Wan 3.0 AI Before and After Video Generator
You cannot go back and film the before. That is the whole reason it is sitting in your camera roll as a photograph — the drive is clean now, the wall is painted, the headlight is clear, and the version anyone would have wanted on video stopped existing the moment you fixed it.
Both ends already happened. The seconds between them are the only part generated here: your before pinned as the first frame, your after pinned as the last, and the change rendered with the sound of it changing in the same file.
Your first one is free. No card.
Alibaba's own published first-and-last-frame material for Wan 3.0 — the prompt verbatim, and the specs `ffprobe` reads off the file they released with it. Their example asks for a fifteen-second one-take with the guzheng, the water, the birdsong and a recited line all named in the same prompt; the file comes back with an audio track.
Alibaba's own first-and-last-frame example, with its sound written into the prompt
A morph is a sound as much as it is a picture.
That is the part everything else drops.
A before and after is not a layout. Set the two photographs beside each other and a viewer reads precisely that: two photographs, chosen by someone with an interest in the comparison. The format earns its attention only when one frame becomes the other while they are watching — and it earns it twice when you can hear it happen.
Frame one, pinned
The change, and the sound of it, in the same render
Watch any good transformation with the audio off and the join stops working. The change needs a transient on the frame it happens — a hiss, a rush of water, a hard mechanical release — and on every other route to this format that is a second purchase, dropped in afterwards and nudged into place by hand. Wan 3.0 samples the sound with the picture, which is also why turning it off does not make the clip cheaper.
The other half is what you hand over. first_frame takes the photograph of how it was, last_frame takes the photograph of how it ended up, and the render is the travel between two states you already had evidence of. That matters more here than anywhere else on this site, because a tool that takes one photograph and produces the second for you has invented the result your ad is about.
The trade is exact, and better known before you upload than after. Pinned frames are their own family, and that family will not mix with reference material — attach a first or last frame and you cannot attach a separate product photo, face or clip in the same run. In a before and after that costs almost nothing, because the product is already in both photographs. It costs you something in one case: if you wanted a named creator on camera as well, that is a different shape of ad, and it is a person in a room holding the thing.
The model repaints every pixel. What comes back resembles your photographs closely — it does not reproduce them. Hold onto that distinction; it is what keeps a before and after ad out of trouble later. How the two frame modes work covers the mechanics.
- Photographs you already own
- 2
- Longest single clip, natively multi-shot
- 30s
- The picture and the sound of the change
- One pass
- Watermarks, on every tier
- 0
Photographs you already own
Longest single clip, natively multi-shot
The picture and the sound of the change
Watermarks, on every tier
- No slider and no split-screen layout — both ends go to the model, and the change happens inside one frame.
- No presenter narrating the transformation — if you want it spoken, that is the UGC ad page.
- No invented after — no second photograph means you describe the ending instead, and you say so.
- No burnt-in captions — the MP4 arrives clean, so the caption goes on last, where you can still fix a word.
Four ways to hand over the before and the after
All four run the same model at the same per-second rate. What changes is how much of the ending you are still asking the model to decide — and in a before and after the ending is the claim, so that is not a stylistic question.
- Both states pinned · 1 image each
Pin both ends
The photograph of how it was in one slot, how it ended up in the other. The clip has to open on your first image and close on your second, and the render is the path between. The most controlled route here, and the only one where nothing about the result is the model's idea.
- Start exact, finish written · 1 image
Pin the before, describe the after
One photograph and a sentence. Use it when the second shot does not exist yet, or exists but was taken from somewhere else entirely. The opening is exact and the ending is written — which also means the ending is a description, and what you can honestly describe has a floor under it.
- Borrow the timing · up to 5 clips, 15s total
Borrow a move you like
A clip off your swipe file whose transition works, handed over as a reference so the timing and the camera behaviour come from it rather than from a description. A reference clip is reference material, so this route cannot be combined with route 01 — borrow a move or pin two ends, not both.
All four bill identically: 8 credits a second at 480P, 16 at 720P, 32 at 1080P. Two pinned frames cost the same as none. Every limit and rate, per second.
Your before and after photos were taken
on different days. That is the problem.
Both photographs are real, both are yours, and they still may not cut together. The reason decides whether tonight's render works or just costs 40 credits: the model is not comparing your two images — it is travelling from one to the other. Everything different between them is something it has to cross, and it will invent a camera move to get there.
- Viewpoint
Where you stood
- What has to match
- Where you stood
- Why it matters
- The two frames are one shot's endpoints, so a different viewpoint is a move the camera has to make
- What you get when it does not
- A swing or an orbit nobody asked for, which reads as a cut
- How to fix it tonight
- Re-shoot the after from the original position — the after is still there to photograph
The single highest-return fix on this page, and the one everybody skips.
