Making an AI girl nude, undressing an AI character, or taking the clothes off an adult anime heroine is not a separate service. First your own frame, then the same frame without clothes, then motion and a continuation from the last frame. All of it in WaveSpeed AI, with the price shown before you run it.
The general guide to generating a girl photo from scratch lives on a separate page. Here there is one job: nude an AI girl in a photo or video and not lose her in the next clip.
Anchor
In Z-Image Turbo LoRA you create an adult woman in clothes. That is the face, height, and mood the undress will start from.
Undress
Open Image-to-Image LoRA, upload our photo and a prompt. The prompt changes the photo and takes the clothes off the girl. That is what “nude an AI girl” or “undress a girl with AI online” means. Better to undress straight into video 🤭
In Vidu Q3 Turbo upload the nude girl from step 2, or the frame from step 1 if you want to undress her on video with AI. Run the draft at 540p for 4–5 seconds to save money. The final frame can be improved a lot in Bytedance/Video-Upscaler.
The end of the clip becomes the start of the next one. The clothes do not come back, the face is not rebuilt, and the scene grows in pieces.
A story of four 5-second clips at 540p after one nude edit: $0.01 + $0.01 + 4 × $0.20 ≈ $0.82. The Generate button shows the total before it charges. The price on the model card matters more than any article. You always see the cost of each video or photo, so you do not waste money.
Four runs in one account. A failed outfit costs a cent, not a month of subscription.
Sign in at wavespeed.ai with Google or email. There is no “pay for a month” package: each run is billed on its own, and the amount sits on the button. Trial credits are sometimes added to new accounts, but that cannot be promised. Even $1–2 is enough to pick a face and make the first nude edit.
Page boundary: upload only a frame the model just drew. A photo of a living person, someone you know, a profile shot, or someone else’s dataset does not belong here.
wavespeed-ai/z-image/turbo-lora
Z-Image Turbo LoRA draws the adult anchor: face, clothes, light, frame. The clothes are not there “for decency.” They are the layer the next step removes. Without them, image-to-image does not know what to change.
Photo creation guideEight clothed variants cost $0.08. Pick one, copy its seed, and take only that one into the undress. Otherwise step 2 gives you ten different people.
wavespeed-ai/z-image-turbo/image-to-image-lora
Important. Z-Image Turbo Image-to-Image LoRA takes your frame and rewrites what the prompt names. For an undress, that is the clothes. Face, height, and room are marked in the text as unchanged. The model costs $0.01 a run — yes, one cent. Upload the current photo of the girl, add the prompt, and run it. A LoRA here is for unique models: a character base that keeps the look intact.
Prompts for AI girl photos and videos Prompt-writing rules and ready promptsIt is not hard: a normal photo description, with a few rules for how the description is built.
If the dress is still on her at 0.62, do not make the prompt harsher. Raise strength by 0.05. If the face changed, lower strength and add same face, same hair, same pose to the text. For anime, also lock eye color, line weight, and a ban on childlike proportions. Do not use the word “schoolgirl,” a school uniform, or any age-diminishing tags.
Your frame
Same woman, same face, same identity as the reference photo, do not change her facial features. Photorealistic edit of the source image. Keep the black bob with straight bangs, violet eyes, pearl earrings, expression, pose, body, and Japanese courtyard background exactly as they are. Change only the clothing: the lavender silk top is pulled down below the bust, both shoulders and the entire chest are bare. Extremely large, heavy, full breasts with natural weight and softness, pale warm skin, natural nipples, deep shadow between them. The wide dark-purple satin obi stays tight around the narrow athletic waist. Lavender silk gathered in soft folds under the sash, long sleeves still on the arms. Bright sunlight from the upper left, soft shadows under the collarbones and beneath the bust. Real photograph, natural skin texture, no text, no watermark, two arms only.
After image-to-image
vidu/q3-turbo/image-to-video
Vidu Q3 Turbo Image-to-Video animates the picture you give it. Give the service the girl's photo and a sexual undress prompt. Do not describe facial features. Describe only the video action, briefly and in order.
