This is the per-image price list for every uncensored AI image generator this key reaches: 10 image models and 16 tiers, pulled from the live catalog on 2026-11-16, with no stock prompt filter in front of any of them. To try a model before you script it, the browser version runs the same catalog. Below: the table, a model-by-model guide, the bill at 100 and 1,000 images, where other APIs do better, then the request itself.
The per-image table
Each row below is one price tier of one uncensored image model, taken from the live catalog on 2026-11-16 and ordered from the lowest per-image price upward. A tier names the settings its price covers, and a request pays the first tier it matches. The 100-image column multiplies that price by 100 with nothing else added.
| Model | Tier | Per image | Per 100 | Images for $10 |
|---|---|---|---|---|
| CyberRealistic XL Uncensored | size 512x512 | $0.0136 | $1.36 | 735 |
| WAI Illustrious SDXL Uncensored | size 512x512 | $0.0136 | $1.36 | 735 |
| CyberRealistic XL Uncensored | any settings | $0.0204 | $2.04 | 490 |
| WAI Illustrious SDXL Uncensored | any settings | $0.0204 | $2.04 | 490 |
| CyberRealistic XL Uncensored | size 2048x2048 | $0.0272 | $2.72 | 367 |
| WAI Illustrious SDXL Uncensored | size 2048x2048 | $0.0272 | $2.72 | 367 |
| Anime Image Uncensored | resolution 1K | $0.03 | $3.00 | 333 |
| Seedream 5.0 Lite Edit Uncensored | any settings | $0.0313 | $3.13 | 319 |
| Qwen Image 2.1 Uncensored | resolution 1K | $0.0375 | $3.75 | 266 |
| Vision Image Uncensored | any settings | $0.04 | $4.00 | 250 |
| Anime Image Uncensored | resolution 2K | $0.045 | $4.50 | 222 |
| Chroma Uncensored | any settings | $0.08 | $8.00 | 125 |
| Qwen Edit Uncensored | any settings | $0.08 | $8.00 | 125 |
| Qwen Image 3 Uncensored | any settings | $0.08 | $8.00 | 125 |
| Vision Pro Image Uncensored | resolution 2K | $0.09 | $9.00 | 111 |
| Qwen Image 2.1 Uncensored | resolution 2K | $0.20 | $20.00 | 50 |
10 image models and 16 base tiers; on 6 of the 10, enable_safety_checker is listed with a default of false. The image and video models page, every model page and the console all read this one catalog, so the figure here is the figure a request is billed.
The unit is one output image. Send n: 4 to a model that supports n and the charge is four tier prices. On models with more than one tier, the resolution decides which tier applies; the aspect ratio never does.
Choosing an uncensored AI image generator by the job
Start from what you have. If the job begins with an image of yours, use an edit model (Qwen Edit Uncensored, Seedream 5.0 Lite Edit Uncensored). If it begins with a prompt, use a text-to-image model, and pick between those by the settings you need, because the base tiers only span $0.0136 to $0.20 per image.
Anime Image Uncensored
Anime and illustration from a text prompt. Negative prompt and seed are ordinary settings here, which is what keeps one character the same across a whole sheet of frames. Starts at $0.03 per image on the resolution 1K tier. Settings: size 1:1, 3:2, 2:3, 4:3, 3:4, 16:9, 9:16, 21:9; resolution 1K or 2K; output png, jpeg; negative prompt; seed. Id anime-image-uncensored.
Chroma Uncensored
General text-to-image at a single price for every setting: a 2K image is billed the same as a 1K one. Starts at $0.08 per image. Settings: size 1:1, 4:3, 3:4, 16:9, 9:16, 2:3, 3:2, 21:9; resolution 1K or 1.5K or 2K; output jpeg, png. Id chroma-uncensored.
CyberRealistic XL Uncensored
CyberRealistic XL Uncensored is one of the uncensored image models on this key. Starts at $0.0136 per image on the size 512x512 tier. Settings: size 1024x1024, 1024x768, 768x1024, 1408x1024, 1024x1408, 1920x1088, 1088x1920, 512x512, 2048x2048; up to 4 images per request. Id cyberrealistic-xl-uncensored.