How close you were
- What has to match
- How close you were
- Why it matters
- Distance is the same problem in one dimension
- What you get when it does not
- A push or a pull arrives to reconcile the two framings
- How to fix it tonight
- Stand at the same distance, or crop both images to the same framing before uploading
Cropping is free and fixes most pairs that are otherwise fine.
The light
- What has to match
- The light
- Why it matters
- A change in light is itself a visible change, and the viewer credits it to the product
- What you get when it does not
- The improvement looks like better lighting — the one reading this format cannot survive
- How to fix it tonight
- Shoot the after at the same time of day, or in the same shade
The most damaging mismatch, because it undermines the claim rather than the craft.
One thing held still
- What has to match
- One thing held still
- Why it matters
- A change is only legible against something that does not change
- What you get when it does not
- The whole frame moves and there is nothing to measure the change by
- How to fix it tonight
- Name it in one clause: the door frame, the badge on the grille, the pot on the left
Costs one sentence in the prompt and does more than any other sentence in it.
You cannot re-shoot the before. You can almost always re-shoot the after. The wall is still painted, the car is still clean. Stand roughly where you stood, hold the phone roughly where you held it, wait for roughly the same light, and take the second photograph again — four minutes, and the pair travels between itself with no invention required.
If your before is a sample you can make again, do the opposite: make the second one match the first.
And for the next job: when you take a before, you now know you will want the after to match it. A mark on the floor, a doorway edge in the corner of frame, a note of where you stood. What a product photo needs before you upload it.
Six ways a before and after stops being usable
None of these are about taste. Each one is a reason the finished file cannot go into an ad manager, which makes it a reason the credits are gone.
Your vertical ad came back horizontal
On adaptive, the shape is read off the frame you pinned, and Alibaba recommends adaptive for this mode precisely because the frames should decide it. Before photos are overwhelmingly landscape phone shots. Their own published example goes in at 2560×1440 and comes back 1920×1080.
The shape of your ad is decided in your camera roll, not in the generator. Pin a vertical before, or crop both photographs to 9:16 before uploading. Forcing the ratio against a landscape frame only crops or pads it.
Flagged on uploadIt reads as a cut, not a change
The two photographs disagree about viewpoint, distance or light, and the model built a camera move to reconcile them.
Re-shoot the after to match the before, or crop both to the same framing, or drop to one pinned frame and describe the ending.
Free to fixThe photo was refused before anything ran
Each image has to be 240–8000px on every side, inside an 8:1 ratio, 20MB or under, in JPG, PNG, BMP or WEBP. A PNG with transparency is refused, and so is the HEIC an iPhone saves by default.
Settings › Camera › Formats › Most Compatible so new photos save as JPEG, and export the older ones. A screenshot is the quickest fix for an oversized file.
Checked before you spendThe uploader will not take your product shot as well
Pinned frames and reference material are two families, and a request mixing them is refused rather than resolved.
Pick one. A before and after almost never needs both — the product is in both photographs already. If it is not, pin nothing and hand over references instead.
Blocked in the uploaderThe number on the label is mush
Small type is where generative video is weakest, and the render redraws every pixel rather than compositing your image through.
Keep the claim out of the rendered frame. Shoot it as a close-up, say what has to stay readable, and put any figure on as a caption in your editor — which is also what the rules want.
Shot choiceThe take you liked has vanished
A finished clip comes back as a link that expires 24 hours after the run.
Download it, or run it signed in — a signed-in run is kept in your library instead of behind a link with a clock on it.
Save it
A run that errors out is refunded automatically. One that completes and is simply wrong is not — which is the whole argument for trying a before and after at 480P first, where the same model, the same pair and the same sound cost a quarter of the finished version.
A before and after is a claim,
and two photographs make it before you write a word
Worth reading even if you never generate anything, because it already applies to the before and after photographs on your feed. This format states a result in pictures without ever writing the sentence down — and the regulator reads the pictures.
- The images are the claimThe FTC works through an ointment advertised as maintaining "joint health and mobility", shown with a before photograph of a woman using a walker and an after of her dancing: "the before-and-after images, along with the references to joint health and mobility, likely convey a claim that the product can dramatically improve the symptoms of arthritis." Nobody wrote "cures arthritis". Two pictures did.
- "Results may vary" buys nothingThe same guidance: "attempts to disclaim dramatic results with statements like 'Results not typical' don't cure the deception." A before and after pair is read as representative of what people generally get. If your best job is not your average job, show the average one — or print the real figure beside the clip, in the same size as everything else.
- The photographs are your evidence, the video is your adThis follows from something mechanical rather than legal: every pixel is redrawn, so a generated clip resembles your photographs and cannot be produced as a record of them. Keep the two originals, untouched, with their dates — that is what proves the result happened. Generate the advertisement for it freely, and never let the generated file be the only copy.
