DEV Community: Louis Chi

Jewellery AI video prompts: how to make stones sparkle without a fake star (12 prompts)

Louis Chi — Sat, 14 Nov 2026 16:12:49 +0000

Jewellery is the hardest product for AI video, because the buyer is looking at stones, prongs and a clasp, and one wrong detail is a wrong product. The most useful thing we learned came from a real necklace job where half a sentence decided the result.

The sparkle sentence

We were making a clip for a small shop's pendant necklace. The first prompt asked for "brilliant sparkles". The model gave one big cross-shaped star on a metal prong. It looked like a sticker, not a stone.

The fix was to describe the sparkle, not ask for it. "Dozens of tiny white pinpoint glints… inside the facets", plus "no star-shaped flare, no glow", put the light inside the stones.

A frame from the second of two takes: a hand lifts the pendant off cream satin into the sun. Made on Kling 4 from four of the shop's own photos, 5 seconds, 1:1.

Most of the prompts below reuse that wording.

Setup

Save the piece once from 3 to 5 sharp photos: straight on, a macro of the stones or setting, the clasp or back, and one on a hand or neck for scale. In the prompts, "the ring", "the necklace" and so on stand for that saved piece.

Rings

1. Ring on the hand (5 s, 1:1)
Close-up: a woman's hand with a soft natural manicure, wearing the ring, rests on cream silk and turns slowly toward the camera, the ring catching small sparkles in warm window light. Shallow depth of field, gentle slow push-in, elegant and natural.

2. Ring on black velvet (5 s, 1:1)
Macro shot: the ring rests on rich black velvet and turns slowly while a soft beam of light sweeps across it; dozens of tiny white pinpoint glints flicker inside the stone's facets as it turns. No star-shaped flare, no glow. Deep black background, shallow depth of field.

3. Ring box opening (5 s, 9:16)
Top-down on a light wooden table in soft daylight: two hands open a small velvet ring box toward the camera, revealing the ring standing in its slot; the camera pushes in until the ring fills the frame, small pinpoint glints across the stones.

Necklaces

4. Pendant lifted into the sun (5 s, 1:1)
The necklace lies on cream satin silk in natural afternoon sunlight from a window. One woman's hand enters from the left, pinches the top of the pendant and lifts it slowly toward the camera, the chain trailing back onto the silk; as it rises, dozens of tiny white pinpoint glints twinkle across its stones, with small flashes of rainbow fire inside them. No star-shaped flare, no glow; the metal stays plain and polished.

5. Pendant turntable (5 s, 1:1)
The necklace hangs from a small clear stand on a slowly turning white turntable against a light grey seamless background, soft even studio light, the whole pendant in sharp focus, the chain falling straight. Only use it if your photos show the back of the pendant; otherwise the model invents it.

6. Gift under the tree (8 s, 9:16)
A small white gift box with a red satin ribbon sits under a glowing Christmas tree; a pair of hands pulls the ribbon loose and lifts the lid. Looking into the box: the necklace rests on white tissue paper, fairy lights twinkling around it as the camera pushes in slowly.

Earrings

7. Earrings on velvet (5 s, 1:1)
Macro shot: the pair of earrings rests side by side on rich black velvet while a soft beam of light sweeps slowly across them; tiny white pinpoint glints flicker across the stones. No star-shaped flare, no glow. Deep black background, shallow depth of field.

8. Floating earrings (5 s, 9:16)
The pair of earrings floats weightlessly in mid-air against a soft peach-to-lavender gradient, turning slowly through half a turn while drifting gently up and down, a soft round shadow on the floor below. Bright soft studio light, locked-off camera, crisp detail on the stones.

9. Earrings with falling petals (5 s, 9:16)
Slow motion: soft pink rose petals drift down through warm light around the earrings, which hang from a small pale stone stand against a blush-pink background; the camera pushes in gently until the earrings fill the frame.

Bracelets, watches and bridal sets

10. Bracelet on the wrist (5 s, 1:1)
Close-up: a woman's wrist wearing the bracelet rests on cream silk and turns slowly toward the camera in warm window light, small pinpoint glints moving along the links. Shallow depth of field, gentle slow push-in.

11. Watch macro (5 s, 1:1)
Extreme macro of the watch: the camera glides slowly across the dial as the seconds hand sweeps, light rolling over the polished case and the hour markers, then pulls back to show the whole watch resting on dark leather. Warm low-key light, shallow depth of field.

12. Bridal set flat lay (5 s, 1:1)
Top-down on warm beige linen: the bridal set (necklace, earrings and ring) lies in the centre as a few small wedding props, a sprig of white flowers and a lace ribbon, slide into the frame one by one around it. Soft even daylight. Save the set as one piece, photographed together.

Rules that kept the stones and metal exact

Let the photos carry the design. Do not describe the metal colour, the stone count or the shape in the prompt. A second description competes with the photos, and the model may split the difference.
Keep the move small and the piece front to camera. A slow turn of up to about 45 degrees, a lift or a push-in shows the stones. A full spin asks the model to invent any side your photos do not show.
Check the clip frame by frame against your photos: the outline, the number of stones, the prongs and the clasp. If one detail drifts, change one thing in the prompt and run it again, not everything at once.

Try them

The full list, with the template that runs each prompt, is on the jewellery prompt page. The necklace job, take by take, is in how to make a necklace video with AI. The first clip is free (3 seconds on Kling 3.0 Omni, no card, no watermark); a single video after that starts at $2.99.

This article was written by an AI agent for RefVid from the facts on the linked page; the necklace clip is real, unedited output of a shop's own photos.

Shoes AI video prompts: 10 copyable sneaker prompts (6 tested on the same shoe)

Louis Chi — Sat, 14 Nov 2026 16:12:01 +0000

A good shoe prompt tells the model where the camera sits, what the feet or the shoe do, and what the ground or set is. It does not describe the shoe. Your reference photos say what the shoe is.

Below are 10 prompts. Six of them made real clips of the same invented running shoe, AERO 1, word for word. The other four are template prompts first shot with other products. Every clip was made from three reference photos on Kling 3.0 Omni, 5 seconds, 720p, no sound.

Setup: save the shoe once

Take 3 to 5 sharp photos on a plain background: the outer side profile, a three-quarter front view, the back with the heel, and the sole. In the prompts, "the shoes" stands for that saved pair.

Length and shape (for example "5 s, 16:9") are settings, not prompt text.

On-foot prompts

Put the camera at shoe height. From ground level the shoe fills the frame and the model spends its detail on the upper and the sole, not on a face.

1. Wet pavement at dawn (5 s, 16:9)
A runner wearing the shoes strides across wet city pavement at dawn. Low tracking shot at ground level follows the shoes as they hit the asphalt, small splashes rise from shallow puddles, streetlight reflections ripple on the wet ground, the coral heel counter with the AERO 1 text stays visible. Slow motion, cinematic running-shoe commercial, shallow depth of field, cool blue light with warm highlights.

A frame from the clip made with this exact prompt. Change the heel detail to your own.

2. City street walk (5 s, 9:16)
Low camera at ankle height tracking alongside a pair of feet wearing the shoes as they walk briskly down a sunlit city sidewalk, shadows of passers-by, shallow depth of field. Close-up of the shoes landing mid-step on the pavement, a little dust lifting, the camera slowing to a stop beside it.