Qwen Edit Uncensored
Qwen Edit Uncensored is one of the uncensored image models on this key. Starts at $0.08 per image. Settings: up to 4 images per request; a source image is required. Id qwen-edit-uncensored.
Qwen Image 2.1 Uncensored
Generation and editing in a single model. Pass image_urls to edit, mask_url to confine the edit to one area, and background transparent for a cut-out. Starts at $0.0375 per image on the resolution 1K tier. Settings: size 1:1, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16, 21:9, 9:21, auto; resolution 1K or 2K; output png, webp, jpeg; seed; mask; an optional reference image. Id qwen-image-2.1-uncensored.
Qwen Image 3 Uncensored
Qwen Image 3 Uncensored is one of the uncensored image models on this key. Starts at $0.08 per image. Settings: resolution 1K or 2K; up to 4 images per request. Id qwen-image-3-uncensored.
Seedream 5.0 Lite Edit Uncensored
Edit only: each request needs a source image. One price whatever the size, and n accepts up to 15 variations in one call. Starts at $0.0313 per image. Settings: size 2K, 3K, 1:1, 4:3, 3:4, 16:9, 9:16, 3:2, 2:3, 21:9; up to 15 images per request; a source image is required. Id seedream-5.0-lite-edit-uncensored.
Vision Image Uncensored
A single price for every setting plus an optional reference image: quick drafts whose cost you know in advance. Starts at $0.04 per image. Settings: size 1:1, 4:3, 3:4, 16:9, 9:16; an optional reference image. Id vision-image-uncensored.
Vision Pro Image Uncensored
The higher-grade version of Vision Image. 1K or 2K output, jpeg or png. Starts at $0.09 per image on the resolution 2K tier. Settings: size 1:1, 4:3, 3:4, 16:9, 9:16, 2:3, 3:2, 21:9; resolution 2K; output jpeg, png. Id vision-pro-image-uncensored.
WAI Illustrious SDXL Uncensored
WAI Illustrious SDXL Uncensored is one of the uncensored image models on this key. Starts at $0.0136 per image on the size 512x512 tier. Settings: size 1024x1024, 1024x768, 768x1024, 1408x1024, 1024x1408, 1920x1088, 1088x1920, 512x512, 2048x2048; up to 4 images per request. Id illustrious-xl-uncensored.
Qwen Image 2.1 Uncensored and Vision Image Uncensored accept image_urls as an optional field, so the same id serves a fresh render and a variation on a reference. Anime Image Uncensored, CyberRealistic XL Uncensored, Qwen Image 2.1 Uncensored, WAI Illustrious SDXL Uncensored charge 1K and 2K output at different tiers, so on those models the resolution you send is what moves the cost.
Put plainly, by task. A product in several colourways, or a new background, is an edit job, since the source image carries the brief. Comics and illustration go to the anime model, whose negative prompt and seed hold a character steady across panels. Lettering inside the picture, or a cut-out on transparency, goes to Qwen Image, which lists both settings. High-volume drafts go to Vision at its single tier, and the final render of that brief goes to Chroma or Vision Pro at 2K. All of them sit behind one endpoint and one key, so moving between them means changing the model id and nothing more.
Video shares the key and the catalog. LTX-2.3 Image-to-Video Uncensored, Wan 3.0 Image-to-Video Uncensored, Wan 3.0 Video Uncensored is billed per second of output, with the rate set by resolution. The per-second figures are in the per-second video price post and on the LTX-2.3 Image-to-Video Uncensored model page, and the browser route is the uncensored AI video generator page.
Which settings move the price
size on the image endpoint uses the WIDTHxHEIGHT string from OpenAI. The API takes the two numbers, matches them to the closest aspect ratio in the model’s list, and where the model has a resolution setting derives it from the longer side. Ratio never affects the charge; resolution can. If a model’s size parameter takes arbitrary pixel values, your string is passed through as written.
