Skip to main content
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.

MOSS-Image LoRA Review: Uncensored NSFW, Exact Text, Your Own LoRAs

Pick MOSS-Image LoRA to run community style and portrait LoRAs where lettering must stay exact; skip it for edits, reference images or quick drafts.

Open the Playground

Spicy Index

58.5/ 100

Preliminary

Freedom Score

54.0/ 100

Range if the rest pass: 54.0–100.0

Content levels

Tested 5 of 6 probe types · 5 runs

  • Mild · The lightest prompts in the test

  • Moderate · Where filtered models start to refuse

  • Strong · Mid-scale; must be tested before a model is ranked

  • Very strong · Where most models start to hold back

  • Maximum · The top of the scale, scored in two halves: class A and class B

Price
$0.025 per image
Median time
29 s
Success rate
Collecting data

MOSS-Image LoRA runs OpenMOSS's open 5-billion-parameter text-to-image model, released on 16 October 2026 under Apache-2.0, with up to three of your own LoRA files per request. It is the same base model as MOSS-Image: with no LoRA loaded it returned the same picture for the same prompt and seed. It makes pictures from text only, with no reference images and no editing, and SpicyAPI adds no content filter, so the page label is Uncensored.

What MOSS-Image LoRA made in our tests

Straight from our test runs. Copy a prompt or open it in the playground to try it yourself.

  • Text to image

    moss_psychpop_style, a 1969 psychedelic concert poster: a woman with long wavy hair and flowers in it, in a fringed suede vest, dancing with her eyes closed, swirling orange, magenta and teal patterns around her. Large melting bubble letters across the top: "SUMMER OF LOVE", and at the bottom: "GOLDEN GATE PARK · AUGUST 3".

    Try this model

  • Text to image

    moss_beauty_style, close-up editorial portrait of a man in his thirties with warm brown skin, a short groomed beard and damp curls, a loose white linen shirt open at the collar, soft window light, dewy skin, calm direct gaze, neutral warm background.

    Try this model

  • Text to image

    moss_beauty_style, full-length editorial fashion photograph of a woman in her late twenties with long copper hair, in a floor-length red satin gown, leaning against a marble column in a sunlit palazzo hallway, soft gold light, dewy skin, confident glance at the camera.

    Try this model

  • Text to image

    moss_vgouache_style, an adult couple slow dancing on a lantern-lit terrace by the sea at night, she in a red silk dress, he in a white linen shirt

    Try this model

Make your own with MOSS-Image LoRA

Open the Playground

The verdict

We tested it on 21 October 2026, with a follow-up on 22 October, using fictional adults only. All five intensity levels came back as written on the base model with no LoRA, and with an unfiltered community LoRA at 0.7 the Strong-and-above probes also came back as written. Nothing was refused. LoRAs trained on either of the two published checkpoints, Base and SFT, loaded, in both the diffusers and the ComfyUI file formats, and three stacked LoRAs ran normally.

What sets this endpoint apart is the lettering. With a poster-style LoRA at scale 1, the title and the date line on a concert poster were still spelled correctly, and a portrait LoRA ran at the 2k tier. Pick it for posters, covers and styled portraits where the words must be right; skip it if you need to edit an image, feed in references or iterate fast, because it takes about 20 seconds at 1k and about a minute at 2k. We did not run a separate blind quality test on the LoRA endpoint, so no quality score is quoted; the score card on this page shows the live Spicy Index and Freedom Score.

Try MOSS-Image LoRA →

Key facts

  • Content limits: no platform filter, and the model weights do not review prompts. Nothing was refused or toned down at any of the five levels in our runs.
  • LoRAs: up to 3 entries per request, each {path, scale}, scale from 0 to 4 (1 is the default). path is a direct HTTPS link to a .safetensors file; put the LoRA's trigger word first in the prompt.
  • Parameters: prompt (1 to 12,000 characters, or a JSON layout), aspect_ratio (15 options, default 1:1), resolution (1k, 1.5k or 2k), output_format (JPEG, PNG or WebP) and seed. One image per request.
  • Not available: negative prompt, guidance or step controls, reference images, editing.
  • Sizes: at 1k, 1:1 is 1024 by 1024 and 16:9 is 1360 by 768; at 2k, 9:16 is 1536 by 2720.
  • Speed: about 18 seconds at 1k with a LoRA loaded (a new file adds time on its first use), about 38 seconds at 1.5k and about a minute at 2k.
  • Price: billed per image by resolution tier; loading LoRAs, the aspect ratio and the file format do not change it, and failed requests are not charged. See the model page.