3. Gym sprint start (5 s, 16:9)
Floor-level close-up in a modern gym with a dark rubber floor: an athlete wearing the shoes pulls the white laces tight, then drives off the floor into a sprint start and the shoe flexes as it pushes away from the camera. Hard side light, fine chalk dust floating in the air, crisp sports commercial look, shallow depth of field.

Prompt 3 on AERO 1. Change "white laces" to your own.

Studio prompts

4. Rotating plinth (5 s, 16:9)
The shoes slowly rotate on a round white plinth in a minimal studio with a smooth grey backdrop and a soft coral glow behind it. Static camera at shoe height while the plinth turns, soft key light gliding over the knit texture, the white foam midsole and the translucent blue outsole. Clean premium product commercial, sharp focus.

5. Turntable packshot (5 s, 1:1)
The shoe, matching the reference appearance and proportions, is on a circular white studio turntable against a light grey seamless background, under soft even studio light, with the entire product in sharp focus. A full turn needs photos of every side, the inner side included.

6. Floating in mid-air (5 s, 9:16)
The shoe floats weightlessly in mid-air against a soft peach-to-lavender gradient background, turning slowly through half a turn while drifting gently up and down, a soft round shadow on the floor below. Zero gravity, bright soft studio light, locked-off camera, crisp detail on the label.

7. Frozen splash, bullet time (5 s, 9:16)
Bullet time: the shoe hangs frozen in mid-air inside a burst of crystal-clear water droplets and splashes, every drop suspended and perfectly still, while the camera orbits smoothly half a circle around it and ends with the side profile facing the lens. Soft blue-white gradient background, bright studio light, crisp macro detail.

Unboxing and sale prompts

8. Shoebox unboxing (10 s, 9:16)
Top-down shot of a plain kraft box on a light wooden table in soft daylight; two hands lift the lid and fold back white tissue paper. Close-up as the hands lift the shoes out of the box and hold them up, turning them slowly so the side profile faces the camera. The shoes stand on the table next to the open box, the camera gently pushing in.

9. Neon sale drop (5 s, 9:16)
A dark studio set lit by pink and blue neon tubes, stacks of black gift boxes with satin ribbons; the shoes sit on top of the tallest box as the camera rises slowly toward them. Slow push-in on the shoes, neon reflections sliding across them, glittering confetti drifting down. Glossy, premium, high contrast.

10. Gift under the tree (5 s, 9:16)
A white gift box with a red satin ribbon sits under a glowing Christmas tree, warm fairy-light bokeh; a pair of hands pulls the ribbon loose and lifts the lid. Looking into the open box: the shoes rest on crisp white tissue paper, golden lights twinkling around them as the camera pushes in slowly.

Three rules that kept the sole, laces and logo exact

Let the photos carry the shoe. Do not describe the colourway, the logo or the sole pattern unless the shot turns on it, as prompt 1 does with the heel. A second description competes with the photos, and the model may split the difference.
Show the sides your photos show. A side-on walk or a half turn stays on the profile you photographed. A full turn or a top-down shot asks the model to invent the inner side or the insole. Add a photo of that side first.
Keep feet simple. Brisk walking, one step landing or a sprint start reads well. Jumps, tricks and laces being tied in close-up give the model hands, feet and laces to tangle. Check the clip frame by frame against your photos: the outline, the sole, the laces and the logo position.

Try them

All ten prompts, with the templates that run them and the full list of rules, are on the shoes prompt page. The first clip is free (3 seconds on Kling 3.0 Omni, no card, no watermark); a single video after that starts at $2.99.

This article was written by an AI agent for RefVid from the facts on the linked page; AERO 1 is an invented shoe and the clips shown are real, unedited output.

eBay listing video: requirements, and how to make one from photos you already have

Louis Chi — Sat, 14 Nov 2026 16:10:37 +0000

eBay lets you add one short video next to your listing photos. Most sellers skip it because they think it means a camera, a tripod and an evening of editing. It does not have to: if you already have three or four photos of the item, you can make a clean turntable clip from them.

This post covers what eBay asks of a listing video, which clips are safe to upload, and the three steps I use to make one. The example item is AERO 1, an invented running shoe, so no real brand or seller is involved.

What eBay asks for

From eBay's own help page on adding a video to a listing (read 2026-11-09):

One video per listing, ideally one minute or less.
MP4 (MP4 or MOV in the app), at most 150 MB, uploaded at up to 1080p.
Every video is reviewed first. It usually shows within 48 hours, up to seven business days at peak times.
Listings with variations cannot have a video, and links to YouTube are not accepted.

A 5-second clip made from photos is a few megabytes at most. The one below is 0.3 MB, a small fraction of the 150 MB limit.

Three photos in, five seconds out

Reference photo 1 of 3. The listing photos work: front, back, and a close-up of the label or a flaw you want buyers to see.

Reference photo 2 of 3.

The clip is a 5-second, square turntable made on Kling 3.0 Omni with the 360 turntable template, from those three photos, not edited afterwards. You can see the same clip, and the three photos it came from, on the eBay listing video page.

The three steps

Add 1 to 10 photos of the item. The listing photos are enough. The tool saves them as one product, so the model sees the same item on every clip.
Pick the shot. A turntable is the safest. You can also write your own, for example "the sneaker on a white studio turntable, slowly turning, soft even light".
Generate, download the MP4, upload it with Add video in the Photos & Video section of your listing.

Which clips work on eBay

Buyers want to see the real item from every side, so show the item and nothing it does not have: a turntable spin, a slow push-in on the label, the item in a hand for scale, or an unboxing for new stock.

One caution that matters on eBay: small printed text can still blur or change in a generated clip. Check the clip against the item before you upload it. Never use a clip that shows a feature, colour or accessory the item does not have.

Also, a full turn asks the model to invent any side your photos do not show. If you want a full turn, take a photo of every side first.

What it costs

The first clip is free: a 3-second clip on Kling 3.0 Omni, no card, no watermark. After that a single video starts at $2.99, and credits never expire. A failed clip gives its credits back.

If you sell in volume, the faster route is to save each item once and reuse the same turntable template across listings. The product photo to video guide covers how many photos to take and how to shoot them.

This article was written by an AI agent for RefVid from the facts on the linked page; the example item is invented and the clip shown is real, unedited output.

How to Make an AI Drama Video With the Same Characters in Every Episode

Louis Chi — Thu, 12 Nov 2026 16:22:54 +0000

This post was written with AI help (Claude), from our own screenshots and screen recording of the RefVid studio; every step shows the real UI. The characters are from GuaPi, our own AI-made cartoon series.

An AI drama falls apart when the faces drift between episodes. Describe a character in words only and the model draws a new face every clip. The fix is to save each character once, from photos, and start every clip from those photos. Only the words change.

Here is the whole thing in a 4-minute video, then step by step with a screenshot each. Full written guide with the transcript: How to make AI drama videos with the same cast.

1. Save each character once, from photos

In Elements, click New element, give it a name and choose Character. Add three or four views: front, side and a close-up. The first photo is the main one.

2. Do the same for the whole cast

Every character is its own element. Here the cast is Henry the boy, GuaPi the goose and Xiaobai the dog. You never upload them again.

3. Start the episode from the Mini Drama template

Open Templates and click Use template on the AI short drama scene: three shots, 10 seconds, vertical 9:16. You start from a working drama beat, not an empty box.

4. Add the second and third character

Click the + beside the photos and pick a saved element: Henry takes the second lead, then add Xiaobai. One clip holds up to three characters.

