Anima Turbo LoRA vs Anima Turbo: which to pick?

Pick Anima Turbo LoRA when you have a LoRA file for a specific style or character: up to three per request, scale 0 to 4, and loading them adds no separate charge. Pick Anima Turbo when the base look is enough, since it is the simpler request. Both share the same sizes, 15 aspect ratios and edit endpoint, and both bill per image by resolution tier.

Which LoRAs work here?

Files trained for Anima, of which Hugging Face listed 129 on 2026-10-05. The publisher recommends training on the Base build, and such LoRAs apply to Turbo. Pass the direct HTTPS link to the .safetensors file. LoRAs made for other models load without an error and change nothing, so run a with-and-without test first.

Can I use uncensored LoRAs?

Yes. The weights are not filtered and neither is the output. With and without LoRAs our strongest content probe rendered as written, in text-to-image and in image editing, across repeated seeds. Rights and consent for everything you generate stay with you.

What scale should I use?

Start at 1, the full style. Go down to about 0.5 for a light touch and up to about 1.2 to push a little; 1.5 pushed the style harder in our test, and 4 broke the image apart. A scale of 0 is the same as no LoRA.

Do I need the trigger word?

It depends on the file. Our monochrome LoRA stayed in colour until “monochrome, greyscale” was in the prompt, and the PC-98 LoRA changed the style even without “pc98gal_style”. Read the model card and test both ways.

What strength works for edits with a LoRA?

A photo took on the PC-98 style at strength 0.75 with its composition kept. Turning a colour illustration into manga ink needed 0.85, since at 0.8 the colour stayed, and at 0.85 the pose is redrawn. Strength 0 returns the source untouched and 1.0 ignores it.

How fast is it, and how is it billed?

In 60 runs across every aspect ratio, both tiers and both families, model time was 2 to 16 seconds, with 52 of 60 under 10. A first request with a new LoRA can take longer while the file is fetched. Billing is per image by tier, 1k or 1.5k; the 1.5k tier costs more, the ratio does not matter, and LoRAs add no separate charge. Current prices are on the model page.

Why did my LoRA request fail?

Usually the link needs a login, since a gated Hugging Face repository answers with HTTP 401, or it points at a repository page instead of the file, which asks for the direct download link. Use a public URL that downloads the .safetensors file itself. Failed requests are not charged.

Anima Turbo LoRA API: Uncensored NSFW Anime Images

circlestone-labs/anima-turbo-lora/text-to-image

Anime and illustration art on Anima Turbo with up to three of your own style or character LoRAs per request, each with its own scale. Tags or sentences, fifteen aspect ratios, 1k or native 1.5k, and no filter on prompts, weights or results.

From $0.018/image

Run model View API

Input

Text to image

Prompt319/5000

Aspect ratio

Resolutionaffects price

Style LoRAs1/3

Pc98 Gal Style

Seed

Safety checkerOff

SpicyAPI does not run a safety checker on this model. Your prompt and media go to the model as written, and the model itself decides what comes back.

Checking your account before enabling submission and pricing.

Output

Ready for a new run

Example result

Animate

Edit

Split into layers

{
  "model": "circlestone-labs/anima-turbo-lora/text-to-image",
  "input": {
    "aspect_ratio": "4:3",
    "resolution": "1k",
    "loras": [
      {
        "path": "https://huggingface.co/Melon-Bread/anima-pc98-gal-style-v0.2/resolve/main/pc98gal_style-v02_anima.safetensors",
        "scale": 1
      }
    ],
    "seed": 42,
    "prompt": "pc98gal_style, dithering, limited color palette, flat cel shading, thin linework, safe, 1girl, solo, adult woman, mature female, long dark blue hair, red bomber jacket worn open over a black tank top, leaning against an arcade cabinet, neon signs, night arcade, confident smirk, looking at viewer, retro anime aesthetic"
  }
}

Introduction

Anima Turbo LoRA Text to image draws one anime illustration from a written prompt with up to three of your own LoRA files loaded. It runs on Anima Turbo, the distilled 8-to-12-step build of CircleStone Labs' 2-billion-parameter anime model. Pick any of fifteen aspect ratios and one of two resolution tiers. The platform adds no content filter to the prompt, the weights or the image.

