
Labels and process
Undress AI, clothes remover, DeepNude: one process, four labels
New labels appear when the old name gets harder to search for. That rarely means the process underneath is new.
Short answer
Undress AI, clothes remover, nudifier and DeepNude are four labels for one process: a region of a photograph is masked, its contents are discarded, and the gap is redrawn by a generative model. In Philippine search results the newer labels often sit above the older name, and that is a marketing fact rather than a technical one — new wording appears when the old word becomes harder to advertise, not when the engine changes.
What actually differs between services is not the label on the button but what each one publishes about price, storage and ownership. The inventory of what is used here now is on the front page.
Four names, one operation
What each label promises and what it delivers
Read across the table and the promises differ while the operation does not. The third column is the one worth checking: it is the only part that can be verified from outside.
| Label in the advertising | What the wording implies | What the process actually is |
|---|---|---|
| Undress AI | That software performs an action on clothing, as if the garment were a separate layer in the file. | The garment area is masked and deleted, then refilled from noise. Nothing is lifted, because nothing underneath was ever recorded. |
| Clothes remover | A tool that takes something away, like an eraser. | Nothing is removed and kept aside. Pixels are replaced with invented ones that match the surroundings statistically. |
| Nudifier | A transformation of the same person into another state. | The body in the output belongs to no one: it is an average of training data, and it changes on every run. |
| DeepNude | A specific product with a history, which suggests there is an authentic version of it. | A search term. The program it named was withdrawn by its author in 2019 and has no successor; the services using it now are unrelated. |
| Photo editor or filter | An ordinary retouching step, the same family as brightness or a colour filter. | A generative model rewriting part of the frame. A filter changes existing pixels; this discards them and fabricates new ones. |
Every row in the middle column shares one implication — that information about the body is somehow present in the source file and is being brought out. It is not. A photograph records light reflected off the outermost surface, and there is nothing behind fabric for a model to recover. The step-by-step mechanics are in the piece on what differs between these engines.
There is one genuine distinction hiding among the labels, and it is not between any two of them: it is between editing an existing photograph of a person and generating an image from text. That difference decides whose data moves and who can complain, and it gets its own page on two different processes.
Why the wording moves
New labels appear for reasons that have nothing to do with the engine
Four forces push the vocabulary along, and none of them is technical progress. All four are visible from outside.
Platform rules name words
Store and advertising policies are written around terms. Google Play's content policy names non-consensual sexual content created via deepfake or similar technology, so a service describing itself differently buys itself a little time.
A dead product name carries no owner
The 2019 name has no proprietor and no trademark defence, so anyone may put it in a page title. That makes it cheap traffic — and also makes it worthless as a signal of who is behind the site.
Newer wording sounds like a tool
"Editor", "filter" and "remover" borrow the vocabulary of ordinary photo software. The borrowing is the point: it makes an operation on somebody else's image sound like a routine adjustment.
Affiliate pages multiply the terms
Pages paid per registration have an incentive to cover every phrase people might type. Most of the label variety in search results comes from them, not from the services themselves.
The scale behind the vocabulary churn has been counted once. Graphika's report on synthetic intimate-image services recorded 34 provider websites with more than 24 million unique visitors in a single month and referral link spam for such services rising by more than 2,000 per cent on Reddit and X over . A category that large generates new wording constantly, because the wording is the cheapest part of the product to change.
Six checks that ignore the label
What to compare when the names keep moving
Every one of these is readable from outside, and none of them depends on what the service calls itself this month.
A price that exists before sign-up
Three of the six put a complete tariff in front of a visitor with no account, and each of the three counts in a unit of its own: credits at Ainudez, gems at Undresswith, dollars per execution at WaveSpeed. None of those units converts into another without that service's own rate, so two open tariffs still do not line up at a glance. The other three publish no figure until an account exists.
The check that survives any rename is therefore not reading a price but converting one: take the smallest pack, divide it by what one image costs in that service's unit, and the wording on the button stops mattering. Where the figure lives only inside an account that division cannot be done at all, and the moment it surfaces is chosen — after registration, after a friend has been invited for coins, after a job has been left half-run. The row-by-row layout of who charges what is on its own page.
A company with an address
Count how many of the six write down a company and the answer is two. The other four offer a country, a team page, a mailbox or nothing at all — material that reads like an identity and cannot be looked up in any register. That split has not moved while the wording on the buttons has turned over three times, which is the whole point of putting it on this page.
This is the row that survives every rebranding. A service can change its label overnight; it cannot invent a company registration overnight, and it cannot hide the absence of one behind a new word.