5. Write the episode as one prompt

Who does what, and how the beat ends, in your own words. Use the names you saved.

6. Pick the model, the length and 9:16

The video uses Kling 4, 10 seconds, vertical 9:16, 720p. The cost shows above Generate before you click.

7. Generate the first episode

A clip takes a few minutes. One clip is about ten seconds; an episode is a few clips, joined in any editor.

8. Next episode: same cast, new words

Leave the three characters in the composer, write only the new beat and generate again.

9. Two episodes, one cast

The same boy, the same goose and the same dog, because every clip starts from their saved photos.

Before you post

Export vertical 9:16 for TikTok, Reels and YouTube Shorts.
Switch on the platform's label for AI-generated content.
Keep it fast and end on a cliffhanger.

More about making a series with one cast: AI drama maker.

How to Add an AI Voiceover to a Product Video (Script, Voice and Download in One Place)

Louis Chi — Sun, 08 Nov 2026 20:25:56 +0000

Most AI video tools stop at the clip. If you sell on Shopify, Etsy or your own site, the next job is a voiceover for Reels, TikTok or the product page, and that usually means a second tool: write a script, generate a voice, line it up in an editor, export again.

We built the voiceover into the same place the video is made. Here is the whole flow on a real example: four photos of a sofa, a 360° turntable clip, then a Polish voiceover.

1. Make the product video

Pick the 360 degree turntable template. It writes the shot: the product on a white turntable, a fixed camera, one full turn.
Upload up to four photos of one piece (front, side, back, other side) with the front view first. Use photos of a single version of the product, or the model mixes them.
Choose Kling 4 for the smoothest turn, the shape you will post (1:1 for a shop page, 9:16 for Reels and TikTok, 16:9 for a site) and a 10-second clip. Each shape is generated on its own, never cropped.
Generate. A ten-second Kling 4 clip took about four minutes here.

2. Add the AI voiceover

Open the finished video and switch to the Voiceover tab:

A script is already written from the video, short enough to end with the clip (about 14 characters a second; 6 in Japanese and Korean). It only says what the video shows; edit it or click Rewrite.
Pick one of ten languages: English, German, French, Spanish, Portuguese, Italian, Dutch, Polish, Japanese or Korean. The script is rewritten in that language.
Pick one of six voices. Every voice speaks every language.
Click Make voiceover. About fifteen seconds later a new version appears and plays.

The voice replaces the clip's own sound, so a model that always adds sound (Kling 4 does) no longer plays its audio under the narration. The original stays in the list, and every voiceover is kept as its own version, so you can try another voice or language and keep the best one.

3. Download

Download saves the version you picked: an MP4 with the voice as an AAC track, which every player and upload form reads. If you would rather record your own voice, More → Download without sound gives the same clip with no sound at all.

What it costs

A voiceover is priced by the clip's length, in credits: 4 for 5 seconds, 6 for 10, 10 for 15 and 18 for 30. The price is on the button before you click, and a failed voiceover is refunded.

The full step-by-step guide, with screenshots and the transcript, is here: AI voiceover for product videos.

How to keep the same character across AI video clips with reference images (Kling 3.0 Omni, MiniMax H3, Wan, Seedance)

Louis Chi — Sat, 07 Nov 2026 21:42:57 +0000

The most common complaint about AI video is not quality. It is that clip two shows a different person from clip one. You describe "a woman with a copper-red bob and an olive jacket", and the model draws a new woman who fits that description every time you press Generate.

The fix that works is reference images: you show the model the same few photos of your character on every clip, and you stop describing her in words. This tutorial walks through how to do that, which photos to pick, how to write the prompt, and how the main reference models differ: Kling 3.0 Omni, MiniMax H3 (Hailuo), Wan 3.0 and Seedance 2.5. Every image below is a frame from a real, unedited clip we made in RefVid, the tool I work on.

Step 1: Make one invented character and photograph her four ways

Our test character is Mia. She is invented, not a real person, which matters later (some models refuse real-looking faces, and you should only ever use photos of people who agreed to it).

She has four reference images, and the set is the whole trick:

Front, head and shoulders
Three-quarter view
Profile
Full length, so the model knows her jacket, jeans and build

Reference 1 of 4: the front view. The sharpest, most evenly lit photo goes first.

Reference 3 of 4: the profile. Without it, a model has to guess her nose and ear when she turns.

Rules that held up across every model we tried:

Three to five photos beat one. One photo is enough to start, but the model has to invent every angle the photo does not show. More angles hold better when the character turns her head.
Plain background, even light, no filters, no text. Anything in the photo can leak into the clip.
Never change the set between clips. A new haircut photo or a different outfit in clip two is, to the model, a new character. Save the set once and reuse exactly those files.

Step 2: Write the scene, not the person

Once the photos are attached, the prompt should only say where she is, what she does and what the camera does. Do not repeat her hair colour or her jacket. When the words and the photos disagree even slightly, the model blends them, and the blend is the drift you were trying to avoid.

Here is the prompt for the first clip, word for word:

She walks through a neon-lit Tokyo side street at night in light rain, wet asphalt reflecting pink and blue neon light. The camera tracks alongside her at shoulder height in a smooth lateral dolly as she walks.

Kling 3.0 Omni, 5 seconds, from the four photos. Not edited.

Then the same four photos with completely different scenes:

Same photos, new prompt: a windy beach at golden hour, slow push-in to a close-up.

Same photos: a Scandinavian kitchen in morning light, pouring coffee, then smiling at the camera.

Same photos: a misty pine trail, turning back toward the camera.

Four scenes, four lighting setups, one face. For comparison, the prompt-only version of the kitchen scene (no photos, Mia described in as much detail as we could fit) gave a woman who matched the words but not the other clips:

Words only, no reference images. Close, but not Mia.

All of the Mia clips above ran on Kling 3.0 Omni, which takes up to 10 reference images and makes 3 to 15 seconds. If you want to try it without the Kling app or an API key, our Kling 3.0 Omni free online page runs a first 5-second clip on it for free.

Step 3: Keep the photos, switch the model

Reference images are not a Kling feature; most current video models have a reference mode. The useful part is that the same saved photos work on all of them, so you can pick a model per shot and the character stays put. What changes is how many photos each takes, how long a clip can be, and what kind of subject it accepts:

Model	Photos it takes (in RefVid)	Clip length	Good for
Kling 3.0 Omni	up to 10	3–15 s	People, products, pets
MiniMax H3 (Hailuo)	up to 5	5–15 s, with sound	People, short talking shots
Wan 3.0	up to 10	3–30 s	People and products, long single takes
Seedance 2.5	up to 10	4–30 s	Products, pets, illustrated characters (refuses real-looking people)

A new account's free credits cover one short first clip from your photos on several of these: MiniMax H3 (5 seconds at 768p, with sound), Wan 3.0 (4 seconds at 720p) and Vidu Q4, another reference model that takes 1 to 10 photos. If you know these models by their app names, the same free first clip is on our Kling AI free and Hailuo AI free pages (Hailuo is the MiniMax video app).

MiniMax H3: Hailuo's subject reference, now with several photos

Hailuo's original subject reference model, S2V-01, read a single face photo. MiniMax H3 reads several. We sent it the same four Mia photos:

MiniMax H3, 5 seconds at 768p, same four photos, with street sound and her line "Hi, I'm Mia." Back in 33 seconds.

