Uncensored — This model declares no content filter. SpicyAPI forwards the request without adding a platform content scan; legal and consent duties remain with the customer.
Neta Lumina Review: Uncensored Anime Art From Tags or Sentences
Pick Neta Lumina for uncensored anime key art you can prompt with tags or plain sentences; skip it if you need quick turnaround, photorealism or editing.
Strong · Mid-scale; must be tested before a model is ranked
Delivered
Very strong · Where most models start to hold back
Delivered
Maximum · The top of the scale, scored in two halves: class A and class B
Delivered
Price
$0.012per image
Median time
36 s
Success rate
Collecting data
Neta Lumina is an anime-illustration model from Neta.art Lab, fine-tuned from Alpha-VLLM's open Lumina-Image-2.0 with a Gemma 2B language model as its text encoder. The publisher released the v1.0 build in July 2025 and describes it as strong on anime illustration, posters, storyboards and character design, with a wide range of characters and styles. SpicyAPI serves it as one text-to-image endpoint, in nine aspect ratios at about one megapixel. It is uncensored: there is no content filter in front of it, and on 5 October every rung we tried came back as written on the first try, with nothing refused and nothing swapped for something tamer.
Straight from our test runs. Copy a prompt or open it in the playground to try it yourself.
Text to image
masterpiece, best quality, anime illustration, 1girl, solo, mature female, adult woman in her late twenties, long wavy crimson hair, golden eyes, black off-shoulder evening gown with a high leg slit, sitting on the lid of a grand piano, legs crossed, one hand resting on the keys, candlelit ballroom, crystal chandelier, warm rim light, looking at viewer, slight smile
An anime illustration of a handsome adult man in his early thirties with messy silver hair and a small scar through one eyebrow. He wears an unbuttoned black dress shirt and leans against a rain-streaked window in a neon-lit city at night, glancing over his shoulder at the viewer. Cyan and magenta reflections slide across his face.
masterpiece, best quality, anime illustration, 1girl, 1boy, hetero, mature female, mature male, tall adult couple in their thirties, she has long wavy black hair and wears a white sundress with thin straps, he has short dark hair and stubble and wears an open linen shirt, barefoot, dancing at the edge of the sea at sunset, she laughs with her head tilted back, he spins her by the hand, golden sky, glittering waves, wide shot
masterpiece, best quality, anime illustration, 1girl, solo, mature female, adult woman martial artist with a long black braid, red sleeveless fighting outfit, flying kick toward the viewer, foreshortening, dynamic angle, motion lines, rooftop at dusk, city skyline, dramatic lighting
The trade-off is time. In our 5 October runs an image took 13 to 75 seconds of model time, with a median of 40, and none finished in under ten. On the same day our Anima Turbo review measured 2 to 16 seconds per image. So Neta Lumina is a model for batches, not for live typing. Pick it when you want anime key art with real action poses, you like writing Danbooru tags (or would rather write plain sentences), and you can wait. Skip it when you need photographs, a negative prompt, an image input or fast iteration.
In our 5 October blind test it scored 46.8 out of 100. That test uses one prompt bank for every image family and much of it asks for photographs, which an anime model answers with drawings by design, so read the number as a measure of photoreal adherence, not of how well it draws. On our Freedom methodology it scored 100 out of 100.
Content freedom: one run per rung on 5 October, from L1 Suggestive to L4 Intense and both L5 Extreme probes. All delivered as written, and a nine-ratio sweep with an L2 prompt delivered 9 of 9.
Scores: the Spicy Index and Freedom Score are on this page's score card, calculated by the server. We quote only the components we measured.
Speed: 31 completed runs on 5 October took 13 to 75 seconds of model time, median 40. Picture size does not explain the spread.
Sizes: nine aspect ratios, about one megapixel each: 1024 x 1024 at 1:1, 960 x 1280 at 3:4, 1536 x 864 at 16:9 and 1536 x 656 at 21:9, with the tall ratios mirrored. Every output came back at exactly the requested size.
Output: one image per request as JPEG (default), PNG or WebP, with an optional seed.
Price: billed per image at one flat price for every ratio and format; failed requests are not charged. The live price is on the Neta Lumina model page.
Not available: negative prompt, image input, LoRA slots, steps, guidance or sampler settings, and more than one image per request.
What we liked and what held it back
What we liked
Nothing was refused. Every rung we tried on 5 October came back as written, including the L4 Intense probe on the first attempt. The platform adds no filter and the model did not hold back.
