Pick what you need — a photo, or a photo plus video — a model gender and quality tier, and upload a photo of the clothing product. Each result is ready to download as soon as it's generated.
Setup
Model, garment & API key
Everything below is required before generation can start.
Sets the image and video models automatically. Each video clip is 8 seconds (Veo's per-call limit); a second clip is optional on the video step.
Photo credit pricing
Tier
Photo (dress the model)
Standard
…
Pro
…
Ultra
…
Uploading your own master model photo instead of generating one is free — this is the only guaranteed cost for a photo.
Video credit pricing
Tier
Product sheet
Video clip (8s)
Sheet + 8s video
Sheet + 16s video
Standard
…
…
…
…
Pro
…
…
…
…
Ultra
…
…
…
…
Assumes you already have (or generate) a photo of the dressed model — that cost is shown in Photo credit pricing above.
Talking Model pricing (Kling AI)
Length
Price
8 seconds
₹170
16 seconds
₹320
Click to upload a clear photo of the garment (flat-lay or on a hanger works best)
Step 1 of 4
Master model
A neutral, photorealistic base model in a plain studio outfit. Already generated one recently? Pick it below and skip generating again.
The master model wearing the exact uploaded garment — same face, same product, fitted with realistic fabric physics.
The front three-quarter pose is always included. Tick Side and/or Back to add those angles too — pick either, both, or neither. Charged per photo generated.
Your photo is ready — download it above. That's everything for a photo-only run.
Click your favorite to make it the one used for Product Sheet and Video — you can still download all four either way.
Your photos are ready — download above, or continue with your selected favorite.
Step 3 of 4
Product sheet
A single 16:9 image, 10 panels, covering hero shots and construction macros (collar, buttons, print, sleeve, cuff). This sheet grounds the video in Step 4.
A short cinematic commercial in wide/full-body framing — no macro shots, since fine print detail is where video models drift most. Detail macros live in the product sheet image instead. Video generation is the slowest step — it can take a few minutes.
Starting video generation…
Rendering video…
0:00
Single shot holds one continuous take per clip — no reframing, so the garment's print has the best chance of staying consistent. Best for detailed or busy fabric prints. Multiple shots gives more visual variety (a wide hero, a turn, etc.) and works well for simpler or plain fabrics, where print drift is less noticeable.
A talking presenter for non-wearable products — restaurants, salons, shops, and similar businesses. Runs on the same credits as the garment side.
Loading…
Click to choose the product photo
Click to choose the background image
Upload anything your action script needs already present from the first frame — a glass, a table setting, a second item. Without this, the model can only invent it mid-clip, which tends to make it appear suddenly partway through instead of being there from the start.
Before each later part, the last picture of the previous part is redrawn against the original to restore the face and sharpness. It costs 2 extra credits per later part and can sometimes change small things (a held item, hands). The redrawn picture is checked against the real last picture; if it changed the presenter’s size or position it is thrown away and the part carries on from the real last picture. For the most consistent face and body size across parts, leave this unticked and tick “start every part from the original picture” below. Left unticked (the default), each part simply carries on from the previous part's last picture, as it is.
Only matters for videos longer than 8 seconds. By default each later part carries on from the previous part's last picture, where a small held label has gone soft, and the video model may then invent a different label (for example just initials). Ticked, every part starts from the same sharp original picture, so the label looks the same in every part. The cost is that the presenter's pose resets slightly at each cut (the crossfade hides most of it). It costs no extra credits. Don't use it together with a walking or moving action. For a perfectly exact label, also add your real product photo as a cutaway photo in the graphics settings.
YouTube and Meta share the same 16:9 generation, cropped locally for Meta — selecting both together costs one generation, not two. Social (9:16) is a genuinely different native shape and always needs its own separate generation.
Sending it on WhatsApp? In real tests WhatsApp shrank a square (1:1) video to only 368 × 368 pixels, but kept 16:9 and 9:16 videos at about 480 pixels on the short side, which is roughly three times more picture. For the clearest result on WhatsApp, choose 9:16 (best on phones) or 16:9. Use the square only for Meta ads.
Tip: for body movement, describe it concretely, for example "walks two slow steps to her left, then back to the centre" or "turns slowly to show the back of the jacket". A short "walk little" is easy for the AI to ignore.
Words, logos, light effects and photos over the video. Tick this to open the settings. Left unticked, your video is made without any graphics.