resolution is the one that counts. Anime Image Uncensored, CyberRealistic XL Uncensored, Qwen Image 2.1 Uncensored, WAI Illustrious SDXL Uncensored have separate 1K and 2K tiers. Chroma Uncensored, Qwen Edit Uncensored, Qwen Image 3 Uncensored, Seedream 5.0 Lite Edit Uncensored, Vision Image Uncensored, Vision Pro Image Uncensored have a single tier, so the biggest output costs exactly what the smallest does. Where the tier matters, put resolution under input yourself rather than letting the API infer it from size. An explicit setting beats the mapped one, and the bill follows what the model actually receives.
n is a multiplier. CyberRealistic XL Uncensored (up to 4) and Qwen Edit Uncensored (up to 4) and Qwen Image 3 Uncensored (up to 4) and Seedream 5.0 Lite Edit Uncensored (up to 15) and WAI Illustrious SDXL Uncensored (up to 4) return several images from one request and bill the tier price per image. Ask a model without n for two images and you get a 400 telling you to send two requests, with no charge. Four single-image requests and one four-image request cost the same; the difference is one connection and one generation id instead of four.
Settings that cost nothing extra, on the models that list them: output_format, which can be png, jpeg or webp; seed, so a result can be reproduced; negative_prompt, for what should stay out of the picture; prompt_extend, present on some models and on by default there, which lets the model expand a terse prompt before it renders; and background: "transparent" to get a cut-out. Switch prompt_extend off when the wording has to be obeyed exactly, as with a product shot or a layout containing text. None of them moves the tier.
enable_safety_checker appears on 6 of the 10 image models, default false. Omitting it and sending false are the same thing, at the same price. There is no surcharge and no separate id for the unfiltered path.
The bill at 100 and 1,000 images
At these prices the count is what decides the bill, so this table repeats the arithmetic at 100 and 1,000 images with each model on its base tier. You pay ahead, by card, WeChat Pay or crypto, with no ID check: top-ups start at $10, bonus credit lands at +5% at $200, +10% at $500, +15% at $1,000, and a $200 top-up therefore covers 5% more images than the rows show. Those bonus steps are the same constants the console reads.
| Model | Base tier | Per image | Per 100 | Per 1,000 |
|---|---|---|---|---|
| Anime Image Uncensored | resolution 1K | $0.03 | $3.00 | $30.00 |
| Chroma Uncensored | any settings | $0.08 | $8.00 | $80.00 |
| CyberRealistic XL Uncensored | size 512x512 | $0.0136 | $1.36 | $13.60 |
| Qwen Edit Uncensored | any settings | $0.08 | $8.00 | $80.00 |
| Qwen Image 2.1 Uncensored | resolution 1K | $0.0375 | $3.75 | $37.50 |
| Qwen Image 3 Uncensored | any settings | $0.08 | $8.00 | $80.00 |
| Seedream 5.0 Lite Edit Uncensored | any settings | $0.0313 | $3.13 | $31.30 |
| Vision Image Uncensored | any settings | $0.04 | $4.00 | $40.00 |
| Vision Pro Image Uncensored | resolution 2K | $0.09 | $9.00 | $90.00 |
| WAI Illustrious SDXL Uncensored | size 512x512 | $0.0136 | $1.36 | $13.60 |
Two real jobs from those rows. A shop with 250 products wanting four variations of each needs a thousand edits, $80.00 on Qwen Edit Uncensored. Six panels on each of 24 comic pages is 144 panels, $4.32 at 1K on Anime Image Uncensored or $6.48 at 2K. No plan is involved: each accepted generation debits the balance once, and redoing a single panel costs a single panel.
Two guarantees hold the sum together. The quote comes from the request itself, before any work starts, and is taken when the model accepts the job, so a batch can never exceed tier price times count. A failed generation is refunded whole without you asking, so failures never appear on the bill.
Chat models on the same key are priced per token; the pricing page shows each one’s per-token input and output rates, read from the same catalog.
Other image APIs, compared plainly
Venice puts image, text, video and audio models behind a single API dashboard with zero data retention, and the same account carries its consumer app, so one login covers everything. Its API and plan prices move and are not recorded here; take them from venice.ai/pricing rather than from any figure in this post (description taken from Google’s results page, 2026-10-19).
The first organic hit for that search is a post whose endpoint is described by model: Seedream 5.0 Lite and Seedream 4.5 builds, per its page title on 2026-10-19. That is how an image API should be sold, by model name, and this post follows suit.