What we liked and what held it back

What we liked

  • Text survives a style LoRA. The poster-style LoRA at 1 kept both quoted lines of lettering correct.
  • Base and SFT files both load. LoRAs trained on either checkpoint worked, so you are not limited to files trained on one of them.
  • Both file layouts work. The diffusers format (pytorch_lora_weights.safetensors) and the ComfyUI format each loaded.
  • Style LoRAs show clearly at 1. A gouache LoRA redrew a plain scene as a painting, and a portrait LoRA gave a dewy editorial finish.
  • Stacking works. Three LoRAs at 0.7, 0.8 and 0.5 ran together.
  • Same price however many LoRAs you load. Zero, one or three cost the same.
  • No content filter. Every level of our ladder came back as written on the base model.

What held it back

  • Text only. No reference image, no editing, no variations from a source picture.
  • Not fast. About 20 seconds at 1k and about a minute at 2k, plus extra time the first time a new file is used.
  • Lettering still slips occasionally. On the base model, a wordmark came out with letters stuck together on a second seed. Rerun with another seed when exact spelling matters.
  • Some files do little. A realism-finish LoRA at 1.0 changed the picture only slightly, and two Civitai files that need a sign-in returned HTTP 401 when linked directly.
  • The link has to be the file. A repository page link fails with "Failed to load LoRA ... provide the direct download link", and a gated file fails too. Failed requests are not charged.
  • No negative prompt or step controls. Leave out what you do not want.
  • Seeds are close, not identical. Expect similar repeats, not matching pixels.

What the images look like

The gallery shows the four published examples from the MOSS-Image LoRA model page, all text-to-image with a community LoRA at 1. Everything in it is clothed and safe for work; our content-test images are not shown and those results are in words.

A poster with lettering. A 1969-style psychedelic concert poster at 2:3 with a bubble-letter title across the top and a date line at the bottom, made with a poster-style LoRA. Both lines are spelled right.

A close editorial portrait. A man in his thirties in a white linen shirt in soft window light at 4:5, made with a portrait LoRA.

Full length at 2k. A woman in a red satin gown against a marble column in a sunlit hallway at 9:16, 1536 by 2720, with the same portrait LoRA.

A painted look from one sentence. A couple slow dancing on a lantern-lit terrace by the sea at night, as a gouache painting from a single short sentence plus the trigger.

Prompts that worked

Open with the LoRA's trigger word, then describe the picture in plain sentences. Put every piece of text to render in quotation marks and say where it goes. Write every person as an adult.

1. Style LoRA with lettering. Poster-style LoRA at scale 1, aspect_ratio 2:3, 1k.

text
moss_psychpop_style, a 1969 psychedelic concert poster: a woman with long wavy hair and flowers in it, in a fringed suede vest, dancing with her eyes closed, swirling orange, magenta and teal patterns around her. Large melting bubble letters across the top: "SUMMER OF LOVE", and at the bottom: "GOLDEN GATE PARK · AUGUST 3".

2. Portrait LoRA at 2k. Portrait LoRA at scale 1, aspect_ratio 9:16, resolution 2k.

text
moss_beauty_style, full-length editorial fashion photograph of a woman in her late twenties with long copper hair, in a floor-length red satin gown, leaning against a marble column in a sunlit palazzo hallway, soft gold light, confident glance at the camera.

3. Painting LoRA, one sentence. Gouache LoRA at scale 1, aspect_ratio 1:1.

text
moss_vgouache_style, an adult couple slow dancing on a lantern-lit terrace by the sea at night, she in a red silk dress, he in a white linen shirt

Weak: a LoRA link and no trigger word. Strong: trigger first, quoted text with its position, and a second run at scale 0 on the same seed.

How far it goes

SpicyAPI adds no content filter and the weights do not review prompts. On 21 October 2026 we ran our standard ladder on the base model with no LoRA loaded: five intensity levels from Mild to Maximum (the test prompts are not published), one run each, plus a male-subject version of Strong and two kinds of Maximum. Every character was written as an adult.

LevelBase model, no LoRAWith an unfiltered community LoRA at 0.7
Milddelivered as writtennot run
Moderatedelivered as writtennot run
Strongdelivered as written (female and male subjects)delivered as written (female and male subjects)
Very strongdelivered as writtendelivered as written
Maximumdelivered as written (both types)delivered as written (one of the two types)

Nothing was refused, nothing was toned down and no clothing was substituted. With the unfiltered LoRA loaded, the look shifted toward casual photography. These are single runs, so read them as "it does not refuse" and not as "every image will be perfect".

Adults only, in every prompt and output. Never depict a real person or likeness without consent, and remember that you are responsible for what you generate and publish and for the licence of any LoRA you load. Generated media is kept for 14 days and then deleted.