Six free slots: Top, Bottom, Left 1, Left 2, Right 1, Right 2. Tick a slot and type any words (a price, an offer, a phone number, a heading — whatever you need), or switch it to Photo / logo to put your logo, a watermark or any picture there. They are drawn on top of the finished video, so the spelling is always exactly what you typed. For each slot you choose the second it appears and the second it disappears; after that it is gone from the screen (tick “stay till end” only if you want it to remain). Adding or changing graphics costs no credits.
Light effects (optional)
Glowing, see-through curved light lines that swirl around the presenter for a few seconds, like in launch ads. Tick one, choose the colours, the seconds it appears and disappears, and where the swirl is centred (Across 50 % / Down 68 % is around the waist of a standing presenter). The dashed orange oval in the small preview shows where it will swirl; the real glow appears only in the finished video.
Photos over the video (optional)
Show a picture (a product, a price list, a certificate, a map…) on top of the presenter while she keeps talking — the voice carries on. Six slots. For each: choose the photo, choose Full screen or Lower half, and the seconds it appears and disappears; after that the presenter is back. The dashed blue boxes in the small preview show where each photo will sit.
Photo size to upload (so nothing is cropped or blurry). The video is made in three shapes and each uses its own area:
YouTube 16:9
Social 9:16
Meta 1:1
Full screen
1920 × 1080 px
1080 × 1920 px
1080 × 1080 px
Lower half
1920 × 540 px (wide strip)
1080 × 960 px
1080 × 540 px
One photo for all three shapes? Use a square-ish picture of about 1080 × 1080 px with Show whole photo (the empty sides get a soft blur). JPG or PNG, any file size up to about 10 MB; very large photos are shrunk automatically. Each row below tells you how your photo will fit.
Preview as:
Every slot, light effect and photo is placed separately for each video shape you ticked above, so check each one.
Preview (uses your composed presenter picture once you have composed one).
Short clips (4, 6 or 8 seconds) with no presenter: a book opening, a bottle turning, a light being fixed, a screw rotating, students walking, a vehicle moving. Use them as cutaways between your talking parts, or on their own.
First picture of your clip
Look at the faces. If they match the photos, press the green button under the script to make the clip. If not, make another picture.
Your photo is the first picture of the clip and the exact look of the object. The script says what should happen to it.
Click to choose the photo of the object
One simple action works best in a short clip. Say what is moving and how (slowly, smoothly). If people appear, say adults, for example "college students" — the video service may refuse clips with children.
Without this, the video model picks a language from how the place looks (a Kerala street can come out in another language even when you typed English). Put the spoken words in quotes in your script, for example: Raju says "Please keep the change."
Give a person, an object or a place a name and describe exactly how it looks. After that, just use the name in any script. Every clip whose script mentions a name gets that description added for the video model, so "Raju" or "the tea shop" is described the same way in every clip. It is saved in this browser, so you can use the same names tomorrow.
Backgrounds: press “+ Add a background”, name it (for example “Raju's kitchen”), describe the place, then add a photo of it or press “Create background with AI” (an empty place made from your words, 2 credits). Write that name in any later script and the clip is set in the same place. A background photo counts towards the 3-photo limit, so one background plus two people fits in one clip. No permission tick is needed for a place.
Face lock: add a clear photo (a front-facing portrait for a person, an outside photo for a shop), or press “Create face with AI” to make a new face from the description (2 credits each time), so no real person is needed. The video model then copies that face or place from the photo, which holds the same person much better than words alone. Clips that use a photo are made by the Gemini Omni model and cost more credits. Use up to 3 photos in one clip. For 4 or 5 people (or people plus a background), keep “Make the first picture first” ticked: the picture can hold up to 5 photos, and the clip then starts from it.
Describe what can be seen: for a person the age, build, hair, clothes and colours, helmet or glasses; for a vehicle the type and colour; for a place the signboard words, colours, building and what is around it. Keep each description under 220 characters.
Longer clips cost more credits. Pick the length that lets your whole script finish.
Sending on WhatsApp? It shrinks a square (1:1) to 368 × 368 pixels but keeps 9:16 and 16:9 at about 480 pixels. Choose 9:16 or 16:9 for WhatsApp.
Type what the voice should say. It is read aloud and put on the clip. About 2 words fit in each second.
