stating duration
Add a short clause that names the clip length and connects it to motion pacing:
... the motion completes within a short clip of a few seconds.
Or, more directly, state the target and let the pacing blocks do the rest:
Duration: a short clip, roughly 4 seconds.
Keep the duration clause separate from the action description. The model should read it as a constraint on the whole shot, not as part of the subject's movement.
Stating aspect ratio
Name the framing orientation explicitly, and prefer plain language over ratios when it removes ambiguity:
Aspect: vertical 9:16 framing for mobile.
Aspect: horizontal 16:9 widescreen framing.
If a recipe needs vertical framing, say so in the camera block as well — "vertical framing, subject centered" — because composition and framing are related decisions. A vertical clip of a horizontal tracking shot needs different subject placement.
Verifying against real output
Prompts bias the result, they do not guarantee it. After each run, check three things:
- Does the returned duration match the request? If your pipeline expects a specific clip length, inspect the returned asset's actual duration rather than assuming the prompt enforced it.
- Does the framing match? Confirm the returned frame dimensions, especially when you switch between horizontal and vertical recipes.
- Does the pacing feel right for the length? A motion that completes in four seconds looks frantic in a longer clip and truncated in a shorter one. Adjust the pacing wording, not just the duration clause.
Log duration and framing requests alongside the returned values so drift is visible when you compare runs.
Negative Prompts and Artifact Cleanup
Negative prompts are most useful when they are short and specific. The instinct to paste a wall of generic quality words — "bad quality, low resolution, ugly, deformed" — usually backfires: it dilutes the signal and can push the model away from perfectly fine outputs.
A practical starting list
Keep the list focused on artifacts you have actually observed:
Negative: text overlays, watermarks, logos, extra fingers,
duplicate limbs, warped faces, flickering lighting,
sudden camera shake, jump cuts
That list covers the most common failures in generated video: unwanted text, anatomy errors on hands and faces, and temporal instability. Add to it only when you see the specific problem.
Matching negatives to recipes
Negatives are best scoped per recipe rather than global:
- Product shots: add
reflections jumping, floating object, mismatched shadows. - Orbit shots: add
geometry deforming, object warping during rotation. - Tracking shots: add
disconnected leash, sliding feet, background speed mismatch. - Reaction shots: add
unnatural eye movement, frozen expression.
What to leave out
Avoid stacking generic quality adjectives. If a run produces a soft image, the fix is usually a sharper lighting or style block in the positive prompt, not another negative. Negatives remove specific unwanted content; they do not add quality. Treat them as a cleanup pass, and revisit the list whenever the artifact you're fighting changes.
Iterating on Real Outputs Without Wasting Generations
The fastest way to burn through generations is to change three things at once and then guess which one helped. Treat prompt iteration like a controlled experiment.
The one-variable loop
- Pick a baseline recipe. Save it verbatim, including the labeled blocks.
- Change exactly one block. If the camera motion felt wrong, edit only the camera block. Leave subject, action, environment, lighting, style, and negatives untouched.
- Regenerate and compare side by side. Watch both clips at the same time, focusing on the block you changed.
- Keep or revert. If the change helped, promote it to the baseline. If not, revert and try a different wording in the same block.
- Log the diff. Record the baseline ID, the block changed, the new wording, and a one-line verdict.
What to compare
Compare clips on observable properties, not vibes: did the motion direction match? Did the duration land where requested? Did the framing hold? Did the artifact list change? Write down the answer for each run — memory is unreliable after a dozen clips.
When to abandon a recipe
If two or three single-variable attempts in the same block don't move the result, the problem is likely structural. Maybe the camera move conflicts with the subject's action, or the scene is too busy for the shot size. In that case, change the shot structure — a different camera block or a simpler environment — rather than continuing to tweak adjectives.
Reuse across projects
Once a recipe stabilizes, copy it into a prompt library with its verdict notes. Over time you accumulate proven blocks for specific shot types, and new scenes start from a tested baseline rather than from scratch. That is the payoff of the recipe approach: iteration compounds instead of resetting.
Wiring It Into Your Pipeline
The recipes in this guide are designed to survive the transition from manual playground to programmatic calls. A few practical habits make that easier.
Store prompts as structured data. Keep each recipe as a record with named fields for subject, action, environment, camera, lighting, style, negatives, duration, and aspect. Assemble the final prompt string at call time so you can log the exact text that produced each output.
Pin a model. Use a single model identifier in your config and log it with each generation. The PixVerse V6 Max text-to-video model is documented on its model page, including request parameters — start there when you build the client.
Validate the response, not just the request. After each call, check returned duration and framing against what the recipe asked for, and store the observed values with the prompt record. Discrepancies are your early signal that a recipe needs revision.
Version your recipes. Treat the prompt library like code: commit changes with a message describing the block that changed and the reason. When a recipe stops producing good results after a model update, you can trace exactly what changed.
A minimal call flow looks like: load recipe → assemble prompt → send text-to-video request → store prompt and response metadata → compare output against recipe constraints. Build that loop once and every new recipe slots into it.
The recipes above are baselines, not guarantees — the real value is in the disciplined iteration loop around them.
Related articles
View all articles →-
Guide
gpt-6.2-sol Getting Started Guide: Capability Boundaries, API Integration, and Token Pricing Essentials
Integration decisions and practical essentials for engineering teams
-
Guide
gpt-6.2-astra Getting Started Guide: Text Model Capability Boundaries, Integration Essentials, and Cost Control
-
Guide
Tripo H3.1 Getting Started Guide: Choosing Among Three Input Modes, Parameters, and Pricing
A practical guide for developers on 3D generation integration decisions and cost control