The differences between S2V-01 and H3 are laid out in our Hailuo subject reference guide.

Wan 3.0: the long take

Most Wan reference-to-video tutorials are ComfyUI workflows written for Wan 2.6 or 2.7. Wan 3.0 takes up to 10 reference images and runs up to 30 seconds in one pass, which is the easiest way to keep a character through a long continuous action instead of stitching 5-second clips. Write the actions in order, one per sentence, so the model knows the sequence. More in Wan reference to video online.

Seedance 2.5: for products and drawn characters

Seedance 2.5 refuses photos of real-looking people before the clip runs. For a product, a pet or an illustrated character it holds detail very well. Here is AERO 1, an invented running shoe, from three product photos:

Seedance 2.5, 5 seconds at 720p, from three photos. The knit upper, heel tab and white sole match the photos.

For an illustrated character, a reference sheet that shows it from several sides works as one of the photos. Details in Seedance reference to video.

Checklist

3–5 photos of one subject: front, three-quarter, profile, full length.
Plain background, even light, sharpest photo first.
Save the set once; never swap photos between clips of the same story.
The prompt says place, action and camera. Never appearance.
Pick the model per shot: Kling 3.0 Omni or MiniMax H3 for short character shots, Wan 3.0 for long takes, Seedance 2.5 for products and drawn characters.

If you try this and a model still drifts on your character, tell me in the comments which model and how many photos you used. That is usually where the answer is.

Product photo to video with AI: from 1–10 shop photos to 5-second clips that keep the label (real outputs inside)

Louis Chi — Fri, 06 Nov 2026 07:17:43 +0000

Most product pages on Shopify or Etsy still have only photos. A 5-second clip of the product spinning, splashing or being worn shows the back, the scale and the finish better than another still, and you no longer need a shoot to make one. This post walks through how we turn product photos into short clips in RefVid: what kind of AI model keeps the label intact, which photos to give it, which shots to make first, and how to write the prompt. Every image below is a frame from a real output, next to the photos it came from. The full guide, with every template and the current prices, is on our product photo to video page.

A frame from a 5-second clip made from one side-view photo of AERO 1, an invented running shoe, on Kling 4. Prompt: "the shoe on a wet rock by a mountain stream at sunrise, the camera moving slowly along it". Not edited.

Image-to-video vs reference-to-video

There are two ways to turn a product photo into a video, and the difference decides whether your label survives.

Image-to-video animates the photo itself as the first frame. Once the camera moves away from that frame, the model has to invent the product, so it drifts: the logo warps, the stitching changes, the colour shifts.
Reference-to-video treats your photos as a description of the product and builds every frame around it. The shape and label hold through the whole clip, and a second and third clip match the first, which matters when a listing or an ad set needs several videos of one item.

One of the three AERO 1 photos behind the next two clips: a plain background, even light, the label readable.

Two scenes from the same three photos, the same shoe in both (Kling 3.0 Omni in RefVid).

RefVid is reference-to-video: you save the product once from its photos, and every clip in any scene starts from that saved product.

Step 1: give it the right photos

One sharp photo is enough to start: your main product-page angle, on a plain background, with the label readable and the colours accurate. If that is all you have, RefVid can draw three more angles from it with the same shape, colours and label text, so the video model sees more than one side.

With three to five photos the result gets steadier: add a side, the back and a close-up of the label. You can save 1 to 10 photos per product. If the photos are already on your shop page, paste the product link instead of uploading; RefVid reads the photos from Shopify stores, Amazon listings and shop pages that describe the product in structured data.

Leave out photos with price stickers, props that are not part of the product, or phone-screenshot clutter. A screenshot of a listing brings the status bar and the app's buttons into the video, so crop it first:

Adding a photo in the studio. When a photo has a phone screenshot's shape, the studio offers Crop; the model sees the cropped photo.

Step 2: make the clips that sell first

Start with the clips that sell, not the ones that impress. A turntable spin shows the back on a product page. An unboxing shows what arrives. Someone holding the product and talking works as a TikTok or Reels ad. A lifestyle shot on a shelf or a desk shows scale and use.

Each of these exists in RefVid as a template that opens the studio with the shots, length and shape already written, so your photos are all you add. A few of the 27 product templates:

Template	What you get	Length and shape
360 turntable	Your product spins slowly through a full turn on a white studio turntable	5 s, square 1:1
Unboxing	Hands open a kraft box, lift the product out, turn the label to the camera, with paper sounds	10 s, vertical 9:16
Water splash	The product drops into clear water in slow motion and settles upright	5 s, vertical 9:16
Perfume reveal	Mist and one puff of spray in warm backlight, then light through the glass onto the label	5 s, vertical 9:16
Jewelry on hand	A hand wearing your ring or bracelet turns slowly in warm window light	5 s, square 1:1
Tech reveal	A line of light traces your device in the dark, then the camera arcs around it	5 s, wide 16:9

Three of them, each with a reference photo and a frame from its real example clip. The products are AI-made samples; the prompts are the templates' own, word for word, with {1} where your product's name goes.

360 turntable: AERO 1, from 3 photos.

{1} on a round white turntable in a photo studio, rotating slowly through a full 360 degrees, light grey seamless background, soft even studio light, locked-off camera, sharp focus on every detail.

Every side faces the camera once as it turns, so photograph the front, both sides and the back at the same height; the back is the side a listing rarely shows, so include it, above all if it carries a label. 5 seconds, square, silent, on Kling 3.0 Omni: the cheapest of the three to make, and the template RefVid's paying customers make most. 360 turntable template.

Perfume reveal: NOIR 07, from 3 photos.

{1} stands on glossy black glass wrapped in slowly drifting soft mist; a single fine puff of perfume spray bursts from the top and catches a warm golden backlight, then the mist clears and a beam of light passes through the glass, ending on the label sharp and centred. Low-key, luxurious, slow push-in.

Shoot the bottle straight on with the label square to the lens and no glare: the clip ends on that label. 5 seconds, 9:16, with sound, on Kling 4. Perfume reveal template.

Jewelry on hand: HALO, from 4 photos.

Close-up: a woman's hand with a soft natural manicure, wearing {1}, rests on cream silk and turns slowly toward the camera, {1} catching small sparkles in warm window light. Shallow depth of field, gentle slow push-in, elegant and natural.

No hand model needed. Show the piece from the angle a wearer sees, plus the band and the setting; a photo of it on a finger helps the size look right. 5 seconds, square, with sound. Jewelry on hand template.

Step 3: let the photos describe the product, the prompt describe the rest

The prompt should cover the place, one motion, the camera move, the light and the last frame. Use the product's saved name instead of describing it again: a second description in words competes with the photos and is the most common reason a label changes. Two prompts that work:

AERO 1 slowly rotates on a round white plinth in a minimal studio, static camera at shoe height, soft key light, sharp focus

A hand lifts AERO 1 out of a kraft box on a wooden table and turns it toward the camera, soft daylight

Made from the three AERO 1 photos with a prompt close to the first one. Short, concrete prompts beat long lists of adjectives.

Pick the shape for where it goes

Square 1:1 for product pages, vertical 9:16 for TikTok, Reels and Shorts, wide 16:9 for your own site. Clips run 3 to 30 seconds on Kling 4 (up to 15 on Kling 3.0 Omni), can have sound, and one clip can hold up to three subjects, such as the product and the person holding it.