Voice over
Paste a script and get it read aloud as an audio file (MP3 and WAV), with no video. Works for long scripts of several minutes and for Indian languages. Written in Hindi, Kannada, Tamil, Telugu, Malayalam, Marathi, Bengali, Gujarati or English. For an English script you can pick "English (Indian accent)".
0 characters
The same voice reads the whole script. The voice names are Google's.
Dub video
Upload a video you shot. You get the same video with the speech in another language: a new AI voice, and the lips changed to match it. Up to 5 minutes.
Click here to choose a video, or drop it here
MP4 or MOV, up to 5 minutes and 300 MB. For the best result, one person talking, face clearly visible. A 720p export keeps the file small.
The tool works out who speaks when, and whether each speaker sounds like a woman or a man. Each speaker gets a different voice. The original person's own voice is not copied.
What to expect: the original background sound is replaced by the new voice only (add your music again in your editor). It works best when each shot shows only the person who is talking. Two people talking at the same time, or several faces in one shot, may not come out right.
Enhance song / audio
Upload a song or an audio clip. Tick what you want done and you get a ready MP3. You only pay for the steps you tick. Up to 8 minutes.
Click here to choose a song, or drop it here
Up to 8 minutes and 120 MB.
Makes the singer clearer and fuller: less muddy, more presence, smoother loud and soft words. Use it when the voice sounds dull.
Turn the background music down (or up) without touching the voice.
Much quieterSame as nowLouder
Takes out hiss and steady background noise.
Gently pulls notes that are a little sharp or flat onto the right note. It stays the same voice and the same timing. Slides and ornaments (meend, gamaka) and natural vibrato are left alone. It cannot add notes, power or emotion the singer did not sing, and it does not fix timing. Works best when one clear voice is singing.
Start with Light. Listen to the preview, then press “Mix again” to try another strength. It costs nothing extra to try.
Your client likes how another song sounds? Choose that song here. The finished MP3 is shaped to be as loud, as full and as wide as the reference (its tone, loudness and stereo width are measured; nothing is copied from it). Combine it with BGM adjust to keep the music a little lower.
Click here to choose the reference song, or drop it here
The reference song (MP3, WAV, M4A or a WhatsApp audio). It is only measured and is not kept.
Clone the voice
When the recording is noisy or the voice breaks up, the singer is re-sung in a clean voice. By default we pick a clean 10 seconds of the singer from the same song as the voice sample, so it stays the same singer. The whole voice is then converted to that voice, keeping the words, melody and timing; the music is kept as it is. If your client has sent a voice sample of their own (about 10 seconds), choose “upload” below: the song is then sung in their voice instead.
Voice sample
Click here to choose the voice sample, or drop it here
About 10 seconds of one voice, recorded in a quiet place with no music behind it (MP3, WAV, M4A or a WhatsApp voice message). Calm, steady singing with a few long notes can sound more natural in a song than speaking (speaking also works). A voice in a similar range to the singer's (a man's voice for a man's song) usually works best. A longer recording is fine too (up to 5 minutes): its clearest 10 seconds are used, and if some of it is spoken or sung higher, that part is kept as well and used for the high parts of the song.
The sample is sent to the voice service to make the cloned voice. It stays with this song on our server and is removed with it (within about 1½ hours).
Singer's own voiceOnly the cloned voice
The voice converter often sings some notes wrong. The tool moves the cloned voice back onto the original singer's notes. Parts it cannot match (or where the notes are too far off) keep the singer's own voice. If too little of the song was cloned, try Relaxed and press “Mix again” (quick, nothing extra to pay), then listen for notes that sound wrong.
A voice converter often cannot sing the notes that are very high or very low for the voice sample, so those parts keep the singer's own voice. With this ticked, such parts are converted one more time: for high notes with the singing lowered a few semitones (and, if that is still not enough, with the voice sample raised a little); for low notes with the voice sample lowered (back to the recording itself when it had been raised) and, if that is still not enough, with the singing raised a little. The result is moved back onto the original notes. Short stretches that kept the singer's voice inside a part that was cloned (a dropped word, for example) are also converted again on their own, as short pieces of a few seconds. It only fills what kept the singer's voice, and it uses a few extra voice-service calls (no extra credits). It is not used with the Strict melody check.