Choose a mode by starting material

What you have	Mode
a written description	Anima Turbo LoRA Text to image
a picture to restyle	Anima Turbo LoRA Image to image

Key features & highlights

Community LoRAs by direct link

Hugging Face lists more than a hundred LoRAs trained for Anima: retro game styles, manga ink, painterly looks, characters and unfiltered styles. Pass the direct file link and a scale; nothing is uploaded to us in advance.

Three at once

Up to three LoRAs per request, each with its own scale. In our test a PC-98 style at 1, a monochrome LoRA at 0.6 and an aesthetic LoRA at 0.25 loaded together and the main style still read clearly.

Seconds per image

Across 60 test runs the model took between 2 and 16 seconds per image and usually less than 10. The first request with a new LoRA can take longer while the file is fetched.

Uncensored with custom weights

Our strongest content probe came back as written with a LoRA loaded, with three stacked and with none. The weights are never reviewed and the output is never filtered.

Request parameters

Request parameters for Anima Turbo LoRA · Text to Image
Parameter	Required	Description
`prompt`	Yes	Put the LoRA's trigger words first, then tags or sentences for the picture, up to 5,000 characters. There is no negative prompt. at least 1 characters · at most 5000 characters
`aspect_ratio`	No	Fifteen options from 3:1 to 1:3; 1:1 by default. The ratio changes the shape, not the price. `1:1` · `16:9` · `9:16` · `4:3` · `3:4` · `3:2` · `2:3` · `21:9` · `9:21` · `2:1` · `1:2` · `3:1` · `1:3` · `5:4` · `4:5` · Default: `1:1`
`resolution`	No	1k is about one megapixel and is the default. 1.5k is about 2.4 megapixels, generated natively, and costs more. `1k` · `1.5k` · Default: `1k`
`loras`	No	Up to three {path, scale} entries trained for Anima; scale 0 to 4, 1 is neutral and 0.5 to 1.2 is the useful range. Use a public direct link to the .safetensors file. Default: `[]` at most 3 items
`seed`	No	Keeps a set of images comparable while you change one LoRA scale at a time. It does not guarantee the exact same pixels on every run. `≥ 0` · `≤ 2147483647`

Pricing

Billed per image by resolution tier. The aspect ratio does not change the price, and loading LoRAs adds no separate charge.

Pricing
Specification	Price
`1k`	$0.018 / image
`1.5k`	$0.036 / image

Use cases

Style batches

A style LoRA plus a list of prompts becomes a consistent set of illustrations, sprites or thumbnails. Fix the seed per layout and swap only the subject line.

Character sheets

Keep a character LoRA and a seed fixed and change only the outfit, place and light between prompts. The face and build stay recognisable across the whole set.

Uncensored stills for your platform

Load the LoRA your audience expects and write the image directly. There is no content flag and no second review of the result.

Explore key use-case tutorials

Prompt guide & pro tips

Trigger words first

Check the LoRA page for its trigger and put it at the very start. Some styles need plain tags instead of a made-up word: our monochrome LoRA only went black and white once "monochrome, greyscale" was in the prompt.

Compare against no LoRA

Run the same prompt and seed without the LoRA. If the two results are identical, the file was trained for another base model: such LoRAs load without an error and do nothing.

Stay near 1

The scale field accepts 0 to 4, but 0.5 to 1.2 is where the results are usable. At 4 the image broke apart in our test.

Use direct, public links

The link must download the .safetensors file itself. A repository page or a file that needs a login fails the request, which is then not charged.

