In image-to-video, the source frame is already a bundle of decisions: subject, styling, color, lens perspective, background, object placement, and often brand detail. The prompt should not redesign all of that. Its main job is to direct change over time.
The most reliable workflow begins by judging whether the image can support the requested motion. Then you establish a motion hierarchy, choose a camera plan, decide whether one frame or two boundary frames are appropriate, and review the result against the source.
Open the Wan 3.0 image-to-video workspace when your source is ready.
First question: should this image move?
Not every attractive still is a good motion source. A source image may look polished while hiding problems that become obvious as soon as the scene changes.
Inspect five areas before writing a prompt.
1. Subject readability
The main subject should be easy to identify at the delivery size. Check the silhouette, face or product edges, clothing boundaries, and contact with the ground or supporting surface.
If the subject blends into the background, a generation may have to invent where one ends and the other begins.
2. Occlusion and missing information
Ask what the requested motion will reveal. A three-quarter turn may expose an ear, shoulder, product side, or room area that the source never shows. The larger the reveal, the more invention the model must perform.
Use modest movement when the hidden geometry is important. If the unseen side must be accurate, provide a more suitable source or consider a reference-led workflow with additional approved views.
3. Pose and balance
Look for ambiguous hands, crossed limbs, cropped joints, unsupported objects, or a body position that cannot naturally continue. A source can freeze an impossible moment that still appears plausible as a photograph; motion exposes it.
4. Background structure
Repeated railings, shelves, windows, text, tiles, and crowds create many continuity obligations. Decide whether they are essential. If the subject is the priority, a cleaner background often gives the motion more room to remain readable.
5. Delivery composition
Choose the final aspect ratio before generation. Cropping a wide source into a vertical output can remove the very context the motion depends on. Leave visual space in the direction the subject or camera will travel.
Choose the frame-control mode
Wan 3.0 supports both first-frame and first/last-frame image-to-video workflows.
Use a first frame when
- the opening appearance matters but the ending can emerge naturally;
- the motion is subtle or exploratory;
- you want to test several actions from one approved still;
- the exact final composition is not a requirement.
Use first and last frames when
- the clip must travel between two approved visual states;
- a product needs a specific closed and open configuration;
- a transition must arrive at a defined layout;
- you are designing adjacent clips that need compatible boundary frames.
Two frames add control, but they also add a feasibility test. The subject, perspective, lighting, and scene geometry must have a believable path between them. A front-facing portrait and a distant overhead city view do not become a coherent transition simply because both images look good.
Write a motion hierarchy
Separate motion into four layers before turning it into prose.
| Layer | Question | Example |
|---|---|---|
| Primary subject motion | What is the one action the viewer must read? | The designer rotates the watch once to catch the light |
| Secondary motion | What supports the main action? | Sleeve fabric settles as the wrist turns |
| Environmental motion | What makes the space feel alive? | A narrow reflection moves across the stone counter |
| Camera behavior | How does the viewer’s position change? | Slow 10-degree orbit, then hold |
If all four layers are “fast,” the scene has no hierarchy. Keep one dominant motion and let the other layers support it.
Use a motion-first prompt
Suppose the source image shows a watch designer seated at a stone workbench, holding a finished watch near a window.
Weak prompt:
A cinematic luxury watch advertisement, beautiful studio, professional designer, dramatic lighting, premium look.
Most of that information is already in the image, and none of it says what moves.
Stronger prompt:
Preserve the designer’s face, charcoal shirt, hand placement, watch geometry, stone workbench, and cool window light from the source image. The designer slowly turns the wrist once so a narrow highlight travels across the watch crystal, then steadies the watch toward camera. Fingers maintain a natural grip; the free hand remains resting on the bench. The shirt cuff settles slightly after the turn. Make a gentle 10-degree clockwise camera orbit at the same height, then stop. Keep the background tools soft and stationary. No dialogue; quiet room tone and one subtle metal clasp sound. End with the watch face readable and hold the composition for one second.
The prompt does four useful things:
- identifies what should remain stable;
- gives the subject one physical action;
- limits camera travel;
- defines a reviewable ending.
