Getting the best results

Use a clear, well-lit photo. Avoid extreme close-up crops — the AI video model needs a fairly normal width-to-height ratio to work with, so a photo that's already cropped very tight (just a face filling the whole frame, for instance) tends to produce stranger motion than a photo with a bit of breathing room around the subject.

Good lighting matters more than resolution. A well-lit photo on a basic phone camera will usually animate better than a dim, blurry photo from a nicer camera.

The video can only animate what's already in your photo. This is the single most common source of "instructions not followed." The model brings your one flat photo to life — it does not add people, hands, or objects that aren't already visible in it. If you upload a photo of just the product and write a prompt like "a woman smiling warmly while holding the product," there's no woman in the frame for the model to animate, so it falls back to a generic move on the product itself (usually a small pan or zoom) and nothing you described happens. Match your prompt to what's actually in the photo:

• Product-only photo → describe camera movement around the product: "camera slowly orbits the product, studio lighting" (or just use the Product Showcase template's default prompt).
• Want a person reacting to or holding the product → you need a source photo that already shows that person with the product. A prompt can direct how an existing hand, face, or body in the photo moves — it can't invent one that isn't there.

Don't have a photo of a person already holding the product? Use "Merge product + person photo" on the upload step instead of a single upload — upload the product photo and a photo of a person separately, and the app combines them into one realistic photo for you. Check the merged result carefully before using it (hand/product contact is the part most likely to look a little off) — retry with a clearer description of the pose if it doesn't look right.

A talking template can make a held object "float away." If your source photo shows someone holding something, be careful with templates like UGC Testimonial whose default prompt says the person is speaking "enthusiastically" — the model can take that as license to gesture with both hands while talking, which pulls the hands away from whatever they were holding a moment ago (the object then looks like it's floating in front of them, disconnected). Fix it by writing your own motion prompt that explicitly keeps the hands anchored, e.g. "Person smiles and talks to camera while holding the [product] steady with both hands, minimal hand movement, subtle handheld camera movement, warm lighting" — the phrase "holding it steady" is what stops the model from freeing the hands to gesture.

This is the single biggest factor in output quality, once your prompt actually matches what's in your photo. A few rules of thumb:

One clear action per scene, not a story. The AI video model gets 5 seconds to work with per scene. Asking it to do two or three things in sequence — "a hand picks up the product, then the person smiles, then holds it to her chest" — usually produces something awkward, because the model tries to cram multiple beats into a few seconds and the transitions look unnatural (a hand and a face just "appearing" instead of smoothly moving into frame).

Instead, split multi-step actions into separate scenes:

• Scene 1: A hand reaches in and picks up the product, smooth motion, soft lighting
• Scene 2: Close-up of a woman smiling warmly while holding the product close to her chest

Name the gesture, don't describe what it means. If you want a hand sign in a scene, use its actual name — "thumbs up," "OK sign," "peace sign," "wave" — rather than describing the meaning you're going for. A prompt like "hand action says very good" isn't a gesture the model recognizes, so it tends to default to some generic pose (a raised finger is a common fallback) instead of the sign you had in mind. Combined with cramming in other actions — e.g. "hand moves to pick packet, person looks at camera, hand action says very good" — the model has three unrelated things to render in 5 seconds and usually collapses down to just one, chosen somewhat arbitrarily. Split it instead: Scene 1: A hand reaches in and picks up the packet, smooth motion, bright even lighting / Scene 2: Close-up of a person smiling at the camera and giving a thumbs-up, bright even lighting.

Describe it like a shot, not a sentence. Camera/motion language works better than narrative language. Compare:

• Weaker: "A person walks up to the counter and picks up the item and looks happy about it."
• Stronger: "Slow push-in on a smiling person's face, warm natural light, subtle handheld movement."

Common motion phrases that work well: "camera slowly orbits the subject," "smooth slow-motion pan," "gentle dreamy camera drift," "punchy fast zoom-in," "subtle handheld movement," "product rotates on a turntable," "soft studio lighting."

Vague phrases produce vague — or minimal — motion. If a phrase isn't close to one of the ones above, the model often can't commit to a direction and falls back to a small, generic movement (a slight up-and-down bob is a common default for product shots when the prompt doesn't give it clear direction). For example, "camera moving around" doesn't tell the model which way to move — "camera slowly orbits around the product" does. Likewise, "full focus from a distance" mixes two different ideas (focus and camera distance) into one confusing phrase — split them into a normal shot description instead: "wide shot, camera further back, product in sharp focus." If a reel comes back with barely any motion, this is usually why — try rewording with one of the specific phrases above before assuming something's broken.

If a scene comes out wrong, don't fight it with a longer prompt — simplify it. Cut it down to the one motion you actually want and drop everything else.

Match the photo to the pose you want. The AI never sees anything beyond your one flat photo — it has no real 3D understanding of the room or the person's body. If your photo shows someone facing sideways or away from the camera, and the prompt implies they should end up facing forward (e.g. "speaking to camera"), the model usually can't convincingly rotate a whole body — shoulders, hips, feet — in 5 seconds. Instead it takes a shortcut and just swivels the head, which looks unnatural (the classic "body facing one way, head facing another" look). To avoid this: start with a photo where the person is already roughly facing the camera if that's how the final video needs to look, and save prompted "turns" for small ones — "slight turn of the head and shoulders toward camera" — rather than a full-body rotation.

Keep captions short — one punchy line reads much better on a 9:16 vertical video than a full sentence. Think headline, not paragraph: "Real results, real fast" rather than "We deliver real results very fast for all of our customers."

If generating music with AI, describe it the same way you'd describe a mood board — genre, tempo, and instruments, not a vague feeling:

• Weaker: "happy upbeat music"
• Stronger: "lo-fi hip hop, chill, slow tempo, piano and soft drums"

Tag-style prompts (comma-separated descriptive words) consistently produce better results than full sentences.

Use the preset buttons (Birthday, Wedding, Travel & Tour, etc.) when a reel matches one of those categories — this keeps your "My Reels" library organized and searchable. For anything more specific, type a custom name instead — "Mysore Garden — Fruits & Spices" is far more useful to search for later than "Product Shoot" alone. You can leave the "show on video" box checked if you want the name visible on the finished video, or unchecked to just use it as an internal label.

The default look ("Made with Reelcheyyu," bottom-right, white) works fine for most cases. Tick "Customize" if you want your own text, a different corner, one of the bundled fonts, or a specific color — useful if you're delivering finished reels to a client and want their brand name on it instead of yours.

When joining several reels into one longer video, order matters — tick them in the sequence you want them to play, not just the order they happen to be saved in your library. The number badge on each selected reel shows you the current play order.

Each reel card in "My Reels" has a small "Delete" link — use it to remove reels you don't need anymore. This is permanent and also removes the video file itself, so it can't be recovered afterward.

If a generated video looks visually off (warped face, strange motion, blurry regions), it's almost always a prompt or source-photo issue rather than a bug — try a clearer photo, simplify the prompt, or split a complex prompt into two scenes as described above. If the app itself isn't behaving (stuck on "Queued," errors on generate, blank preview), that's worth flagging separately — that's a technical issue, not a prompting one.