API quick start

# Generate the idempotency key once; reuse it when retrying after a lost response.
export SPICY_IDEMPOTENCY_KEY="${SPICY_IDEMPOTENCY_KEY:-$(uuidgen)}"
curl -X POST https://api.spicyapi.ai/api/v1/jobs/createTask \
  -H "Authorization: Bearer $SPICY_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $SPICY_IDEMPOTENCY_KEY" \
  -d '{
  "model": "circlestone-labs/anima-turbo-lora/text-to-image",
  "input": {
    "aspect_ratio": "1:1",
    "prompt": "A ceramic teapot on a linen cloth, soft window light, muted palette",
    "resolution": "1k"
  }
}'

# Copy data.taskId from the response into TASK_ID, then poll the status.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.spicyapi.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
  -H "Authorization: Bearer $SPICY_API_KEY"

See the API reference for authentication, the complete schema, polling, and error handling.

Related models and next tasks

Missing the model you need?

Tell us what to add and where you would use it. Requests go to the people who decide what gets onboarded next.

Give Anima your own style, one LoRA file at a time

Anima Turbo LoRA loads up to three LoRA files onto Anima Turbo by CircleStone Labs, a 2-billion-parameter anime model built on NVIDIA Cosmos-Predict2 and distilled for 8 to 12 sampling steps. Hugging Face listed 129 Anima LoRA adapters on 2026-10-05, and files trained for Anima apply to this Turbo build; the publisher recommends training them on the Base build. Images took 2 to 16 seconds in our runs, with 52 of 60 under ten. Nothing filters the prompt, the weights or the image: with and without LoRAs, our strongest content probe rendered as written.

Open the playground

Give Anima your own style, one LoRA file at a time

LoRAs per request: Up to 3, each {path, scale}; path is a direct HTTPS link to a .safetensors file
Scale: 0 to 4, default 1; about 0.5 to 1.2 is the useful range, and 4 broke our image apart
Speed: 2 to 16 seconds per image; the first request with a new LoRA can take longer while the file is fetched
Editing: One source image, strength 0 to 1 (default 0.8), with the same loras array; the result keeps the source's shape
Price: Billed per image by tier, 1k or 1.5k; LoRAs add no separate charge, and failed requests are not charged
Content filter: None; our strongest content probe rendered as written, with and without LoRAs

Trigger words: Check whether the LoRA wants its wordOur monochrome LoRA stayed in colour until “monochrome, greyscale” was in the prompt, while the PC-98 LoRA changed the style even without its trigger “pc98gal_style”. Behaviour differs from file to file, so read the model card, then test the prompt both ways.
Wrong base model: A file for another model fails silentlyWe loaded a LoRA trained for FLUX and got no error and no effect: the output was pixel-identical to running without it. If the same prompt and seed look the same with and without the file, it was almost certainly trained for another base model.
Limits: What it will not doMore than three LoRAs per request is rejected, one request returns one JPEG, and there is no negative prompt. The link must be public and point straight at the file: a gated Hugging Face repository answers with HTTP 401 and a repository page link asks for the direct download link. Neither failure is charged.

Working with LoRAs on Anima Turbo LoRA

Run the with-and-without pair first

A with-and-without pair tells you whether a LoRA did anything. Same plain prompt, same seed, once with no LoRA and once with the PC-98 LoRA at 1: the first came back as modern anime and the second in a PC-98 era game look, even though the prompt never used the LoRA's trigger word.

Climb the scale in small steps

Scale 0.5 to 1.2 is the useful range, and you reach it in small steps. We ran the PC-98 LoRA at 0, 0.5, 1, 1.5 and 4 on one seed. 0 is identical to no LoRA, 0.5 is a light touch, 1 is the full style, 1.5 pushes harder, and 4 broke the picture into abstract shapes (the last tile). Our prompt carried the trigger words, which is why the first tile already leans retro.