“Same as the original singer” keeps the singer's notes in the singer's own octave, which is the usual way. The other choice is experimental: where the singer sings far higher than the voice sample (about 7 semitones or more), that stretch is converted and sung one octave lower (and a stretch far below the sample one octave higher): the same notes, so it still fits the music, but nearer the pitch of the sample. The voice service does not handle singing that has been moved an octave well: on the first real test (a singer about 8 semitones above the voice sample) the voice came out hollow, like a ghost, and the singer's own voice came back about three times as often. That is why it is off by default; listen to the preview before you choose it for a song. Under the result you are told which stretches were moved. Changing this converts the whole voice again (a few minutes).
Sometimes the cloned voice is missing for a moment in the middle of a cloned line (a short word or note the voice converter would not sing, even after the second tries), and the singer's own voice then pops up in the middle of the new voice. With this ticked: a loud moment like that (under a second, between two cloned stretches) is filled with the cloned voice. It has the singer's notes and the singer's loudness and the colour of the new voice, but not the words of that moment; that is better than a hole in the song. (If a loud moment cannot be filled, because the cloned voice next to it is too short, too quiet or on other notes, it keeps the singer's own voice but turned down by 7 dB, so that the original does not stand out; turned off, the whole song would drop there, because the voice is most of what is heard.) A quiet moment is left without the singer's voice (a very short pause that the echo of the cloned voice covers). And the very quiet tail of the singer's voice that is left for a moment right after the new voice stops (the fading end of a note, where the original singer seems to come back) is turned off. Breaths and longer stretches that kept the singer's voice are not touched. Under the result you are told how many moments were filled or turned down and how many tails were turned off, and where a loud moment could not be filled. Untick this to hear the singer there, and press “Mix again” (quick, nothing extra to pay).
A voice converter keeps its voice close to the pitch of its sample. When the singer sings several semitones higher than the sample (a man's speaking voice for a woman's song, for example), the converted notes drift away from the song's and many parts need repair. With this ticked, the sample is raised part of the way (2 to 4 semitones, never all the way) before it is sent, so the converted notes land closer to the original's while the voice stays close to the person who recorded it. It does nothing when the sample is already close to the song's pitch. The voice can sound a little different from the sample: if it no longer sounds like your client, untick this and press “Mix again”.
A song is rarely sung at one pitch: a verse can be close to your sample and the chorus 6 to 11 semitones above it. For the chorus the converter gives a voice that is far too low, and the program has to lift it a long way afterwards. A voice lifted that far loses its body and sounds thin, “in the nose”. With this ticked, every part gets a sample raised for its own pitch (up to the pitch of the lower notes of that part, at most 7 semitones). A part that is not sung much higher than the rest gets the same sample as before. Under the result you are told which parts got a higher sample. If a high part now sounds too unlike your sample, untick this to raise the sample once for the whole song, as before, and press “Mix again” (the voice is converted again).
The voice is converted in pieces, and every piece is checked and repaired on its own. In shorter pieces the new voice stays closer to the sample (in long pieces it can drift towards the singer's own voice) and one bad piece spoils less of the song. The price is the same.
A converted voice comes back dry and a little dull next to a produced vocal, and that can sound like another singer. With “Match”, the original singer's brightness (more at 1–5 kHz) and soft echo are measured on the singer's own voice in the same song and given to the cloned voice. The cloned voice keeps its own sound: the lift is limited (at most 10 dB with “Match”; “Strong” goes about 40 % further) and nothing is taken from the music. Too bright or too echoey? Choose Gentle and press “Mix again” (quick, nothing extra to pay).
A voice converter leaves a faint growl (“grrr”) on low notes and some noise between the tones of the voice, which shows most in headphones and can sound slightly robotic. This smooths the voice cycle by cycle, only on steady, clearly pitched notes and only below about 4 kHz: the notes, the words and the consonants are not changed, and the tone changes by about 1 dB at most. Still rough or robotic? Choose Medium or Strong; too smooth or too clean? Choose Light or Off and press “Mix again” (quick, nothing extra to pay).
The cloned voice is a dry voice in the exact middle, with nothing around it, and in headphones that sounds close, flat and slightly robotic next to a produced vocal. With this, the soft echo from “Brightness and echo” is spread over left and right, so that it is around your head and not inside it. What you hear in mono does not change at all. It needs the echo (“Brightness and echo” not Off). Press “Mix again” to try it (quick, nothing extra to pay).
Please know: this is experimental. Voice converters are made for speech, so singing can come out with a slightly artificial or different tone. The converter can sing wrong notes, so they are repaired to follow the original melody. Every part is checked against the original timing and melody; a part that cannot be matched keeps the singer's own voice, and you are told which. Listen to the free preview first, then press “Mix again” to try another sample or a lower amount.