A deletion period in days
Only one of the six writes a single number: Undresswith states that files are held encrypted for at most 24 hours and then deleted, while its account, payment and IP records are stated as kept for seven years. Deep Undress writes a promise and a deadline into the same set of documents: nothing is stored, says the upload box; 72 hours, says a content notice; three days, says the task history for that same 72 hours. Ainudez and WaveSpeed use "as long as reasonably necessary". UndressHer writes no period anywhere, and Pornworks publishes nothing readable.
The pattern is not about generosity but about coverage: a short window always attaches to the image, while the traces that actually identify you carry the longest periods. Where exactly the file stops being yours, and the point where the file leaves your device, is traced step by step on another page.
A payment that can be disputed
Ainudez and Undresswith take crypto only, and the latter's documents say refunds in crypto are generally not available. WaveSpeed takes cards, PayPal and a bank transfer. UndressHer takes cards and crypto and is the only one of the six with a written refund window: 14 calendar days, prorated by usage.
For anyone paying from the Philippines there is a second point on this row. None of the six accepts GCash, Maya or an over-the-counter deposit, so a buyer without a card is pushed onto the one rail that has no reversal mechanism at all. That is a bigger difference than a few cents per image, and it is the same regardless of what the service calls itself.
Rules that match the interface
Read the content rules against what the interface actually invites you to do. Undresswith describes automatic filters, an age estimation model and human review within 24 hours, and requires written consent from anyone appearing in the material. Deep Undress makes you tick a box acknowledging that processing somebody else's image without permission may be illegal.
Then read the contradictions. The WaveSpeed terms forbid non-consensual nudity and elsewhere state that the company does not monitor its models' outputs. The UndressHer front page ends its three-step flow with saving the image to your device, while its terms forbid downloading or distributing results and permit use only inside the platform. When two documents disagree, the binding one is the one that counts in a dispute. What comes out of all this, and the output nobody describes in writing, is examined separately.
The label does not change the law
Philippine law weighs the act towards a person, not the vocabulary on a button. Section 12 of the Safe Spaces Act of 2019 covers uploading and sharing, without the victim's consent, of media containing photos or video with sexual content, with a penalty in Section 14 of prisión correccional in its medium period or a fine of ₱100,000 to ₱500,000, or both. The Anti-Photo and Video Voyeurism Act of 2009 punishes copying, distributing and publishing intimate images taken without consent with three to seven years and the same fine range, and Section 6 of the Cybercrime Prevention Act of 2012 raises any such penalty by one degree when ICT is used.
Calling the operation an edit or a filter changes none of that, and neither does the fact that the pixels were fabricated: what the law protects is the person in the image, not the authenticity of the file. Where a child is depicted, the Anti-OSAEC and Anti-CSAEM Act of 2022 writes computer-generated images into its definition explicitly and sets life imprisonment and a fine of not less than ₱2,000,000 for production. The money side, and what to check before paying, is collected on its own page.
Common questions
About the labels and what is behind them
Answers built from the services' own documents, platform policies and the text of Philippine statutes.
Are undress AI and the older name the same thing?
As a process, yes: mask, delete, redraw. As a product, no — the 2019 program was desktop software that ran on the user's own machine, while everything under the newer labels is a web service computing on somebody else's servers. The shared word is the operation; the difference is where your file ends up.
Why do the newer labels rank higher in search here?
Because they are the words the services and their affiliates currently advertise with. Platform policies name specific terms, so the marketing moves to wording that has not been named yet. That movement says nothing about which engine is running underneath.
Is a clothes remover the same as a photo filter?
No. A filter changes pixels that exist. This deletes a region and fabricates new pixels that never existed in the source file, which is why two runs on the same photo produce two different bodies.
Does a different label mean a different level of quality?
Not in any way you can verify. None of the six services publishes an output resolution for a still image or a paired before-and-after example, so quality cannot be compared before a transaction whatever the service calls itself.
Does calling it an edit change the legal position?
No. The Safe Spaces Act and the Anti-Photo and Video Voyeurism Act are written around what was done to a person and whether they consented, not around the name of the tool. Where a child is depicted, the Anti-OSAEC Act covers computer-generated material explicitly.
Is there any service that advertises no restrictions at all?
None of the six. Four describe filters or express prohibitions, one publishes a takedown procedure with deadlines, and the sixth publishes nothing readable. The promise of no limits appears on third-party pages paid per registration, not in any service's terms.
Does this desk operate any of these tools?
No. Nothing here can be run, no photos are accepted, and no images are produced. This is an editorial examination of what third-party services publish about themselves.