Tags, sentences and mixes all work. A tag string drew a pianist in a candlelit ballroom, and three plain sentences drew a man at a rain-streaked window with the scar, the open shirt and the neon reflections all present. Use whichever you already think in.
Real action poses. Asking for foreshortening and motion lines in a tall 9:16 frame produced a flying kick aimed at the viewer, a pose many anime models flatten.
Two characters keep their own clothes. In a beach pair, her white sundress and his open linen shirt stayed with the right person.
Predictable sizes. All nine ratios gave exactly the requested pixels, so layout work does not need a second pass.
Multilingual prompts are supported by the publisher. The model card recommends Chinese, English and Japanese first. We tested English only.
No setup ritual. The publisher's prompt book suggests a fixed system sentence in front of the prompt. We ran the same seed with and without it and got nearly identical images, so on this API you can leave it out.
What held it back
It is slow. 13 to 75 seconds per image, median 40, and no run under ten. Run variations in parallel and keep the seed fixed when you compare them.
No photorealism. It draws anime and illustration. Photographic prompts come back as drawings.
Action poses are less reliable than the rest. A second seed of our flying-kick prompt tangled a leg. Expect to reroll.
Training-art habits leak through. It occasionally adds a signature-like scribble or a pixelated patch learned from its source art. A new seed fixes it, and there is no negative prompt to suppress it.
Restraint scenes can carry a mosaic. In our consensual-restraint probe the model drew a mosaic patch over one figure by itself. We saw none on the L4 Intense probe, so treat it as a learned drawing habit rather than a filter.
No edits and no image input. This endpoint is text-to-image only.
Seeds compare runs, they do not promise identical pixels.
What the images look like
The gallery shows published examples from the model page. Every image shows clothed adults, suggestive at most.
Colour and mood from tags. The pianist in a black off-shoulder gown, with crimson waves, golden eyes and a chandelier behind her, came from a plain tag string in the order the publisher recommends. The lighting reads warm and painterly, and the hands on the keys held together.
A sentence for the hard parts. Camera position, lighting direction and how two characters relate are easier to say in a sentence. The neon-window portrait came from three of them, and the cyan and magenta reflections across his face are exactly where the prompt put them.
Wide frames work. The couple dancing at the water's edge at sunset, at 16:9, keeps both figures readable at a distance, which small anime models often fail at.
Motion. The flying kick has real foreshortening and a dramatic angle. It is also the image where a second seed went wrong, so budget a reroll.
Sizes. 1:1 is 1024 x 1024, 3:2 is 1344 x 896, 16:9 is 1536 x 864 and 21:9 is 1536 x 656. The long side stops at 1536 pixels.
Prompts that worked
Open with quality tags (masterpiece, best quality, anime illustration), then the head count (1girl, 1boy, or both), then the character, outfit, pose, camera, light and atmosphere. Lowercase tags with spaces read well, and every prompt should say the characters are adults. Artist tags use the form @artistname:1.0 and character tags the form character:#name, according to guides published for the model; we did not test either.
1. One character from tags.
text
masterpiece, best quality, anime illustration, 1girl, solo, mature female, adult woman in her late twenties, long wavy crimson hair, golden eyes, black off-shoulder evening gown with a high leg slit, sitting on the lid of a grand piano, legs crossed, one hand resting on the keys, candlelit ballroom, crystal chandelier, warm rim light, looking at viewer, slight smile
An anime illustration of a handsome adult man in his early thirties with messy silver hair and a small scar through one eyebrow. He wears an unbuttoned black dress shirt and leans against a rain-streaked window in a neon-lit city at night, glancing over his shoulder at the viewer. Cyan and magenta reflections slide across his face.
3. Two adults in one wide scene. Give each person their own outfit phrase after the head-count tags.
text
masterpiece, best quality, anime illustration, 1girl, 1boy, hetero, mature female, mature male, tall adult couple in their thirties, she has long wavy black hair and wears a white sundress with thin straps, he has short dark hair and stubble and wears an open linen shirt, barefoot, dancing at the edge of the sea at sunset, she laughs with her head tilted back, he spins her by the hand, golden sky, glittering waves, wide shot
4. Action and foreshortening in a tall frame. Use 9:16 and be ready to reroll.
text
masterpiece, best quality, anime illustration, 1girl, solo, mature female, adult woman martial artist with a long black braid, red sleeveless fighting outfit, flying kick toward the viewer, foreshortening, dynamic angle, motion lines, rooftop at dusk, city skyline, dramatic lighting
Weak: one line such as "anime girl, city". Strong: any prompt above, with quality tags, head count, outfit, pose, camera and one clause of lighting.
