Image to image
Uncensored AI image editor
An uncensored AI image editor reworks a picture you already have from a written instruction, adult material included, and does not quietly tone that instruction down before the model reads it. On Vellria that work is done by the editing models: attach your picture in Studio or send its address to the API, say what should change, and pay for each run from prepaid credits. Invented adults can be edited freely, while turning a photo of someone who exists into explicit content without that person's agreement is prohibited by the terms.
What an image to image edit does
Image to image means the model starts from pixels instead of a blank canvas. You hand over one picture or several, write what should be different, and a new image comes back built from that input. Typical requests are another outfit, a changed setting, clothing removed or added, a different light, or the same figure in another pose. Whatever you leave unmentioned is what you are asking the model to hold, so an instruction that names the change and the parts to keep works better than one that describes the whole scene again.
As an NSFW image editor it runs on the same terms as generation here. An adult instruction reaches the editing model in the words you chose; nothing swaps it for a milder version, and our automatic check is not looking for adult vocabulary. That is what uncensored AI image editing amounts to in practice: the edit that comes back answers the edit you asked for, or the run fails and your history shows it.
Models
Seedream 5 Lite Image Edit
Takes up to fourteen references and redraws the images you provide.
Image to image
5 credits
Wan 2.7 Image Edit
Edits from up to nine images, keeping the source proportions. Extra images are free.
Image to image
6 credits
Qwen 3 Image Edit
Edits using the images you provide as references, preserving character and scene.
Image to image
5 credits
Seedream 5 Pro Image Edit
The flagship editing model. Takes up to ten references and preserves identity and lighting.
Image to image
8 creditsWhat the catalog tells you before an edit runs
Every editing model needs at least one picture and turns a job down when none is attached. How many it will read differs between models; each model's page prints that range from the live catalog, and every card on this page is tagged image to image. A model that takes several lets you split the evidence, with one picture carrying the person and another carrying the room or the garment.
Beyond the first picture, some models add a small charge for each extra reference and others add nothing; the total shows beside the run button before you commit. Output size belongs to the model as well. Editing models usually offer a list of shapes and sizes, and where a model keeps the proportions of the picture you sent, its page shows that in place of a list, so crop before uploading when the result should be framed differently.
Some editing models also accept a negative prompt, a seed you can fix between runs, or a reasoning switch, and each model page lists which. The fixed seed earns its place in editing: hold it along with the picture, change one phrase, and more of the difference between two outputs comes from that phrase.
Editing a picture in Studio or over the API
In Studio, choose an editing model and add your picture in the reference box, either by uploading the file or by pasting an https address; the panel shows how many pictures that model expects before anything runs. Write the change, pick a size where the model offers one, and start. The job carries on in the background and the result lands in your Gallery, so leaving the page does not lose the edit.
From code, upload each file to /v1/uploads as a raw body whose Content-Type matches it, hold on to the returned address, and list those addresses in image_urls beside the model id and your instruction. The call returns a generation id at once; poll it until the run reaches a final state. Files can be JPEG, PNG, WebP, BMP or GIF, uploading requires a positive credit balance, and the API reference documents the whole body.
A picture made here can go into an edit too, just not by its link, because a finished file is served behind your sign-in and the provider rendering the edit cannot fetch it from there. Over the API, `POST /v1/generations/:id/reference` copies a finished image into an upload and returns an address the provider can read; in Studio, download the picture and upload it as the reference. For the starting picture itself, use the uncensored AI image generator. When one character has to survive many edits, keeping a character consistent sets out the method, and illustration or original-character work, where an edit is one pass in a longer loop, has its own page, the AI art workflow for figures and characters.
What an edit costs
An edit is priced by the model and the output size you pick, plus any per-reference charge that model lists, and the credits leave your balance when the run starts. There is no subscription and no free tier: balances are funded by card or another method available at checkout, and one left idle costs nothing. If the provider fails the job, the credits return automatically once it is marked failed, and pricing has the top-up packages.
Real people, and what the automatic check cannot see
An editor raises a question that writing a prompt from nothing does not, since a file you upload might picture anybody at all. Editing a photo of a real, recognisable person into sexual or explicit content without their consent is forbidden by the terms, as is any other recognisable depiction of a real person made without consent, and being famous exempts no one. Breaking that rule costs you the account: the terms close it without notice.
Nothing automated enforces that rule for you. Our check reads the text of the prompt, looks for minors and refuses a prompt it catches before any credit is taken; it does not open the uploaded file or attempt to identify the face in it, which leaves keeping real people out to the terms and to whoever uploads. The model's provider may also run its own filter, and when that stops a run the credits are refunded. The rules for NSFW generation set out the whole policy, age requirement and all.
Frequently asked questions
Is there an uncensored image to image AI?
Yes. The editing models in this catalog are image to image models, and an adult instruction reaches them as written. Each takes at least one picture, reads up to the number its model page shows, and returns a new image built from it. Sexualising a minor is prohibited, and an instruction the automatic check recognizes as doing so is refused; editing a real person without consent is prohibited by the terms.
Can I edit a photo of a real person?
Not into anything they have not agreed to. The terms forbid any recognisable depiction of a real person made without that person's consent, famous people included, and an edit counts as much as a fresh render. Nothing automatic inspects the uploaded picture for who is in it, so that responsibility sits with the account doing the uploading.
What happens to the pictures I upload?
Every upload gets its own address built around an id nobody can guess, and the model's provider reads it from there without signing in, which also means anyone who has the address can open it. An upload is kept exactly as long as a generated file would be; after that the file is deleted and the address reports it as expired, though a cached copy can linger for up to a day.
Is the editor free to use?
No. Vellria has neither a free tier nor trial credits, so each edit comes out of a balance you fund by card or another method available at checkout, and no subscription sits behind it. An edit that fails at the provider gives its credits back automatically.