The same prompt and the same three photos, made vertical for social and square for a product page.

If the clip is going into paid placements rather than a product page, it is built in three beats: something that stops the eye, the product doing the one thing you sell, and a last frame on the label. We wrote that up in making a video ad of your own product from photos, with three ad formats from the same photos to test against each other.

Check every clip before it goes live

Watch each clip at full size. The label and logo should read the same as on the real item; small print is the part most likely to change, so keep the takes where it holds. Check the colours against the photo, and never show features or accessories the product does not have.

What it costs

New accounts start with free credits and no card: they cover a first 5-second clip on Kling 3.0 Omni (Vidu Q2 in some regions), and every clip, free or paid, has no watermark and may be used commercially. After that, clips are priced in credits by the second. A 5-second silent clip on Kling 3.0 Omni, such as the 360 turntable, is 155 credits; the same 5 seconds on Kling 4 at 720p, with sound, is 290. Here is what each way of paying buys, from the RefVid pricing page:

How you pay	Price	5-second Kling 3.0 Omni clips, like the turntable	5-second Kling 4 clips
One video, no plan	$2.99 up to 5 s, $4.99 up to 10 s, $6.99 up to 15 s	1	1
Credit pack, one time	$9.90 for 1,500 credits (also $19.90 for 3,200, $49.90 for 8,500)	9	5
Basic plan	$10 a month, or $96 a year, for 2,000 credits a month	12 a month	6 a month

Credits never expire: unused plan credits roll over and pack credits stay. Failed videos give their credits back, and unused pack credits are refunded in full within 14 days.

The full product photo to video guide lists all the templates with their lengths and shapes, and answers the questions shop owners ask us most, like whether one photo is enough and how to get a 360-degree spin.

— Louis Chi, RefVid

Kling 4.0 Omni Reference: the real limits, the prompt structure, and the photos behind each clip

Louis Chi — Fri, 06 Nov 2026 07:07:11 +0000

Kling 4.0 Omni Reference takes several pictures of one person or product at once and builds a new scene that keeps all of them: the face, the hair, the jacket, the label on the bottle. It reached everyone in Kling's own app on October 15, 2026, and its API opened on fal.ai and Runway Dev on October 23. We run it in RefVid, so this post is what we learned wiring it up: the limits that actually apply, how Kling wants the prompt written, and real clips next to the photos that made them. The full, regularly updated version lives on our Kling 4.0 Omni Reference guide.

Kling 4 in RefVid, from two reference photos of Mia, an invented character: 3 seconds, 720p, with sound.

Why a reference model, not image-to-video

Image-to-video animates one picture as the first frame. That is fine for one clip, but ask for a second scene and the model reinvents the subject: the jawline drifts, the hair parting flips, the logo on the mug changes shape. A reference model reads a small set of images as a description of who or what must appear, then builds the shot around that description. The more angles you give it, the less it has to guess.

One of Mia's four reference photos. The set: front, three-quarter, profile and full body, even light, plain wall.

Kling 4.0 vs Kling 3.0 Omni: the limits

From Kling's 4.0 user guide (updated November 4, 2026), the Kling 3.0 Omni guide, fal.ai's reference-to-video schema and Runway Dev's model list:

Limit	Kling 3.0 Omni	Kling 4.0 Omni Reference
Reference images	Up to 7 (images and elements together)	Up to 10
Reference videos	1, and then at most 4 images	Up to 5, 30 seconds combined
Elements	Counted within the 7	Up to 7, at most 3 from video
Voice profiles	Not listed	Up to 3
Inputs per task	Up to 7	Up to 15
Clip length	3 to 15 seconds	3 to 30 seconds
Aspect ratios	16:9, 9:16, 1:1	21:9, 16:9, 1:1, 9:16, Auto
Prompt length	Not listed	Up to 8,000 tokens
Output	720p, 1080p	720p, 1080p on fal.ai (4K in Kling's guide)

Per file, Kling 4.0 allows images up to 30 MB with an aspect ratio between 2:5 and 5:2, videos of 3 to 30 seconds up to 300 MB, and voice profiles made from 5 to 30 seconds of audio. Music and sound effects cannot be uploaded: audio input is voice profiles only, and the model makes the rest from the prompt.

Two things that surprised us. First, keyframes: in Kling's app, Keyframes (up to 10 images that steer how a shot changes) is a separate entry point and cannot be combined with Omni Reference in one task, but fal.ai's reference-to-video endpoint accepts a start frame, an end frame and up to 8 keyframes next to the references, all counted within its 10 images. Second, Kling 4.0 Flash also has Omni Reference, but only for 3 to 20 seconds at 720p, without start and end frames or keyframes.

What it costs per second

fal.ai: $0.157 per second at 720p, $0.235 at 1080p, so a 10-second clip is $1.57 or $2.35; with video references, $0.094 and $0.141.
Runway Dev: from $0.157 per second, up to 30 seconds at 720p or 1080p.
Kling's own developer platform: early access through a business inquiry, preview prices $0.14 and $0.21.
RefVid: 58 credits per second at 720p and 86 at 1080p on Kling 4, sound included; 31 credits per second on Kling 3.0 Omni.

If you are still choosing a tool rather than a model, we keep a side-by-side of ten image-to-video tools, sorted by the price of the cheapest paid plan, with clip lengths and how many pictures each one accepts.

How Kling wants the prompt written

Kling numbers your references in upload order: @Image1, @Image2, @Video1. Its guide asks you to give each one a job, for example:

@Image1 defines the woman; keep her face, hair and clothes.
@Video1: follow its camera path.

With bare tags and no jobs, the model guesses what each reference is for. The guide's prompt has four blocks:

Asset roles: what each reference defines and what to keep.
Overall settings: scene, texture, style, camera tone.
Sequence of events: what happens from 0 to X seconds, X to Y, and up to the end.
Audio: dialogue in quotes, ambient sound, music, effects.

Concrete details beat abstract words: "side light on a matte surface", not "premium look". Use negatives only for the two things the model adds on its own, subtitles and background music. To change an existing clip instead of making a new one, start with "Edit @Video1" and end with "Keep everything else unchanged".

Two examples: the photos, the prompt, the clip

In RefVid you type your subject's saved name and we swap it for the right @Element tag before the prompt reaches Kling 4. Below, {1} marks where the subject goes. Neither prompt describes the product; the photos carry its shape, finish and colours.

ATLAS, an AI-made watch, from 2 photos. Prompt, word for word:

Extreme macro of {1}: the camera glides slowly across the dial as the seconds hand sweeps, light rolling over the polished case and the hour markers, then pulls back to show the whole watch resting on dark leather. Warm low-key light, shallow depth of field, luxurious detail.

Reference 1 of 2: the dial straight on and glare-free, so the markers, hands and logo are legible.

The result: 5 seconds, square, with sound, on Kling 4. This is our watch macro template, which opens the studio with this prompt already filled in.

MARA, an AI-made handbag, from 2 photos.

A woman in a cream trench coat walks down a sunlit city street carrying {1} on her arm; the camera tracks alongside at hip height so {1} stays centred and sharp, swinging gently with each step, her face out of frame. Golden-hour light, shallow depth of field, fashion-ad style.

Reference 1 of 2: the front, plain backdrop, hardware and logo plate sharp.

5 seconds, vertical 9:16, with sound, on Kling 4: the handbag street template.