Self-hosting is cheaper per image and gives total control. On Reddit and in the local studios on GitHub, people run checkpoints on their own GPU: an image costs electricity, and checkpoint, LoRA and sampler are all their choice (results page read 2026-10-19). The price is the card, the setup time and the hours one machine spends on a batch. A per-image API wins when no GPU is available, when the job runs from a server, from a phone or inside an agent, and whenever the total has to be known before the first image.
The consumer tools on that results page, Perchance and Mage among them, suit someone who wants a web page and nothing to configure. This post is for the script, the app or the agent calling an endpoint that must know the cost of each call. The web-page version of these same models is the uncensored AI image generator page.
This key wins on the arithmetic. The per-image price is in three places before you send anything, the model page, the catalog endpoint and this table, and they agree. The request shape is OpenAI’s, so your existing SDK needs a base URL and a key and nothing else. Image and chat models draw from one prepaid balance on one key, no subscription is required, and a failed generation comes back without a support ticket. Every image model page, Anime Image Uncensored for one, carries its tiers and a ready-made first request.
Three ways to send the request
A single image in OpenAI’s format. size gives width by height in pixels, which the API translates to the model’s own ratio and resolution settings; n becomes the model’s image count where one exists.
curl https://api.unrestrictable.ai/v1/images/generations \ -H "Authorization: Bearer sk-unr-live-…" \ -H "Content-Type: application/json" \ -d '{"model": "qwen-image-2.1-uncensored", "prompt": "a lighthouse at dusk, oil on canvas", "size": "1024x1024", "n": 1}'The call blocks until the image is ready, up to roughly 100 seconds, then returns a URL on this API. Generated files stay for 30 days.
{ "id": "<generation id>", "created": <unix seconds>, "data": [{ "url": "https://api.unrestrictable.ai/v1/files/<path>.png" }]}The same thing from the OpenAI Python SDK. Whatever a model accepts beyond the OpenAI fields goes under input inside extra_body. Here the resolution is set explicitly and a negative prompt added, both of them settings the anime model lists.
from openai import OpenAI client = OpenAI(base_url="https://api.unrestrictable.ai/v1", api_key="sk-unr-live-…") image = client.images.generate( model="anime-image-uncensored", prompt="a courier on a rooftop at night, rain, neon signs, anime key visual", size="2048x2048", extra_body={"input": {"resolution": "2K", "negative_prompt": "text, watermark", "output_format": "jpeg"}},)print(image.data[0].url)Editing begins with an image you supply. Edit models require image_urls; text-to-image models that list the field accept it as an option.
curl https://api.unrestrictable.ai/v1/images/generations \ -H "Authorization: Bearer sk-unr-live-…" \ -H "Content-Type: application/json" \ -d '{"model": "qwen-edit-uncensored", "prompt": "replace the sky with a thunderstorm, keep the building", "input": {"image_urls": ["https://example.com/source.png"]}}'When a batch should not keep a connection open, post to /v1/generations and put the model’s native settings under input. The API returns 202 Accepted immediately, carrying the generation id and the price, and you collect the file once it is done.
curl https://api.unrestrictable.ai/v1/generations \ -H "Authorization: Bearer sk-unr-live-…" \ -H "Content-Type: application/json" \ -d '{"model": "qwen-image-2.1-uncensored", "input": {"prompt": "a lighthouse at dusk", "resolution": "2K", "size": "16:9"}}'# answers 202 with the generation id and its price; fetch the result withcurl https://api.unrestrictable.ai/v1/generations/<generation id> -H "Authorization: Bearer sk-unr-live-…"To get the bytes inline instead of a URL, send response_format: "b64_json" on the image endpoint, as OpenAI’s image reference documents. Any client needs three things, base URL, key and model id; the docs page gives them for the OpenAI SDK.
These builds run with no stock prompt filter, and enable_safety_checker is false by default. One rule applies to each image model: any prompt involving a minor gets content_not_allowed back, and nothing is charged.
The URL you get back is good for 30 days on this API; move the file into your own storage before then. Request bodies are not stored unless the key opted in to the console’s 7-day replay, so the prompt is gone once the image exists.