Loading your own LoRA

Point loras at up to three {path, scale} entries; for Hugging Face, link the /resolve/main/ address of the file. We ran the endpoint's published cases and all three landed: a poster-style LoRA at 1 changed the whole look with the lettering intact, a portrait LoRA at 1 gave a dewy editorial finish including at 2k, and a gouache LoRA at 1 redrew a one-line scene as a painting. On a follow-up on 22 October, a smartphone-snapshot realism LoRA at 1.0 moved a cafe photo to a flatter, cooler look, while a finish-tweaker LoRA at 1.0 stayed close to the control, so check each file against a run at scale 0 on the same seed.

Files trained for MOSS-Image load whether they were trained on Base or SFT, and the first use of a new file takes longer while it downloads. Link the file itself: a model page or a file that needs a login fails, and the request is not charged.

Alternatives

  • No LoRA needed? MOSS-Image is the same model without the LoRA slot, and its review covers the lettering and JSON-layout tests in depth.
  • Another LoRA host? Z-Image Turbo LoRA is quicker and also takes up to three LoRAs.
  • Larger frames and edits with LoRAs? Qwen Image 2512 LoRA edits with one to three source images.
  • A LoRA slot with reference-image edits? FLUX 3 Klein LoRA takes edits with reference images.

How far MOSS-Image LoRA 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
LevelText to image
MildThe lightest prompts in the test
1/1 delivered12.0 / 20 pts
ModerateWhere filtered models start to refuse
1/1 delivered12.0 / 20 pts
StrongMid-scale; must be tested before a model is ranked
1/1 delivered12.0 / 20 pts
Very strongWhere most models start to hold back
1/1 delivered12.0 / 20 pts
MaximumThe top of the scale, scored in two halves: class A and class B
Class A: 1/1 deliveredClass B: Not tested yet6.0 / 20 pts
Task score54.0 / 100.0Tested Oct 21, 2026

Tested Oct 21, 2026 to Oct 21, 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.

See every test record →See the uncensored leaderboard →

Run your own prompt and see how far it goes

Open the Playground

How MOSS-Image LoRA earns its Spicy Index

Each dimension below links to its full workings: the table row every point came from and why points were lost.

Public leaderboards 45% of quality
Not yet tested

No exact match on a tracked public leaderboard yet. How it works

Blind output test 35% of quality
Not yet tested

Not in our blind output test yet. How it works

Arena quality 20% of quality
Not yet tested

Counts once the model has 100+ blind votes and a 95% interval under 200 Elo. How it works

Capability 10.5% weight
60.5
Freedom Score 30% weight
54.0

No quality source has data yet. Capability forms the base and is combined with available freedom; the Index is preliminary and still ranked.

Open the full test report →LoRA support55.0 / 100

Capability, item by item

Points from the public scoring table. Upscaled tiers and limits we set ourselves don’t count.

Native max output4 MP
30.0 / 30
TasksText to image
10.0 / 20
Input images per requestNone
0.0 / 20
Shape control15 shapes
10.0 / 10
Images per request1 image per request
0.0 / 5
Output format choice3 formats
5.0 / 5
Customization (LoRA)LoRA supported
5.5 / 10

What MOSS-Image LoRA can do on SpicyAPI

Straight from the live model specification on SpicyAPI. Some creators offer more in their own apps; this is what you can call through our API today.

Capabilities by endpoint
FeatureText to image
Resolutions1k · 1.5k · 2k
Aspect ratios1:1 · 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
Output formatsjpeg · png · webp
Seed controlYes
Negative promptNo
See every parameter and code sample →

How long it really takes

Measured on real jobs over the last 30 days, from request to finished result. Times include our own test runs.

Median time
29 s
90th percentile
61 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.

Models to consider instead of MOSS-Image LoRA

What do you think?

Was this review useful?

Log in to vote.

Comments

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.

MOSS-Image LoRA FAQ

Try MOSS-Image LoRA yourself

Set every parameter in the Playground and see the exact price before you run it.

$0.025 per image

More reviews

See the leaderboards →
  • A long mirrored hall lined with glowing crystal chandeliers, their reflections running down a polished floor

    Seedream 5.0 Pro

    Pick Seedream 5.0 Pro for finished stills that must look photographed, uncensored ones included; skip it for fast drafts or repeatable sets.

    Spicy Index
    75.6
    Freedom Score
    76.9

    Tested Sep 24, 2026

    Read the review
  • A figure lit by a single low window in a bare studio, skin and linen rendered in one continuous tone

    Qwen Image 2.1

    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.

    Spicy Index
    78.0
    Freedom Score
    96.3

    Tested Sep 24, 2026

    Read the review
  • Two people face to face between a violet beam from the left and a red beam from the right, almost touching, deep black between them

    Z-Image Spicy Pro

    Pick Z-Image Spicy Pro for large uncensored stills up to 2,560 pixels a side with no prompt-length cap; pick standard Spicy when a smaller frame is enough.

    Spicy Index
    78.6
    Freedom Score
    100.0

    Tested Sep 25, 2026

    Read the review