Use this when the client says the result is quieter than his song. When “Match a reference song” is ticked, the loudness follows the reference instead.
What to expect: the voice is separated from the music by AI, which is very good but not perfect. You may hear a faint "watery" sound on the voice, or a little echo or backing vocal left in the music. A poorly recorded voice becomes better, not studio quality. The finished MP3 is set to a standard loudness. You can listen to a free 30-second preview before you pay.
Create song
Record your singing on a phone or recorder, with no music. We listen to it, work out the key and the speed, add music that goes with it, and then make the voice clearer and cleaner. You get a ready MP3. From 8 seconds up to 5 minutes.
Click here to choose your recording, or drop it here
Sing alone, close to the phone, in a quiet room. 8 seconds to 5 minutes.
1. The music
Leave Key and Feel on “Find it” first. After the music is made we show what we found; if the music sounds wrong against your voice, choose the right key and press “Mix again”.
Very softSame as the voiceLouder
2. Your voice (all optional)
Clearer and fuller: less muddy, more presence, even loudness. No extra charge.
Takes out hiss, fan and room noise from the phone recording.
Gently pulls notes that are a little sharp or flat onto the right note. Same voice, same timing; slides, ornaments and natural vibrato are left alone. It does not add notes or fix timing.
What to expect: the music is made by software, so it is a simple, tidy accompaniment, not a studio recording. It works best with one clear voice and no other sound. If you do not keep a steady beat, there are no drums. The key is found automatically and can be wrong for unusual tunes; you can choose it yourself. You can listen to a free 30-second preview before you pay.
Write a song
Tell us what the song is about. AI writes the lyrics, you choose the music, and AI makes the song (4 minutes at the most). Then you send a voice sample of about 10 seconds, and the song is sung in your voice. You can listen to free 30-second previews before you pay.
1. What is the song about?
The more you tell us, the better the lyrics.
Choose the kind of voice that will sing it: the same as the voice sample you will send. It decides how high the tune goes.
Keep the headings in [square brackets], like [Verse 1] and [Chorus], on their own lines. Short lines (6 to 10 syllables) are easier to sing and to clone. Please do not copy the words of an existing song.
3. The music
Melody
Rhythm
Support
SlowMedium
A slower song is easier to sing and to clone. The length of the song depends on the speed.
4. Make the song
What to expect: an AI music service writes the tune and records the song, with a singer and the instruments you chose. It takes a few minutes. Then you listen to a free 30-second preview, send your voice sample, and the singer is replaced by your voice. This is new and experimental: the cloned voice can sound a little different from your own, and a part that is too high or too low for your voice may keep the AI singer's voice. You can try another voice sample before you download.
A free 30-second piece from the middle of the song. In the next step the AI singer is replaced by your voice, and the music stays.
Your song, sung in your voice
A free 30-second preview from the middle of the song. The MP3 you download is the whole song.
Your voice sample
Send about 10 seconds of the voice that should sing this song. The whole song is then sung in that voice and the music is kept. The song is moved up or down by a few notes to suit your voice.
Click here to choose your voice sample, or drop it here
The sample is sent to the voice service that makes the cloned voice. It stays with this song on our server and is removed with it (within about 1½ hours).
More settings for the cloned voice (optional)
How much of your voice
The AI singerOnly your voice
Melody check
Brightness and echo
Smooth the voice
Width of the echo (headphones)
Clean video voice
Upload a recording with a noisy voice: a screen recording, a talk, a phone or camera video. You get back the same video, the same length, with the noise taken out of the voice. The picture is not changed or re-compressed. Up to 30 minutes and 600 MB.
Click here to choose a file, or drop it here
MP4, MOV, MKV or an audio file. Up to 30 minutes and 600 MB. A big file takes a few minutes to upload; sending only the sound (MP3 or WAV) is much faster, but then you get the clean sound back instead of a video.
What to expect: the voice is cleaned in pieces of about 5 minutes (cut in the quiet gaps between words), then joined again, so you still get one file with the whole recording. Very loud noise can leave the voice sounding slightly processed. You can listen to a free before and after of 30 seconds first. Long recordings take several minutes to finish; you can leave this page open while it works.
Export for marketplaces & social
Ready-to-upload crops for Amazon, Flipkart, Myntra, Meesho, Meta Ads, and YouTube — generated from what you already made, no extra credits.