Errors, and which ones are billed
Every failure on the image endpoint comes back in one error shape, and the billing rule fits in a sentence: nothing is taken until the model accepts the job, and a job the model fails is refunded. The codes below are from the API reference.
400 invalid_request: an out-of-range setting, a video model, ornsent to a one-image-per-request model. No charge.401 invalid_api_key,403 scope_denied: a wrong key, or one scoped to other model ids. No charge.402 insufficient_balance,spend_cap_reached,monthly_limit_reached: the quoted price exceeds the balance, or a cap you put on the key in the console. No charge; top up, or raise the cap, then resend.404 model_not_found: the id is missing fromGET /v1/catalog/media. No charge.413: the body exceeds 1 MB. Pass image URLs instead of inline bytes.429 rate_limited, or its busy-model form: back off and retry. No charge.502 generation_failed: the model could not finish the image. No charge, and any amount already taken is returned.504 generation_pending: the model was still working when the image endpoint stopped waiting. This one is billed, since the job carries on; the response carries the generation id, and the file turns up atGET /v1/generations/{id}.
For a batch: money leaves only for a finished image, at its tier price, or for a pending image that completes later at the same price. Everything else leaves the balance untouched.
Where the tier prices come from
Every model in GET /v1/catalog/media carries a list of tiers. A tier is three fields: when, the settings it covers; price_usd, the per-unit price to four decimals; and list_usd, the developer’s own published price when one exists. Tiers are tested in order and the first match is billed, which is why a one-tier model with an empty when costs the same at any size.
- Only resolution changes the tier on multi-tier models; ratio, output format and seed never do.
nmultiplies the tier price by the number of images. A model with nonparameter produces one image per request and says so rather than billing twice.- An add-on tier, where one exists, is added to the base tier whenever its named settings apply; the first table lists add-ons on their own line.
- A failed generation is refunded in full. One still running when the image endpoint stops waiting is billed and keeps going; its id is in the 504 response and the file lands at the generations endpoint.
Those tiers are readable by any script, no key needed, exactly as this page reads them, so the resolution per job can follow the live figure:
curl -s https://api.unrestrictable.ai/v1/catalog/media | jq '.data[] | select(.id == "qwen-image-2.1-uncensored") | .pricing'The reply is that id’s tier list, when and price_usd per entry. Caching it for a few minutes is enough: the catalog changes only when a model is added or its price changes, and this page is rebuilt from it on the same cycle.
The console shows these tiers next to a panel that runs the model on one of your keys, which lets you confirm a price against a real charge before scripting the batch. The media page is the list itself; this post is the per-image arithmetic laid over it, refreshed from the catalog on every render.
Questions
Yes. Send the image in image_urls to Qwen Edit Uncensored and it is edited with no stock prompt filter, at $0.08 per output image. Qwen Image 2.1 Uncensored edits too when image_urls is present, and accepts a mask_url to limit the edit to one region. See the Qwen Edit Uncensored page.
Qwen Image 2.1 Uncensored and Vision Image Uncensored accept an optional reference image in image_urls; the edit models need one. Put the URL of your image in the request and the model builds on it rather than on a blank canvas.
Every one of the 10. Nothing rewrites or blocks the prompt, apart from the refusal of any prompt involving a minor. Chroma Uncensored costs $0.08 per image whatever the settings, and Qwen Image 2.1 Uncensored starts at $0.0375. See every uncensored image and video model.
Make a key in the console, set the OpenAI SDK's base URL to https://api.unrestrictable.ai/v1, and call images.generate with an id such as qwen-image-2.1-uncensored. You get back a URL to the image and pay the model's tier price times n. To skip the code, the same models run in the browser from the uncensored AI image generator page. See the setup page.
Venice sets its own API and plan prices and changes them, so this post quotes none. Check venice.ai/pricing for the current number and set it against the per-image table above, which is read from this catalog each time the page renders.
Yes: chroma-uncensored on this key, $0.08 per image on a single tier for every setting, through either the OpenAI image endpoint or the generations endpoint. See the Chroma Uncensored page.
Every model above is on the uncensored model list with its live price per million tokens. Per-token pricing sets the whole catalog side by side, and the family directory groups the builds by base model.