Stack up to three
Up to three LoRAs can be stacked in one request. Three at 1, 0.6 and 0.25 worked together and the main style still read clearly. The two-LoRA band picture in the gallery below used the PC-98 LoRA at 0.8 and an aesthetic LoRA at 0.5. Keep the main style highest and lower the others, then raise them in small steps; a fourth LoRA is rejected.
Give it a link it can fetch
The LoRA must be a public, direct link to the .safetensors file. It is fetched when the request runs, so the first request with a new LoRA can be slower; one of ours took almost three minutes, the others between 2 and 26 seconds. Link to the file itself, not the page that describes it.
Keep the usual tag habits
LoRAs change the look, not the prompt grammar. Start with “masterpiece, best quality” and a rating tag such as “safe”, then the character and scene tags, then the LoRA's own trigger words if its card lists any. For an artist's style, put @ in front of the name.

Who uses Anima Turbo LoRA

Comic and manga makers
Ink pages from one monochrome LoRA
With notverriegod/monochrome-grayscale_style at 1 and “monochrome, greyscale” in the prompt, we got black-and-white pieces with screentone: a swordsman on a rainy bridge and a woman leaning on a motorcycle. Without those words the same file stayed in colour, so put them first.
Game and retro-art studios
A PC-98 look on demand
Melon-Bread/anima-pc98-gal-style-v0.2 gave us a noir detective, a seaside town at dusk with no characters, and an arcade portrait. Its trigger word is “pc98gal_style”. The band at the bottom left combines it at 0.8 with an aesthetic LoRA at 0.5.
Creators with photos to restyle
A photo straight into a LoRA style
Left is a source photo, right is the result of the edit endpoint with the PC-98 LoRA at strength 0.75. The porch, the guitar and the pose stayed where they were while the rendering changed completely. Use the same loras array on edit as on text-to-image.
Illustrators changing medium
A colour illustration into manga ink
Turning a finished colour illustration into manga ink needed strength 0.85: at 0.8 the colour stayed. At that strength the pose is redrawn, so the right-hand picture shares the setting and the mood with the left, not every line.

From first LoRA to finished batch with Anima Turbo LoRA

Find a LoRA trained for Anima
Read its model card for the trigger words and a suggested strength, then copy the direct download link to the .safetensors file.
Run it with and without the file
Same prompt, same seed, one run each. If nothing changes, the file was probably trained for another base model.
Ladder the scale
Hold everything else still and try 0.5, 1 and 1.2. Keep the value where the look lands without taking over the subject.
Add a second or third LoRA
Bring each extra file in at a lower scale than the main one and raise it in small steps. A fourth is rejected.
Restyle photos on the edit endpoint
Send the same loras array with a source image, start at strength 0.75, and move in small steps. Then run the batch.

Anima Turbo LoRA prompts worth copying

PC-98 noir

With the PC-98 LoRA at 1, keep the trigger tags at the front and write the scene in plain tags.

pc98gal_style, dithering, limited color palette, flat cel shading, thin linework, safe, 1boy, solo, adult man, mature male, detective, trench coat, fedora, loosened tie, standing under a streetlamp in the rain, noir, serious expression, retro anime aesthetic

Manga ink portrait

With the monochrome LoRA at 1. The first two tags are the ones that switch the colour off.

masterpiece, best quality, safe, monochrome, greyscale, 1girl, solo, adult woman, mature female, short black bob, leather jacket, hand in pocket, leaning on a motorcycle, city backstreet, wind, cool gaze, ink shading, screentone

A photo into PC-98

For the edit endpoint with the PC-98 LoRA at strength 0.75. Describe the whole picture, not only the change.

pc98gal_style, dithering, limited color palette, flat cel shading, thin linework, safe, 1boy, solo, adult man, mature male, dark curly hair, short beard, blue chambray shirt with rolled sleeves, jeans, barefoot, sitting on wooden porch steps playing an acoustic guitar, golden hour, field and trees behind him, retro anime aesthetic

All Anima models2

Anima TurboText to image · Image editingFrom $0.012per image
Anima Turbo LoRACurrentText to image · Image editingFrom $0.018per image

LoRA guide

Which LoRAs load on Anima Turbo LoRA

Workbench