WAN 2.2 Image-to-Video Prompt Guide: Direct the Change

By ToolNivo editorial

Try the AI Video Prompt Generator

A WAN 2.2 image-to-video prompt should explain how a supplied image becomes a moving shot. The image establishes visual information; the text directs change. Confusing those roles can produce a long description that says little about what should happen.

The official Wan2.2 repository contains distinct tasks and implementations. This guide concerns the image-to-video writing problem, not installation, hardware recommendations, or a claim that every WAN service uses the same configuration. ToolNivo does not run WAN or provide a dedicated WAN integration; its Universal prompt option can help organize a portable brief.

Verify the actual task you selected

A model family can include multiple tasks. Do not assume that a text-to-video workflow, an image-to-video workflow, and a speech-driven workflow are interchangeable because they share a family name.

Check the selected implementation's input requirements and available controls. If you use a hosted service, its interface may expose only some options. If you work with the official repository, follow its documentation for technical configuration rather than inferring settings from a creative article.

The writing method here remains focused on the shot: starting image, motion, continuity, and ending.

Read the reference like a starting frame

For an example, imagine a photograph of a small red kite resting on a grassy slope with its string visible. Identify what the image establishes: the kite's shape, the slope, the light direction, and the camera viewpoint.

Now identify what the image does not establish. It may not show a person holding the string, the sky above the crop, or the underside of the kite. Asking for a high aerial flight would require substantial invention beyond the visible information.

A modest first action could be the kite lifting slightly in a breeze and settling back. That uses the reference's existing relationships without demanding a new landscape.

Describe motion in relation to the image

Use the supplied red-kite image as the starting composition. A light breeze lifts one edge of the kite briefly above the grass, then lets it settle back onto the slope. Keep the kite's red surface, shape, and visible string consistent. The camera remains at the original viewpoint. Grass moves gently in the same breeze, and the soft daylight stays stable. End with the kite resting again.

This is an original creative example, not a tested WAN result. Its purpose is to make the intended change and return state explicit.

The action also has a physical cause: the breeze. That helps the scene read as one coherent event rather than an unexplained transformation.

Separate small animation from a new scene

If you instead want the kite flying high above a coastline, you are asking for a much larger scene. The original photograph may still communicate the kite's appearance, but it is no longer a strict opening composition.

Choose a workflow and reference plan appropriate to that goal. You might prepare a new starting image with the kite already airborne, then describe its motion. Do not demand both perfect preservation of the ground-level photograph and an immediate unrelated aerial view without acknowledging the transformation.

This distinction is useful for products, portraits, and landscapes as well. A small movement and a complete relocation are different tasks.

Keep camera movement proportional to the available view

A stable camera can help you evaluate subject motion. A gentle approach can reveal texture. A large orbit requires more unseen geometry and background information.

For the kite, a static view is a sensible baseline. If you later test a slight approach, keep the kite fully visible and avoid changing the wind strength at the same time. That lets you compare the camera choice rather than a completely different animation.

If the output invents unwanted areas beyond the reference, reduce the movement or change the input. A longer negative list does not necessarily provide the missing visual information.

Use prompt expansion carefully

The official WAN project documents prompt-related tooling and different execution options. If your chosen workflow expands or rewrites a prompt, inspect the expanded text when available. Added details can change the creative intention.

A simple kite movement should not silently become a dramatic storm with a running child and an aerial chase. If expansion changes the task, simplify or adjust that feature according to the implementation's documentation.

ToolNivo's builder is separate from any WAN prompt-extension system. It does not call that system, host its models, or configure its options.

Write continuity anchors that matter

For this shot, the key anchors are the red kite, the string, and the grassy slope. You do not need to describe every blade of grass. Focus on the objects whose change would make the action confusing.

A continuity sentence is a direction to evaluate, not an exact lock. Watch for changes in the kite's corners, string attachment, or relationship to the ground. If those fail, simplify the movement and consider whether the reference is clear enough.

Do not claim that a particular word guarantees geometry preservation. The actual output remains the evidence.

Plan the output without inventing supported settings

Keep your intended frame shape, duration, and audio plan in project notes. Then set supported options in the implementation you use. The official repository and a hosted wrapper may not present identical choices.

If the task is visual animation, do not assume it produces a soundtrack simply because another model in the family handles audio-related inputs. Verify the specific task. You can always maintain a separate sound-editing brief for later use.

For the kite example, soft wind ambience may be sufficient. It does not need narration to explain the small movement.

Review in a repeatable order

Compare the opening with the reference. Inspect the main movement at its highest point. Check the return to rest. Then look at the background and frame edges for unintended changes.

Write a short observation for each stage. “The string detaches during the lift” is more actionable than “bad physics.” “The kite exits the crop during the approach” points toward framing. These observations guide the next revision.

Keep the same reference and settings while testing a wording change, when practical. If you change several inputs at once, document that the comparison is not prompt-only.

Common mistakes

• Confusing different WAN tasks because they share a model family name.
• Describing appearance at length but omitting the actual movement.
• Requesting large unseen environments from a tightly cropped reference.
• Treating prompt expansion as automatically faithful to the original idea.
• Assuming ToolNivo hosts WAN or supplies a rendering backend.

FAQ

Can I use ToolNivo for a WAN brief?

You can use its Universal option to organize text and notes. There is no dedicated WAN selector or verified direct integration in the inspected version.

Should I always add more detail to an image-to-video prompt?

Add details that clarify motion or preserve important relationships. Repeating the entire image may not address the missing decision.

Does this article provide installation instructions?

No. Use the official repository for technical setup. This guide addresses creative preparation and output review.

Let the reference and text do different jobs

The image communicates the starting appearance. The prompt explains the change. Keep those roles aligned, verify the selected task, and revise from specific observations rather than from a general demand for better quality.