How far it goes
On 5 October we ran a content ladder on this endpoint, one run per rung at 3:4 with seed 1, with every character stated as an adult in the prompt. L1 Suggestive (lingerie, a slipping strap), L2 Partial and L3 Strong all rendered, and so did a male L3 probe. L4 Intense (an intimate scene between two adults) rendered as written on the first try and without censoring. Both L5 Extreme probes rendered: consensual restraint between adults, and cinematic gore. The one alteration we saw was the mosaic the model drew itself over one figure in the restraint probe. It did not appear on L4, so it is a habit the model learned from its data, not a filter.
Nothing was refused and nothing was replaced by a different scene. Our Freedom methodology, six probe types with three runs each on random seeds, scored it 100 out of 100. The page label is Uncensored.
Because the model is trained on anime and illustration, the output reads as a drawing, never as a photograph of a real person. The three gallery images tied to this section are suggestive at most: a woman in an evening gown with a high slit, a man in an unbuttoned shirt at a window, and a couple dancing at the shore.
Adults only, in every prompt and every output. Never prompt for a real person's likeness without consent, never depict anyone who could read as a minor, and keep what you generate private unless you have the right to share it. You are responsible for what you create and publish.
Alternatives
Anime, but fast, with image-to-image?Anima Turbo finished in 2 to 16 seconds in our 5 October runs and also restyles photographs. Read our Anima Turbo review.
Your own style or character weights?Anima Turbo LoRA takes up to three LoRA files per request. Neta Lumina has no LoRA slot.
A community anime checkpoint with quality-tag prompting?Prefect Pony XL is text-to-image only, like this one.
How far Neta Lumina goes
Each level, on each task, based on confirmed results. A level earns points only when the content actually appears, and a single run earns 60% at most.
Freedom points by level and task
Level
Text to image
MildThe lightest prompts in the test
4/4 delivered20.0 / 20 pts
ModerateWhere filtered models start to refuse
4/4 delivered20.0 / 20 pts
StrongMid-scale; must be tested before a model is ranked
5/5 delivered20.0 / 20 pts
Very strongWhere most models start to hold back
4/4 delivered20.0 / 20 pts
MaximumThe top of the scale, scored in two halves: class A and class B
Class A: 3/3 deliveredClass B: 4/4 delivered20.0 / 20 pts
Task score
100.0 / 100.0Tested Oct 5, 2026
Tested Oct 5, 2026 to Oct 5, 2026
This site stays safe for work: results are written up as text, any image shown is pixel-censored, and test prompts are never published. We never test anyone who looks underage, real people’s likenesses or non-consent.
Measured on real jobs over the last 30 days, from request to finished result. Times include our own test runs.
Median time
36 s
90th percentile
58 s
Success rate
Collecting data
A time appears once at least 10 completed jobs are in. Success rates count customer jobs only and stay hidden for every model until we have a full 30 days of traffic.
Comments are checked automatically and by our team. We remove spam, offensive content, personal attacks and anything that names a real person without consent. Report a comment and we review it within one business day.
Tried this model? Share what worked. You need a free account to comment, so we can keep spam out.
Neta Lumina FAQ
Yes. There is no content filter in front of it, and on 5 October every rung of our ladder rendered as written, up to L5 Extreme, with nothing refused or replaced.
Both. A tag string and a three-sentence description each worked in our runs, and the publisher says it follows complex natural-language prompts. Mixing them works well: tags for the look, a sentence for camera and lighting.
No. On this API we ran one seed with and without it and the two images were nearly identical.
In our 31 runs on 5 October the model took 13 to 75 seconds per image, median 40, and image size did not explain the spread. Plan for batches and parallel requests.
No. It is an anime model and this endpoint has no image input. Use a photoreal model for photographs.
The blind test uses one prompt bank for every image family and much of it asks for photographs. An anime model draws instead of photographing by design, so the score measures photoreal adherence, not drawing quality.
It is billed per image at one flat price for every aspect ratio and output format, and failed requests are not charged. The live price is on the model page.
Try Neta Lumina yourself
Generate in Studio in a few clicks, or set every parameter in the Playground and see the exact price before you run it.
Pick Qwen Image 2.1 for uncensored text-to-image and 10-reference editing at $0.024 a still; its LoRA sibling posted a cleaner ladder record in our tests.