- Which ones you can actually runCleaning, rust and stain removal, restoration, detailing, paint, flooring and landscaping are the best fit there is — an object changed, visibly and objectively. Renovation, organisation and styling work the same way. Cosmetics, colour and nails work if you say how it looks rather than what it treats. A product with measured performance works if you print the measurement.
- Which ones not to generateWeight loss, body shaping, skin, teeth, hair regrowth and anything medical. Meta requires dietary, health and weight ads to be targeted to people 18 or older and separately prohibits imagery such as a close-up on a body area with pinched fat; what it does allow is an ad that illustrates "people using the product or the service, and its impact of using it". Show the product and how it is used instead — the product itself, shot well, or a creator explaining it on camera.
- Say that it was generatedEvery clip from here is completely AI-generated, which puts it inside the disclosure most platforms now ask for. TikTok Ads Manager carries a setting reading "This ad contains AI-generated content" for creative that is completely AI-generated or significantly modified by AI. Check it on every campaign, including duplicates.
- OwnershipClips you generate are yours to publish, sell and bill a client for on every paid plan, with no watermark and no separate licence to buy, under Alibaba's usage policy for the Wan 3.0 model and our terms. The result you depict, and any figure on screen, stay yours to substantiate — exactly as they would be with a camera.
Two settings, two problems: the AI-content disclosure is not the same switch as branded-content disclosure, and a before and after usually needs to think about both.
A ten-second before and after
is 160 credits
Billing is per second of output, so length and resolution are the only two dials that move the number — two pinned frames are not an upload fee, the vertical cut is not a surcharge, and the sound is not an add-on. A ten-second before and after is 80 credits at 480P, 160 at 720P and 320 at 1080P. Six five-second tries at 480P is 240 credits, less than one finished ad at 1080P. Find out whether your pair travels, then finish it once.
What a before and after costs, by shape
- Billed by
- output second
- One 5s try, 480P
- 40 credits
- Six 5s tries, 480P
- 240 credits
- One 8s clip, 480P
- 64 credits
- One 10s clip, 720P
- 160 credits
- One 10s clip, 1080P
- 320 credits
- Free on signup
- 1 clip + 24 credits
$39.90 a month on Pro is 2,240 credits — fourteen finished ten-second clips at 720P, or fifty-six tries at 480P. Set that against what this format actually costs you today: the after can be re-shot for nothing, and the before cannot be re-shot at any price, which is why it has been sitting in a camera roll instead of running as an ad. Every plan and pack · what a failed run costs.
الأرصدة الشهرية نفسها في الحالتين
How to make a before and after video tonight
Five steps. The first three cost nothing, and they are the three that decide whether the other two were worth running.
Go find the pair that already matches
Open the camera roll and look for one job where the two shots happen to agree on viewpoint, distance and light. There is usually one. Start with that job rather than your best result — you are testing the format, not the portfolio.
FreeRe-shoot the after
Four minutes, standing where you stood the first time. The single highest-return thing on this page, and the step everybody skips. While you are there: if the ad is going to Reels, take it vertical, because the shape of the clip is read off the frame you pin.
FreeName the anchor and the join
One clause for the thing that must not move — the door frame, the badge, the pot on the left. One clause for how the change travels: what dissolves, wipes away, dries, lifts or peels. Then name what it sounds like, because the sound is rendered in the same pass and a silent morph reads as an edit even when it is not one.
FreeRun it at 480P for five seconds
40 credits. You are looking for one thing only: whether the frame travels from your before to your after without the model inventing a camera move. Grade, polish and length are all later problems.
40 creditsRe-run the keeper at full length and 720P
Ten seconds at 720P is 160 credits. Keep the whole shot list inside thirty seconds — timecodes past the ceiling are discarded rather than trimmed, so a call to action written at 0:35 does not get shortened, it does not get filmed.
Before the media budget
Every step above moves a decision earlier, to where changing your mind is still free. Prompts with the clip each one produced.
AI before and after video generator questions
What it takes in, what one clip costs, and what the rules require before it runs as an ad.
What is an AI before and after video generator?
Can I use my own before and after photos?
My before and after were shot from different angles. Will it still work?
Why did my before and after video come out horizontal?
Can I pin both frames and attach my product photo too?
How long can a before and after video be?
Does the transition come with sound?
What does one before and after video cost?
Is there a free before and after video generator here?
Should I use this or a before and after slider?
Can I run a generated before and after as a paid ad?
Can I make a weight-loss or skin before and after?
You already have the two photographs.
Pick the job where the before and after happen to match, put them in, and watch five seconds with the sound up. The first one is free — no card, no watermark, and yours to run commercially.
كتبه ويحافظ عليه فريق التحرير في wan-3.runنُشر في آخر تحديث إصدار الأداة 2026.09.4