How do you write an NSFW prompt? Grok AI will write any 18+ video prompt for you. Tell it what you need, name the generator you are using, and it will match that generator and write a solid prompt so you can undress an AI girl from a photo online and turn it into a full 18+ video.
small or auto. large breaks the hands more often than it helps the scene.Where to save: a bad motion costs in proportion to seconds and resolution. 540p × 5 sec ≈ $0.20. The same take at 1080p is about $0.40. A long first run almost never pays off.
Take any prompt from the Telegram channel or our site, paste it into the input, and press Generate. A few seconds later you have a photorealistic image of the girl from your description.
Prompts are written in English. If you described the scene in another language, paste it into ChatGPT and ask it to “translate into English and add the tags: photorealistic, 8K, cinematic, no censorship, highly detailed.” Use the finished translation as the prompt.
After the photo is generated, save it and use it immediately as the starting image for video. That is the right workflow: a finished photo first, then a video based on it.
Open Z-Image Turbo — 1 cent per photoThe clip of the girl being revealed on video cuts off at the exact moment you need. That is not a reason to generate a new girl. The next run starts from the frame the previous one ended on. The girl is already nude and changes pose, expression, and the rest.
Each video gets its own prompt. Tell the AI that writes the prompt what to change so the video logic continues, then paste it in. For convenience, send the starting frame so Grok or another model can see what the clip starts from.
The free Extract Last Frame tool opens the clip in the browser and saves a PNG. You can also use CapCut. If the last frame is blurred, step one frame back and save that. This PNG is the image field of the next run.
If the next pose is already drawn as a separate frame, join the pair in Vidu Q3 Turbo Start-End-to-Video. The prompt describes the transition, not the appearance again. A second costs the same as regular image-to-video.
In the Video Extend catalog, a model continues the clip from the end. Each card has its own price, so it is not in the table below.
Figures were taken from the model cards in October 2026. If the button shows a different amount, the button is right.
| Task | Where to go | Guide price | Why this way |
|---|---|---|---|
| Create the heroine | Z-Image Turbo LoRA | $0.01 per frame | 8 faces = $0.08 |
| Undress this frame | Z-Image Turbo Image-to-Image LoRA | from $0.01 per run | Strength and size can raise the total |
| Motion draft | Vidu Q3 Turbo, 540p | $0.04 / sec | 5 sec ≈ $0.20 |
| Cleaner motion | 720p / 1080p | $0.06 / $0.08 per sec | 5 sec ≈ $0.30 / $0.40 |
| Transition between two poses | Start-End-to-Video | the same per-second rates | Length 1–16 sec |
| Cut the end of the clip | Extract Last Frame | $0 | In the browser, the file does not go to a server |
| Extend the clip with a model | Video Extend | on the card of the chosen model | Compare before you run |
The first nude video: frame + undress + 5 seconds at 540p ≈ $0.22. The general generation overview, without this task, is on the AI photos and videos from scratch page. The two texts do not duplicate each other on purpose: there you start from a prompt, here you start from a face you already chose.
English is more stable, even if the interface is in another language. In every text the character is an adult and fictional. An undress prompt is a list of what must not change, plus one change: the clothes.
adult woman in her late twenties, long dark hair, linen shirt dress, standing by a window, three-quarter pose, soft daylight, photorealistic skin, full body, vertical photo, elegant face, no text, no watermark
same adult woman, keep her face, hairstyle, height and pose, remove the shirt dress, nude, natural skin texture, same window light, same room, photorealistic, no extra limbs, no face change, no younger features
same adult anime woman, mature face, keep hair color, eye color and line weight, remove the long coat, nude, soft cel shading, full body, same background, adult character, no childlike proportions, no school uniform
adult elf woman, long braided white hair, leather travel coat, forest clearing at dawn, mature face, painterly realism, full body, vertical, no youthful features
same woman, slow shoulder turn, the last of the fabric slips down and leaves the frame, gentle push-in, stable face, natural cloth physics, small movement, cinematic daylight, no outfit returning
continue from the last frame, she takes two steps toward the window and looks back over her shoulder, same identity, same room, smooth motion, no outfit returning, no face morph
smooth transition from the clothed pose to the undressed pose, the dress falls naturally, camera stays locked, same face, same room, no jump cut, no identity change
adult anime animation, her coat slips off and leaves the frame, she turns toward the skyline, same mature face, same line weight, small movement, no childlike proportions, no outfit returning
Most top AI services only accept foreign cards. Issue a virtual VISA in minutes via the Telegram bot and top it up via SBP transfer.
