Lance Text to Video was tested with five short prompts built to expose different failure modes: a tracking shot, close hand work, fast animal motion, a small object in a busy frame, and a crowd scene. The goal was not to find one pretty clip. It was to see where a single prompt, one main subject, and a named camera move are enough, and where the scene starts to ask too much of the model.
The five clips already on this page are the evidence. They remain the useful part of the review because each one asks for motion, lighting, and composition at the same time. Lance is available through Lance Text to Video on Wiro. The broader project describes Lance as a unified multimodal model; readers who want the underlying project can consult the official Lance project page, the open-source repository, or the technical paper on arXiv.
What the five prompts tested
These prompts were deliberately different. A model can look convincing in a calm, centered shot and still lose structure when hands enter frame, a subject moves quickly, or dozens of people compete for attention. The set therefore moves from a clear focal subject to the most crowded case.
| Test | Prompt demand | What to inspect |
|---|---|---|
| Neon train | Side tracking, reflections, steam | Whether the train remains the anchor while the background moves. |
| Pastry chef | Hands, glossy food, slow push-in | Finger stability, surface highlights, and focus through a close shot. |
| Snow fox | Fast animal, airborne snow, handheld camera | Legibility of the fox and continuity in the snow trail. |
| Repair robot | Small subject, clutter, slow orbit | Object identity, edges, and parallax around a compact subject. |
| Street festival | Crowd, floats, banners, crane move | Whether small people and background objects stay distinct. |
The wording also gives the model a clear hierarchy. Each prompt names a hero subject first, then a camera move, then secondary motion. That order matters in practice. Asking for ten equally important things makes it harder to tell what the clip should prioritize. A train, a tart, a fox, and a robot all provide a strong visual anchor. The festival intentionally removes that safety net.
Lance Text to Video settings, time, and cost
The Wiro model documentation lists a text prompt plus width, height, frames, steps, guidance scale, and seed. Its documented defaults are 848 by 480 pixels, 50 frames, 30 inference steps, guidance scale 5.0, and seed 0. The original five-output record does not preserve each run’s submitted parameter payload, so these defaults should not be read as a claim that every clip used those exact values. They are the configuration baseline exposed on the model page.
The documentation includes one completed task example with an elapsed time of 6.0 seconds. That is a useful indication of the example task only, not a guaranteed runtime for these five videos: queue time, worker availability, dimensions, frames, and steps can change a real run. The same documentation shows a cancelled example with totalcost of 0.003510. It does not publish a verified per-output price for this review’s five clips, so no per-clip price is asserted here. Treat the shown figure as a documentation example rather than a quote.
For a repeatable first pass, start with the documented 848 by 480, 50-frame, 30-step setup and a fixed seed. Keep the focal subject large in frame. Then change one variable at a time: add a camera direction, simplify background action, or raise the subject’s screen presence. This makes failures easier to diagnose than changing prompt, resolution, motion, and seed all at once.
What each output actually shows
1. Neon train: a clean read on tracking motion
The train stays readable as the camera travels beside it. Window reflections and steam give the scene enough moving detail to test whether the model turns background activity into a smear. The output holds the train as the frame’s main object and keeps the lateral movement coherent. This is the kind of shot Lance suits: one large subject, one direction of travel, and environmental motion that supports rather than competes with it.
2. Pastry chef: close detail with moving hands
This is a more demanding test than it first appears. Hands, utensils, steam, reflective glaze, berries, and a push-in all occupy a tight frame. The output keeps attention on the tart and preserves the warm, glossy look called for by the prompt. The hand action reads as careful rather than frantic, which helps. For food or product work, that is the lesson: use restrained action and a measured camera move when surface detail matters.
3. Snow fox: speed works better with a simple silhouette
The fox test adds fast subject movement but gives it a high-contrast setting. The red body remains easy to follow against snow and dark trees. Snow spray supplies a useful motion cue, so small changes between frames are less distracting than they would be against a blank background. The output suggests a practical prompt pattern: pair a moving subject with a scene effect that naturally carries motion, such as snow, dust, mist, rain, or water.
4. Repair robot: small products need a controlled orbit
The robot stays identifiable despite a crowded bench and reflective metal around it. This output matters for product concepts because it asks the model to protect a small hero object while the camera adds parallax. The slow orbit is a sensible choice. A faster orbit would increase the chance that tools, hooks, and edges distract from the robot. Choose this approach for a quick product mood clip, especially when a central object can stay large enough to read.
5. Street festival: the useful limit test
The festival has the most energy, but it also exposes the trade-off. The model preserves the overall feeling of a moving night celebration, with banners, confetti, and a rising camera direction. Fine detail in the smallest figures and props softens first. That is not a reason to avoid busy scenes. It is a reason to frame them around one readable event: a lead dancer, a float, or a foreground lantern. For a dense scene, avoid treating every person as equally important.
When to pick Lance Text to Video
Pick Lance when the shot can be expressed as one subject plus one intentional camera move. It fits product b-roll, short wildlife moments, transport shots, food close-ups, and cinematic establishing clips where the main object remains prominent. The documentation also exposes image-to-video and video-rewrite capabilities, so the model page is worth checking when a text-only test is not the right starting point.
Use a different approach, or split the work into shots, when the brief depends on many tiny faces, layered choreography, readable text, or several independent actions at once. The festival test shows why. Lance can keep the atmosphere moving, but a complex crowd does not deliver the same fine control as a single train or robot. For model comparisons, see LTX-Video vs Kling vs Seedance, WAN 2.7 Video: 5 Video Prompt Tests, and Top 5 Text-to-Video APIs in 2026.
The practical verdict is simple: make the subject obvious, keep motion readable, and reserve crowd-scale complexity for a separate test. Start a clip with Lance Text to Video on Wiro when that is the shot you need.