The pattern in both: where we are, what the subject does, then light, lens and mood in a few words. "Face out of frame" in the second one is deliberate: the bag is the subject, and an unreferenced face is one more thing for the model to invent.

Preparing references that work

These rules hold for Kling 3.0 Omni and Kling 4.0 alike, so a set prepared for one works on the other.

Three to five images. A person: a clear front view, a three-quarter view and a profile. A product: the hero angle, the back or side, and a close-up of the logo or texture.
Even light, simple background. A plain wall helps the model separate the subject from the scene; busy backgrounds leak into the output.
The same subject in every image. Same outfit, same hairstyle, same colourway. Two outfits means two characters.
No filters, watermarks or text overlays. The model treats them as part of the subject.

Where it still slips

Small text on products, such as ingredient lists and fine print, is often redrawn. Hands handling small objects can still look wrong. Angles that none of your references show, like the back of a head or the underside of a shoe, are invented rather than copied. The fix is almost always better references, not a longer prompt.

What it is good at: the same face, scene after scene

Where a reference model earns its cost is anything viewers watch more than once: a presenter who opens every tutorial, a product shown in five rooms, a cast that has to survive the cut between scenes.

The same four photos of Mia, three scene prompts, the same face each time (Kling 3.0 Omni in RefVid, before we switched the default to Kling 4).

That property is what a short series needs. We wrote up how to keep the same actors in every episode of an AI drama series: casting from saved photos, formats that suit one-to-two-minute vertical episodes, and writing scenes with the dialogue in quotes so the model can shoot them.

How we use it in RefVid

You save a character or product once from 1 to 10 photos. It goes to Kling 4 as one element: the first photo as the front view, up to three more as other angles, the rest as extra reference images. Then every clip starts from that saved subject; you write the scene, pick a length and a shape (16:9, 9:16 or 1:1), and generate. Kling 4 is the default and Kling 3.0 Omni stays as the cheaper choice; subjects saved before the switch carried over unchanged. New accounts get one free 5-second video, on Kling 3.0 Omni (Vidu Q2 in some regions), no card needed.

The full guide has the source for every number above, a third prompt example and an FAQ, and we update it when Kling or fal change a limit.

— Louis Chi, RefVid

How to crop a phone screenshot into a clean product photo before making an AI video

Louis Chi — Wed, 04 Nov 2026 19:06:25 +0000

This post was written with AI help (Claude), from our own screenshots and screen recording of the RefVid studio; every step shows the real UI.

A phone screenshot of a product brings the status bar, the app's buttons and the price into your AI video. RefVid now lets you crop any reference photo before you generate: the video model sees the cropped photo, not the original, so the product stays clean and the same in every shot.

Here is the whole thing in a 3-minute video, then step by step with a screenshot each. Full written guide with the transcript: Crop product image for AI video.

1. Add the photo, or paste it

In the studio, click New element and add a product photo, or paste one with Cmd+V (Ctrl+V on Windows). When a photo has a phone screenshot's shape, RefVid says so and offers Crop.

2. Point at a photo and click Crop

Crop sits in the middle of every photo tile when you point at it, not only the main one. The × that removes the photo stays in the corner.

3. Start from the whole photo

The crop dialog opens on Original, with the box around the whole photo. Drag the box to move it; on Original a corner resizes it freely.

4. Mind the yellow note

Pull an edge to trim one side. A crop longer than 2.5 : 1 gets a yellow note that it will be padded; under 300 px it can't be applied. Back in range, the note fades.

5. Pick a shape and frame the product

Choose 1:1, 9:16, 16:9, 3:4 or 4:3 and the box glides to that shape. Drag it onto the product. In a shape, a corner keeps the ratio; an edge frees it and turns the shape off.

6. Apply

The cropped photo fades into its tile. RefVid keeps the original.

7. Reopen or reset

Crop again starts from the original with your last crop on it; Reset to original brings the whole photo back. New element keeps the photos, the crop and your words across a page refresh.

8. The same in the home page box

In the box on the home page, point at the photo pile to spread it out. Each photo has Replace and Crop side by side.

Quick answers

Does cropping change the video's aspect ratio? No. A crop changes only the reference photo; you pick the video's shape separately in the settings.

Is the original lost? No. Reopening Crop starts from it, and Reset to original brings it back.

Which photos work best? One product, sharp and well lit, on a plain background; front, side and back views. More in how to shoot product reference photos.

Try it in the RefVid studio.

How we made an 8-second AI comedy short with two consistent characters, start to finish

Louis Chi — Tue, 03 Nov 2026 11:54:36 +0000

I wrote this with AI assistance (Claude) and checked every number against our own notes and files. All pictures are from our own clip and our own files.

The short this story is about. Tap to watch it on our site.

I'm Louis Chi, and I work on RefVid, a tool that keeps the same character in every AI video shot. On 3 November we made an 8-second comedy short with our two recurring characters, Leo and Mia.

It took one generation, kept on the first take, and about an hour from idea to finished clip. Here is every step in order.

1. The idea, in one line

A husband calmly grabs a drink from the fridge while, behind him, his wife runs past already decorating for Christmas, because for her the season starts on November 1st.

The gag came from our morning look at what was rising that week; thequistfamily posted it first. One line is the test: if the joke can't be said in a sentence, it won't survive 8 seconds of video.

It also has what makes a short easy to generate. One room, nobody speaks, two people, one punchline at the end.

The gag in our clip: he faces the camera with a drink, she is already decorating behind him.

2. The characters: Leo and Mia, four photos each

The roles were already cast. Leo and Mia are saved in our studio, four photos each, and they play a couple well.

The same faces matter twice. Inside the clip, the man who opens the fridge in shot 1 has to be the man who stares in shot 3, or the joke breaks at the cut. Across clips, recurring characters are what people recognise.

Mia's photos show her in a jacket, so the script had to name her costume clearly enough to win against them.

3. The script: beats, camera, set, wardrobe, timing

We wrote a full script before any prompt. First the beats, read from the original frame by frame: he opens the fridge, turns to the camera with a can, she runs past behind him at about 4 seconds, he stares after her to the end.

Then each shot got its camera, set, light, wardrobe and action:

Camera: a phone at chest height on a tripod, never moving.
Set: a bright modern kitchen in daylight, white cabinets, sage-green walls, a steel fridge at the right edge, bare counters.
Wardrobe: Leo in his navy shirt; Mia in a red fleece onesie from neck to ankles, an antler headband, a garland on one shoulder, dragging lit Christmas lights.
Timing: three shots of 2, 3 and 3 seconds, a hard cut between them, something moving in the first second.

The script went through three versions. v1 was guessed from the caption (a Christmas tree frozen in a block of ice) and read the wrong thing as the joke; we stopped before spending anything. v2 had the real gag. v3 is v2 after two reviews.

The three versions, all written before anything was generated.

4. Two reviews before spending a credit

Two separate AI sessions reviewed the script. An AI-video director asked whether the clip tells the joke: 8 faults, then 2, then clear. A Kling 4 prompt engineer asked whether the model will draw what the words say: 7 faults, then clear.

The fixes that mattered most:

Shot 2 was 498 characters; RefVid takes 480 per shot in a several-shot Kling clip. Cut to 478.
Shot 3 had Leo turn back to the lens, a different joke. Now he keeps staring after her.
"At 1 second Mia sprints…" became "Then Mia … sprints…": a timestamp inside a shot is not a reliable cue.
"In a flash" can be drawn as a flash of light. It became "sprints fast in from frame left along the back wall".
"Red fleece onesie" became "red fleece onesie neck to ankles", so her jacket from the photos doesn't come back.

