Contents
- What a template does that a prompt does not
- The motion recipes described on the product site
- Match the template to the crop
- When a custom line helps — and how to write one
- A one-variable test plan
- Common prompt mistakes
- Frequently asked questions
Every image-to-video generation answers two questions: what does the picture look like, and what should change over time? The photograph answers the first. A template, a custom prompt, or a mix of the two answers the second. Choosing badly between them is one of the most common reasons a clip comes out wrong even when the source photo is good.
Playbox AI is an Adults 18+, template-led image-to-video generator, so the template is the default way to describe motion, with an optional short custom line on top. This guide explains what each approach is good at, how to match a template to your crop, and how to write a custom line that helps rather than interferes. It builds on the workflow guide, which covers how to review the result.
Everything here describes how the product is presented on this site's About page. Template names, limits and options can change between releases, so check the live interface before you rely on any specific detail.
What a template does that a prompt does not
A prompt-first video tool begins with a blank box. You must describe camera movement, pace, lighting and performance in words, and the quality of the result depends heavily on how well you phrase it. A template-led tool starts from a prepared motion recipe: the recipe carries the movement pattern, and your photo carries the subject.
That has three practical consequences:
- The skill floor is lower. You do not need to know how to describe a slow dolly or a head turn; you pick one that exists.
- The ceiling is capped. You can only get the movements that the recipes offer, plus whatever a short custom line can steer.
- Results are easier to compare. The same recipe on two different photos tells you about the photos, which is exactly what you want when testing sources.
A recipe also pre-answers the question that most often decides whether a clip works: how much movement to request. Too much is the single most common cause of smeared motion and drifting faces.
The motion recipes described on the product site
The About page describes a small set of prepared treatments. The table below lists them with the kind of source each one tends to suit, as a planning aid rather than a promise of results.
| Recipe (as described) | Movement level | Source that tends to suit it | Main risk |
|---|---|---|---|
| Portrait with subtle facial motion | Low | Close, sharp, evenly lit face | Very little; the safest first choice |
| Cinematic camera move | Medium | Clear subject with usable background on every side | Background shimmer if the scene is busy |
| Travel-style breeze and light | Low to medium | Outdoor or window-lit scenes with soft light | Colour or light shifts if lighting is mixed |
| Anime-style motion | Medium | Stylised or illustrated sources | Style drift on photographic sources |
| 3D product spin | Medium to high | A single object on a plain background | Geometry that changes as the object rotates |
The movement levels are our own judgement of how much each treatment asks of the model, not a figure published by the product. Use them to choose an order of testing: begin at the top of the table and work down only as each one passes the four checks.
Match the template to the crop
The crop of your source sets how much room the model has to move things, and the template decides how much movement it will try. The two have to agree:
- A close portrait suits small head and camera movement. There is little spare background, so a large camera move or a big turn will run out of image almost immediately.
- A wider source can support more. If there is real background on both sides and above, a camera move has something to reveal. Our framing guide explains how to check this with crop overlays.
- A busy background argues for less camera movement. Every edge must be re-drawn from a new position, and detail such as foliage or shelves is the usual cause of boiling.
- A source with a tiny face argues for a recrop first. No template will fix a face that occupies too few pixels.
When the selected motion asks for detail the source cannot supply, artefacts multiply. The fix is rarely a different template; it is usually a source change followed by the same template again.
When a custom line helps — and how to write one
The product site describes a custom prompt field that accepts up to 500 characters. That is generous, and most of it should go unused. Short, concrete phrases about camera path, mood and light direction work better than long prose, because the recipe is already supplying structure and a second long description competes with it.
Use a custom line when a recipe is close but not exact. Typical reasons:
- You want the camera movement slower or smaller than the default.
- You want to name the direction of a movement, such as toward or away from the camera.
- You want to hold something still, such as the background or the framing.
A reliable structure is the same one recommended in the workflow guide:
Subject action + pace + camera behaviour + continuity condition
For example: "The subject slowly turns a few degrees toward the camera. Camera makes a very slow push forward. Continuous shot, steady lighting." Notice what it does not contain: no description of clothing, room or appearance. That information is already in the photograph, and restating it risks contradicting it.
Image cues and written cues must agree. If the source shows dust clouds or motion blur and the prompt asks for stillness, the model has conflicting instructions. Runway's published image-to-video prompting guide documents a case of exactly this kind, and notes that blurry hands or faces in an input image may be intensified during animation. When a video ignores an instruction, inspect the source before adding more words.
A one-variable test plan
Use a fixed source and change one thing per attempt. A sensible sequence for a new photo:
- Attempt 1: the lowest-motion recipe, no custom line, shortest render.
- Attempt 2: the same recipe with a short custom line about camera pace only.
- Attempt 3: a medium-motion recipe, no custom line.
- Attempt 4: the best recipe so far at a longer length, only if the short version has passed.
Score each attempt with the four checks from the first-clip walkthrough — identity, limbs, background, motion intent — and record the result. By attempt four you will know whether the template, the custom line or the source is responsible for any remaining problem. Keeping that record is the habit that makes the repeatable workflow possible.
Common prompt mistakes
- Describing the picture again. Clothing, setting and appearance are already in the image.
- Stacking actions. A turn, a gesture, flowing hair and a camera move together make diagnosis impossible.
- Using intensity words by reflex. "Dramatic" and "fast" raise displacement and therefore failure risk.
- Contradicting the source. Asking for stillness from a blurred source, or for a pan with no background to reveal.
- Rewriting the prompt before checking the first frame. A defect visible at the start is a source problem.
- Changing template and prompt together. You will not know which change mattered.
Frequently asked questions
Should I always add a custom prompt?
No. Run the template alone first. Add a short line only when the result is close and you can name exactly what to change, such as the pace or direction of a movement.
How long should a custom prompt be?
As short as the instruction allows. The field can take far more text than is useful. One or two sentences about action, pace and camera are normally enough.
Which template is best for a beginner?
The one with the least motion that suits your crop — typically a portrait treatment with subtle facial movement. Fewer moving parts means a clearer signal about the source.
Can I use a template meant for a different style?
You can test it, but expect a lower pass rate. Treat a mismatch, such as an anime-style treatment on a photographic source, as a harder test and run it after your simpler pairings.
Why does the same template behave differently on different photos?
Because the photograph supplies identity, light, detail and framing. The template only supplies the movement pattern. That is why the same recipe is a good way to compare sources.