Wan 2.7 Video Guide: Prompts, Cost, and Workflows

Wan 2.7 Video is best treated as a shot generator, not a one-click finished-film machine. The practical advantage is control: give it one visible subject, one readable action, one camera instruction, and a clear reference frame, then judge whether the motion survives from the opening frame to the final frame. That disciplined approach produces more usable footage than asking for a complete advertisement, three scenes, dialogue, a logo reveal, and a call to action in one prompt.

This guide explains the production decisions that matter: when to choose Wan 2.7, how to budget by seconds and resolution, how to prompt motion without over-directing it, and how to build a repeatable approval loop. The current Cliprise catalog lists Wan 2.7 Video at 32 credits per second for 720p and 48 credits per second for 1080p. Those values were verified against the live pricing data on October 18, 2026; use the pricing page as the final check before a large batch.

If you are new to the medium, start with the broader AI video generation guide. If you already have a still image and need to animate it, the image-to-video guide explains the surrounding workflow.

Wan 2.7 in one minute

The model belongs in a multi-model workflow because no generator wins every shot type. Wan 2.7 is a sensible first route for image-led motion, physical movement, product reveals, environmental shots, and deliberate camera work. For native dialogue or sound-led generation, compare it with the options in the AI video with audio guide. For multi-reference production, Seedance 2.0 may be a better fit.

What makes a usable Wan 2.7 shot

A visually attractive frame is not enough. A production-ready shot must pass four independent checks:

1. Frame integrity: the subject, product, hands, faces, labels, and background structure remain believable.
2. Motion integrity: the requested action has a clear start, path, and stop instead of melting into unrelated movement.
3. Camera integrity: the camera move does not fight the subject action or unexpectedly reframe the scene.
4. Edit integrity: the opening and closing frames give an editor a clean place to cut.

Evaluate those criteria separately. A take can have beautiful texture but fail because the product rotates into impossible geometry. Another can preserve the subject perfectly but remain too static for the intended placement. Marking only “good” or “bad” hides the reason a generation failed and leads to random prompt rewrites.

The single-shot rule

Write the brief as something that a camera crew could capture without a cut. “A runner ties a shoe, stands, sprints through a rainy street, enters a stadium, and raises a trophy” contains several scenes and time jumps. A better first shot is: “Low tracking shot beside a runner accelerating on a wet city street; shoes strike shallow puddles; storefront reflections stretch across the pavement; camera maintains a constant two-meter distance.”

The second version describes visible motion and a camera relationship. It does not ask the model to infer an edit.

Wan 2.7 cost planning

Wan 2.7 pricing scales with duration and resolution. At the verified rates, the planning formula is simple:

estimated credits = duration in seconds x credits per second

The table is a budget estimate, not a promise that every requested duration or mode is exposed in every interface. Check the current controls before production. The useful lesson is that an unproven 1080p prompt costs 50% more than the equivalent 720p test. A team that explores at 720p and promotes only accepted shots to 1080p preserves more budget for meaningful variations.

A 12-generation test budget

For a new concept, plan three rounds of four variations:

Do not generate twelve unrelated prompts. The purpose of a test matrix is to learn which variable caused the improvement. Keep a simple record of prompt version, input asset, resolution, result URL, pass/fail reason, and selected take. The seed and consistency guide offers a useful framework for preserving repeatable creative decisions even when a model does not expose every control in the same way.

Prompt structure that controls motion

Use six blocks in this order:

1. Subject: who or what must remain stable.
2. Action: one visible verb with direction and speed.
3. Environment: only details that can affect the shot.
4. Camera: position, movement, and framing.
5. Light and finish: time of day, contrast, texture, palette.
6. Constraints: what must remain unchanged or absent.

Reusable formula

[Framing] of [specific subject] [single action with direction and pace] in [environment]. The camera [one movement] while maintaining [distance or composition]. [Lighting and visual finish]. Keep [identity, geometry, clothing, product details] unchanged; no cuts, no extra subjects, no text.

Product reveal prompt

Tight three-quarter shot of a matte black running shoe resting on wet stone. The shoe rotates slowly clockwise by thirty degrees as fine droplets slide from the sole. The camera performs a subtle six-inch push-in while keeping the side logo centered. Cool dawn light with one warm rim reflection, premium sports-commercial finish. Preserve the exact shoe shape, lace pattern, sole geometry, and logo placement; no hands, no added text, no scene transition.

Why it works: the prompt separates object motion from camera motion, defines a limited rotation, and states which product attributes cannot drift.

Character motion prompt

Medium full-body shot of one cyclist walking a bicycle through a narrow night market from left to right at a relaxed pace. The camera tracks parallel at waist height, maintaining the same distance. Steam from food stalls crosses the foreground; practical red and white lights reflect on damp pavement. Natural documentary movement, restrained contrast. Keep the rider's face, jacket, bicycle frame, and bag consistent; no crowd crossing the subject, no cuts.

Environmental prompt

Wide locked-off shot of a glass cabin beside a mountain lake at blue hour. Wind moves the tall grass in gentle waves from right to left while low fog drifts across the far shoreline. Interior lamps gradually brighten without changing the exterior exposure. Realistic landscape cinematography, clean shadows, subtle film grain. No people, no camera movement, no weather transition.

What not to write

Avoid adjective piles such as “epic, stunning, viral, premium, cinematic, gorgeous.” They do not explain what the pixels should do. Replace them with observable direction: lens distance, subject path, camera path, light source, surface response, and duration of action.