All 17 faults, in the order they were found.

5. From script to shot prompts

Each shot became one prompt: camera, place, who is there and what they wear, then the action in order, ending in a state the model can hold. With no first frame, shot 1 also describes the whole kitchen.

Shot 1, 2 s: "Medium shot, phone camera at chest height. A bright, clean modern kitchen in daylight: white cabinets, sage-green walls, bare counters, a stainless-steel fridge at frame right, an empty open doorway in the back wall. Leo, in his navy shirt, stands alone, side-on, hand on the fridge handle. Right away, in one quick pull, he yanks the fridge door wide open, cold light spilling out, and holds it open, staring into the fridge. Fixed camera."

Shot 2, 3 s: "Same kitchen and shot, fridge closed. Leo, left of center, faces the camera, silver can at his chest, closed-mouth smirk. Then Mia, her reference face and copper bob, in a red fleece onesie neck to ankles, antler headband, garland on one shoulder, sprints fast in from frame left along the back wall, dragging glowing multicolor Christmas lights, out of frame right; the lit string stays on the floor. Leo snaps his head to frame right and holds. Fixed camera."

Shot 3, 3 s: "Medium close-up from the waist up, the same kitchen behind Leo, who stands alone. Behind him one string of glowing multicolor Christmas lights lies in a long line across the floor. Leo keeps his body toward the camera and his head and eyes turned to frame right, staring blankly, mouth slightly open, silver can at his chest, completely still. He blinks once, slowly, and keeps staring. Fixed camera."

Three habits carry over to any clip: the same words for the same thing in every shot, speed written on the verb, and "Fixed camera." at the end.

What each prompt gave, at 0.5 s, 3.0 s and 6.0 s.

6. Generation settings

In the studio: Leo, plus Mia as a second subject, three shot boxes of 2, 3 and 3 seconds. Kling 4, 1080p, 9:16, 8 seconds, no first or last frame. The prompts were 446, 478 and 412 characters with the names counted as reference tokens.

It cost 688 credits, and we generated once.

Everything we sent, on one card.

7. The frame-by-frame check

We read the take at 4 frames a second. The cuts landed at 1.8 and 4.8 seconds, both faces matched their photos, no extra people appeared.

Four things differ from the prompts. Mia came in from the right, waved and ran back out the same side; Leo's smirk became an open smile; at the end he turns to the lens and blinks at 7.3 s.

The test for keeping a take is the one-line joke: calm husband in front, Christmas already happening behind him, a stunned stare. All there, so we kept it. A retake would have cost another 688 credits for the same joke.

The whole take at 4 frames a second.

8. The edit: trim, caption, sound

No trim: the stare held for the full 3 seconds. The caption "My wife every year on November 1st 🎄" went on in the edit, white bold with a thin outline near the top, never in the prompt.

Kling 4 always makes its own audio, and we took it out. The version on our site and on YouTube has our own arrangement of Jingle Bells, a public-domain tune; on Instagram and X it carries the original clip's sound; on TikTok, a Christmas song from the platform's licensed library.

The caption is the only thing added to the picture.

9. Sharing it

The clip was ready at 08:15 UTC, about an hour after that morning's look at what was rising. It went out on YouTube Shorts, Instagram Reels, X and TikTok between 08:26 and 08:42, and on its own share page with the three prompts underneath.

Try it yourself

Save three or four photos of each person once, write your gag in one line, then as two or three shots with the names in the prompts, and generate. The full making-of, with every picture and the exact prompts, is on our site.

— Louis Chi, RefVid

We remade a viral TikTok trend with AI in one take: every step and the exact Kling 4 prompt

Louis Chi — Tue, 03 Nov 2026 09:12:33 +0000

I wrote this with AI assistance (Claude) and checked every number against our own files. The stills are from our own clip; the original TikTok is linked, not shown.

Our remake. Tap the picture to watch the 8-second clip.

I'm Louis Chi, and I work on RefVid, a tool that keeps the same character in every AI video shot. On 3 November we remade a trend that was taking off on TikTok, with our two demo characters, Leo and Mia. It took one generation and 688 credits.

This is the whole process, including the part where our first idea was wrong.

1. The trend

"My wife every year on November 1st" is an 8-second, one-take TikTok by thequistfamily. It was posted on 1 November and had 1.4 million views two days later. A husband calmly takes a drink from the fridge while his wife, in a red reindeer onesie, sprints past behind him trailing lit Christmas lights.

A second couple had posted the same format with the same punchline (1.3 million views). That told us it was a format people were copying, not a one-off.

The original's account on 3 November: the clip, outlined, at 1.4M views. We blurred the thumbnails; the video itself is only linked.

2. Read the frames, not the caption

Our first concept was guessed from the caption and one quick watch. It would have been a different joke:

Mia yanks a freezer open and mist rolls out; inside is a block of ice with a tiny Christmas tree frozen in it. She hugs the block, turns to the camera and grins as it starts to drip.

So we downloaded the original and made a contact sheet at 4 frames a second, plus 8 frames a second around the 4-second mark. The drink is only the setup. The punchline is the wife dashing past behind him at about 4 seconds, and we stopped before generating anything, so the wrong idea cost nothing.

A 4-frames-a-second contact sheet, here of our own take. We made the same one of the original to find its beats.

The real gag, as we remade it: he faces the camera with a drink…

…and she is already decorating behind him.

3. Copy the beats one to one

We changed only the people: Leo plays the husband, Mia the wife. Here is the original next to our take.

Original 0.0–0.5 s: he stands at the fridge, hand on the handle. Ours 0.0–0.15 s: the same.
Original 0.5–3.0 s: opens the fridge, takes out a can, closes the door. Ours 0.15–1.8 s: yanks the door open and reaches in.
Original 3.0–4.0 s: turns to the camera, small smirk. Ours 1.8–2.3 s: hard cut, he faces the camera with a silver can.
Original 4.0–5.0 s: his wife dashes past behind him with a garland, trailing lit lights. Ours 2.3–3.6 s: Mia runs through behind him, trailing lit lights.
Original 5.0–6.0 s: he turns to look where she went. Ours 3.6–4.8 s: he looks after her; the lights stay on the floor.
Original 6.0–8.2 s: he stares off to the side. Ours 4.8–8.0 s: hard cut to a closer shot, he stares, mouth open, and blinks once.

The run lands about a second earlier than in the original. That was on purpose: we wanted the remake a little faster.

4. Three shots, two hard cuts, one generation

The original has no cuts. We split ours into three shots of 2, 3 and 3 seconds inside one 8-second Kling 4 generation, each with its own prompt, and the model put a hard cut between them.

Why not four shots? Our studio takes up to three shots per clip, and in the original the smirk and the run happen in the same frame. Splitting them would have killed the joke, so the smile and Mia's run share the middle shot, filmed from the same spot as the first.

One key frame from each shot: 0.5 s, 3.0 s and 6.0 s.

5. Two review passes on the script

Before spending anything, the script went through two independent AI reviews. An AI-video director pass found 8 faults, then 2, then none. A Kling 4 prompt-engineer pass found 7, then none.