Describe preservation selectively
“Keep everything exactly the same” is too broad to prioritize. Name the elements that would make the result unusable if they changed:
- facial proportions and hairstyle;
- product shape, logo placement, and material;
- wardrobe color and distinctive construction;
- room geometry or a critical background object;
- the original palette and light direction.
Then allow less important areas to breathe. Motion requires new pixels. The goal is not literal frame duplication; it is to preserve the approved identity of the shot while directing a believable change.
Match motion to available evidence
A single image provides appearance evidence, not a full performance model. Scale the prompt accordingly.
Lower-risk motion
- breathing, blinking, a small gaze change;
- a controlled hand or wrist movement;
- light fabric, hair, steam, water, or foliage response;
- a slow camera push or modest lateral move;
- a product part moving along an obvious hinge or track.
Higher-invention motion
- a full body turn from a tight portrait;
- hands manipulating an object hidden by the initial pose;
- a rapid orbit around an asymmetrical product;
- a character leaving frame and returning from another angle;
- a major lighting, wardrobe, location, and action change at once.
Higher-invention motion is not forbidden. It simply demands more tolerance for interpretation or more source information.
Keep camera movement proportional
The camera can reveal information the source does not contain. A slow push mostly enlarges existing evidence. An orbit invents new sides. A crane move can expose floor and ceiling geometry. A handheld run can alter perspective every frame.
Start with the smallest move that serves the shot. Increase camera travel only after the subject motion is stable.
Useful directions include:
- “fixed camera; preserve the original composition”;
- “slow push-in with no change in camera height”;
- “track left at the subject’s pace; keep distance constant”;
- “small clockwise orbit, no more than a three-quarter view”;
- “camera remains still until the action finishes, then pulls back slightly.”
Plan sound around visible events
If audio is enabled, tie sound to what the viewer can see:
- a latch should sound when it closes, not before;
- footsteps need the floor material and pace;
- dialogue needs a named speaker, exact line, language, and delivery;
- ambience should establish the room without masking the primary effect;
- music should have an entrance and function.
For subtle image-to-video work, silence or restrained room tone can protect the visual idea better than a full soundtrack.
Diagnose drift instead of adding more adjectives
The face changes during motion
Reduce head rotation, camera orbit, speed, and simultaneous expression changes. Preserve the specific facial traits that matter and keep the first revision focused on pose range.
The product deforms
Simplify hand contact and rotation. Name the geometry, material, logo position, and moving mechanism that must remain stable. Consider a fixed camera before changing the product description.
The background bends or flickers
Reduce camera travel and environmental motion. Identify the structural lines that need to remain straight. If the background is not important, keep it soft and still.
The motion feels frozen
Replace mood words with a path, speed, and visible effect: what moves, in which direction, over how much distance, and what responds afterward.
The scene ignores the source style
Remove conflicting style instructions. Ask to preserve the source palette, light direction, texture, and framing instead of introducing a new genre at the same time as motion.
The ending feels accidental
Describe a final pose, object position, gaze direction, or composition and request a short hold. For strict endpoints, use first/last-frame mode with compatible images.
Run a controlled first test
For the first generation:
- keep one primary action;
- use a conservative camera move;
- avoid a simultaneous location or wardrobe transformation;
- select a duration that lets the action finish without filler;
- save the source, prompt, settings, and output together.
Review the result next to the source rather than from memory. Check silhouette, identity, product edges, important text, palette, background geometry, and the final state.
Source-to-output checklist
- The source has a readable subject and plausible pose.
- Requested movement does not require excessive hidden geometry.
- The chosen aspect ratio protects the important composition.
- Primary, secondary, environmental, and camera motion are separated.
- Preservation instructions name only critical elements.
- Camera travel is proportional to the available visual evidence.
- Audio corresponds to visible events.
- The final state is defined.
- Revision notes identify the first time and location where drift begins.
When the frame passes this preflight, continue in Wan 3.0 image-to-video. If you need several assets to control identity, movement, or setting, move to the reference-to-video guide instead of forcing one image to do every job.