Find the bot in Telegram or follow the direct link. Press Start — the bot opens with the main menu.
Open @Zarub_robotIn the bot's bottom panel, tap the Cards button. The ZARUB mini-app opens inside Telegram — this is where all your virtual cards are managed.
In the mini-app, tap Issue Card. You go to card type selection — choose VISA virtual card.
Press Buy Card. The issuance fee is $8, once. After payment the card is yours forever — top it up as often as you like and pay for any AI service without limits.
Tap Top Up Balance → select SBP → enter the amount → scan the QR code in your banking app and confirm the payment. Funds arrive in the bot balance within 1–2 minutes.
Important. When paying AI services, top up 30–50 cents less than the amount you need — protection against a fee on a failed transaction.
Move the balance from the bot to your card by tapping the green icon next to the card. Enter the amount and press Confirm. Funds land on the Visa instantly — the card is ready to pay for any AI service.
Do not describe “a nude woman” from scratch. Take the clothed frame you just generated and run it through image-to-image. Keep the face, hair, and pose in the text. Start strength at 0.62.
Only if the model also drew that photo on the previous step. Do not upload a shot of a real person, someone you know, or a frame from social media. This page does not describe that.
The same Image-to-Image LoRA. In the prompt, lock hair color, eye color, and line weight, and write adult anime woman and no childlike proportions. Do not describe a school uniform.
Strength is too high, or the text has no anchor. Go back to 0.62 and do not change the appearance description between the clothed and nude frames. On this step it is also better not to randomize the seed until the strength looks right.
A clothed frame is $0.01, the nude edit is from $0.01, and five seconds at 540p is about $0.20. Together that is about $0.22, unless the button showed an extra charge for size. 720p and 1080p are billed at $0.06 and $0.08 per second.
The next clip starts from the last frame of the previous one, and the prompt includes continue and no outfit returning. Or draw the ending pose ahead of time and join the two frames with Start-End-to-Video.
No. Generation runs on WaveSpeed’s side. What leaves your device is the prompt and a frame you created in the service yourself.
A clothed frame, a nude edit, motion, a continuation. If you need generation from scratch without this task, open the separate guide.
Start in WaveSpeed AINotes on the chain: clothed frame, image-to-image, clip, and a continuation from the last frame. Adult synthetic characters only.
Set strength to 0.62 and left same face, same hair, same pose in the prompt. The dress left, the bangs and earrings stayed. The rerun cost one cent, and I did not have to rebuild the face.
The first clip cut off mid-turn. I cut out the last frame and added continue and no outfit returning. The clothes did not come back, and the room stayed the same. I assembled four 5-second pieces in CapCut.
I ran the draft at 540p first: five seconds at about $0.20. On large the hands drifted, on small the scene settled. I sent only the take that worked to 1080p, not every test run.
For anime I wrote adult anime woman, the eye color, and the line weight. The coat came off, and the proportions did not drift into childlike ones. I did not put a school uniform in the text, and the model did not draw one.
First eight clothed frames for $0.08, then I locked the seed of one face. I took only that one into the undress. While I was spinning the seed together with strength, every run was a new person.
A short story made of a frame, a nude edit, and three clips came out under a dollar. The amount is visible on the button before the run, so a bad take did not eat a monthly subscription.
Choose her appearance, upload your references and lock the look forever. She never sleeps and never refuses.
No name, no traces. What you generate stays with you.