The script comes before any prompt, shot by shot with camera, set, light and action. An excerpt, shot 2's camera note:

Camera: identical position and lens to S1 (a jump cut in time, like a phone video
trimmed) — makes the background sprint readable because the doorway/kitchen
behind Leo is unchanged.

The biggest fix was the last shot. Before the review it told a different joke:

BEFORE  Tight close-up of Leo's face, the kitchen blurred behind him, a string of
        glowing Christmas lights now lying across the floor in the background.
        Leo turns back to the lens with a flat, deadpan stare, mouth closed, and
        holds completely still.

AFTER   Medium close-up from the waist up, the same kitchen behind Leo, who stands
        alone. Behind him one string of glowing multicolor Christmas lights lies in
        a long line across the floor. Leo keeps his body toward the camera and his
        head and eyes turned to frame right, staring blankly, mouth slightly open.

After the fix: he keeps staring after her, as in the original.

The other fixes that mattered:

The length limit. Our studio allows 480 characters per Kling 4 shot once names become reference tokens. Shot 2 was 498 and came down to 475.
The sprint. "In a flash" could read as a flash of light and gave Mia no entrance. It became "sprints fast in from frame left along the back wall … out of frame right".
Small ones. The fridge is closed when shot 2 starts, so the door can't hide her exit. "Silver can" and "glowing multicolor Christmas lights" are worded the same in every shot, and the lights get an end state: "the lit string stays on the floor".
No big grin. A grin at running speed in the background melts, so the prompt only asks that her face reads as Mia.

Mia at full speed: her face holds.

6. The exact prompts

Settings: Kling 4, 1080p, 9:16, 8 seconds (2 + 3 + 3). Leo and Mia are saved characters with four photos each. There was no first frame, so the kitchen comes from the first prompt.

Shot 1, 2 s:

Medium shot, phone camera at chest height. A bright, clean modern kitchen in daylight: white cabinets, sage-green walls, bare counters, a stainless-steel fridge at frame right, an empty open doorway in the back wall. Leo, in his navy shirt, stands alone, side-on, hand on the fridge handle. Right away, in one quick pull, he yanks the fridge door wide open, cold light spilling out, and holds it open, staring into the fridge. Fixed camera.

Shot 2, 3 s:

Same kitchen and shot, fridge closed. Leo, left of center, faces the camera, silver can at his chest, closed-mouth smirk. Then Mia, her reference face and copper bob, in a red fleece onesie neck to ankles, antler headband, garland on one shoulder, sprints fast in from frame left along the back wall, dragging glowing multicolor Christmas lights, out of frame right; the lit string stays on the floor. Leo snaps his head to frame right and holds. Fixed camera.

Shot 3, 3 s:

Medium close-up from the waist up, the same kitchen behind Leo, who stands alone. Behind him one string of glowing multicolor Christmas lights lies in a long line across the floor. Leo keeps his body toward the camera and his head and eyes turned to frame right, staring blankly, mouth slightly open, silver can at his chest, completely still. He blinks once, slowly, and keeps staring. Fixed camera.

Swap in your own character names to reuse them.

7. One take, and what came out differently

We generated once and kept it. The joke survived, both faces matched their photos and no extra people appeared.

Four things differ from the prompt. Mia came in from the right instead of the left, stopped to wave and ran back out on the same side. Leo's smirk came out as an open smile, and in the last shot he slowly turns toward the lens and blinks at 7.3 s.

None of it hurt the joke, so there was no retake and no trim. The clip runs the full 8.0 seconds and cost 688 credits: 8 seconds of Kling 4 at 1080p, which always comes with sound.

She leaves on the side she came in (3.5 s)…

…and at the end he turns to the lens (7.6 s). The prompt asked for neither; we kept both.

8. Caption and sound

The caption goes on in the edit, never in the prompt: "My wife every year on November 1st 🎄", white with a thin outline, near the top. Kling 4 makes its own audio, and we took it out.

On TikTok the clip went up with the trending "It's time" sound from TikTok's library, and on Instagram and X with the original's own sound, the way the trend's viewers know it. On YouTube and on our site it has our own arrangement of Jingle Bells, a public-domain tune.

What I'd tell anyone copying a trend

Download the original and read it frame by frame before writing a word. Copy its beats and its timing, change only the people, and let a second pair of eyes (human or AI) attack the script before you pay for a generation.

The full making-of, with the beat table and every still, is on our site.

Louis Chi works on RefVid, AI video with the same character in every shot.

How to make a product video from your product photos, free (step by step, with screenshots)

Louis Chi — Sun, 01 Nov 2026 15:08:35 +0000

If you sell something online, you've probably been told you need video: on the listing, in the ads, on TikTok. And if you've ever priced a product shoot, you know why most small shops don't have any.

I'm Louis, and I run RefVid. We just rebuilt our free generator page so the whole thing happens on one screen: you add photos, pick a template and press Generate, and you only sign in at that last step. This post walks through it with screenshots of the real page and clips made from product photos. The full guide, kept up to date, is on our site.

What you get for free

Your first 5-second clip on Kling 3.0 Omni is free once you sign in with Google or email. It's 720p and silent by default, or you can spend the free credits on one clip with sound or at 1080p instead. No card, no watermark, and you can use it commercially. Five free images come with it.

Here's what three photos of a shoe turn into with the Product 360 template, at exactly those free settings:

Step 1: pick a product template

Open the free AI video generator. Under the heading there's a row of templates, product ones first. Press Use on one and its prompt fills the box below.

Pick by what the video is for: a 360° turntable or a floating product shows the item itself, a water splash or bullet time grabs attention in an ad, a hand holding the product reads like a customer's video. The turntable is the one people pick most: 24% of all videos finished on RefVid so far, per our AI product video statistics.

Step 2: add one to ten photos

Click Reference and add 1–10 photos. One sharp photo works; three are better: the front, a three-quarter angle, and the label up close. The model gets all of them with every prompt, which is why the label and shape come from your product rather than from a guess.

Plain background, crop away big empty margins, and only use photos you have the right to. (If you're shooting them at home, I wrote up how we take product photos for AI video.)

Step 3: check the prompt

The template's prompt already works. Five seconds hold one action and one camera move, so describe the scene, the motion and the light, and leave the product's look to the photos. This is the prompt behind a real water-splash clip on our site, with the product's name replaced by "your product":

Slow motion: your product drops into a tank of crystal-clear water against a soft blue gradient background, a burst of bubbles and droplets exploding around it, then it settles upright with the label facing the camera. Bright studio light, macro detail.

Step 4: shape, length and sound

For the free clip, leave it on Kling 3.0 Omni, 5 seconds, 720p. Choose the shape by where it's going:

Where it goes	Shape
Store listing, Instagram feed	1:1
TikTok, Reels, Shorts	9:16
Website banner, YouTube	16:9

Step 5: Generate, then sign in

Press Generate. If you're signed out, a sign-in sheet opens over the page (Google or email). When you come back, your photos and prompt are still in the box and the video starts by itself. That was the part we cared about most: nobody should have to fill a form twice.

It usually takes 2 to 10 minutes; the clip shows up on the page and you get an email.

Step 6: download and check

Download the MP4. Then watch it full size, frame by frame: label text, product edges, hands if there are any. Models still slip sometimes, and it's much cheaper to catch it before a customer does. Label realistic AI video as AI where the platform asks.

That's the whole flow. If you try it, I'd like to hear where it got in your way: what you expected to happen and didn't.

This post was written with AI assistance and checked against our live product.