Avoid contradictory camera instructions. “Locked camera, dramatic orbit, handheld energy” asks for three incompatible behaviors. Choose one camera idea per shot.

Text-to-video workflow

Text-to-video is useful before art direction is locked. It explores locations, casting archetypes, compositions, and motion concepts without requiring a first frame.

Step 1: write the acceptance test

Before the prompt, write one sentence that defines success: “The bottle remains readable while the camera moves from profile to three-quarter view.” This protects the evaluation from being distracted by attractive but irrelevant details.

Step 2: generate composition variants

Keep action simple. Test wide, medium, and close framing. If the subject or environment is wrong, do not add more motion language yet. Choose the composition that gives the action enough physical space.

Step 3: add one motion relationship

Specify how camera and subject move relative to each other: parallel tracking, static observation, slow push-in, pull-back, orbit, tilt, or crane. Include distance or framing continuity when it matters.

Step 4: add finish last

Lighting and grade can make a weak take look exciting, so add them after structure and motion work. This prevents a visually polished failure from winning the review.

Image-to-video workflow

Image-to-video is the safer route when the subject is already approved. The source image becomes a contract for composition, identity, wardrobe, product geometry, and lighting.

Prepare the first frame

• Use the target aspect ratio before upload. Do not rely on the model to invent important space outside a crop.
• Remove tiny text unless it must be visible. Generated motion can distort fine typography.
• Leave physical room in the direction of travel.
• Avoid overlapping hands, accessories, or edges that make depth ambiguous.
• Use high-quality source detail, but do not oversharpen skin or surfaces.

The first-frame design guide explains how to design a still specifically for animation. If a source needs cleanup first, use the AI image enhancement workflow.

Describe change, not the image

When a reference already shows a red sports car in a tunnel, the prompt does not need a paragraph repeating those facts. Spend prompt space on what changes: “The car accelerates toward the tunnel exit; overhead lights sweep across the hood; camera follows at a fixed low distance; retain paint color and body geometry.”

Protect the frame edges

Unexpected objects often enter from outside the image. If clean negative space matters, say so: “No people or objects enter from either edge.” For product shots, add “the product stays fully inside frame throughout.”

Choosing Wan 2.7 against other models

This is a routing table, not a permanent leaderboard. Compare the same brief, duration, and input asset. The AI video models ranking gives a broader selection method, while multi-model workflows explains when switching models is more efficient than endlessly repairing one prompt.

Five production recipes

Ecommerce product loop

Start with a clean three-quarter product image. Request a small object rotation or camera push, not both at full intensity. Keep the background stable. Select a take whose end frame can cut back to the opening frame, then finish the loop in an editor. Do not ask the generator to add price, logo, or offer copy; overlay exact text afterward.

Paid social hook

Design the first frame to communicate the subject without sound. Use a single movement in the first second, such as a fast foreground reveal or sudden subject action, then settle into readable motion. Crop and caption for each placement using the aspect-ratio guide.

Fashion motion test

Use a full-body reference with visible hands and space around fabric. Request one action, such as a half-turn or three steps toward camera. Avoid combining walking, spinning, hair motion, fabric motion, and an orbit until identity stability is proven. The AI fashion video workflow provides the surrounding editorial pipeline.

Establishing shot

Use a wide frame with one environmental motion pattern. A slow camera move plus wind, fog, water, or traffic is usually enough. Keep human subjects small or absent if the shot exists mainly to establish place.

B-roll insert

Generate a close detail that can hide an edit: fingers adjusting a control, ingredients falling, light moving across packaging, or machinery cycling. B-roll succeeds when it is specific and brief. For a complete shot list, use the AI B-roll workflow.

Failure diagnosis

Revise one causal variable at a time. Adding five negative phrases to every failure often creates a longer prompt without clarifying the intended action.

Quality-control scorecard

Score every candidate from 0 to 2 on each line. A take scoring below 16 out of 20 should not move to final finishing.

Archive rejected takes with a reason. A failure library becomes valuable: it reveals which reference poses, camera combinations, and prompt clauses repeatedly waste credits.

Use Wan 2.7 when

• You can describe the deliverable as one continuous shot.
• A first frame carries important identity or product constraints.
• Camera movement and subject movement can be stated independently.
• The team can test at 720p before final output.
• The final edit will add exact typography, logos, audio, and calls to action.

Avoid Wan 2.7 when

• The request depends on several scenes or a complete narrative edit.
• Exact in-frame text is the primary success criterion.
• Native dialogue, music, or sound effects must drive the generation.
• The source image has ambiguous anatomy, heavy occlusion, or the wrong crop.
• Stakeholders expect every variation to preserve a person with forensic identity accuracy.

Final workflow checklist

1. Define one shot and one acceptance test.
2. Choose text-to-video for exploration or image-to-video for control.
3. Prepare the correct aspect ratio and clean frame edges.
4. Write subject, action, environment, camera, finish, and constraints.
5. Test composition at 720p.
6. Test motion while keeping the selected composition fixed.
7. Score identity, geometry, motion, camera, and edit points.
8. Promote only the approved prompt and input to 1080p.
9. Add exact text, audio, grade, and brand elements in post-production.
10. Save the winning prompt, source asset, and review notes for reuse.

Wan 2.7 becomes predictable when the unit of work is a shot, the budget is staged, and evaluation is specific. Start the next test in the Cliprise video generator, then route difficult shots to a different model instead of turning one prompt into an unreadable list of repairs.