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.

Qwen Image 3.0 Pro Review: Legible Type, Uncensored, Tested

Pick Qwen Image 3.0 Pro for legible on-image type or a multi-reference edit, uncensored; skip it if a lower price per draft matters more than lettering.

Create in StudioOpen the Playground

Spicy Index

78.8/ 100

Freedom Score

98.0/ 100

Range if the rest pass: 98.0–100.0

Content levels

Tested 6 of 6 probe types · 35 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
$0.04 per image
Median time
66 s
Success rate
Collecting data

Qwen Image 3.0 Pro passed every rung of our content ladder we ran on both endpoints on 24 Sep 2026 — L1 Suggestive (lingerie, a slipping strap), L2 Mature themes, L3 Strong, L4 Intense (an intimate scene between two adults), and both L5 Extreme probe types, L5a (consensual restraint and impact play between adults) and L5b (cinematic gore) — and it puts sharp, correctly spelled type straight into the frame. A 1K image is $0.04.

What Qwen Image 3.0 Pro 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

    A glossy fashion magazine cover, full-bleed portrait of an adult woman in a black lace bodysuit under a sheer open kimono, arms raised adjusting her hair, studio rim light on a deep charcoal background; masthead text 'SPICY' in large condensed type across the top, cover lines reading 'THE LATE SHIFT' and 'ISSUE 07' in small caps, high-end print typography, razor-sharp 2K detail.

    Try this model

  • Text to image

    Neon-drenched night photograph outside a roadside motel. An adult woman in a translucent pink rain poncho worn over a black bikini stands in the glow of a humming vending machine, holding a cold can against her cheek and looking sideways at the camera. Behind her a tall motel sign reads 'VACANCY' in pink neon script with a smaller board underneath reading 'POOL • ICE • CABLE TV'. Wet asphalt reflections, magenta and cyan, moths in the light, razor-sharp signage typography, 2K detail.

    Try this model

  • Text to image

    Cinematic fight poster, underground arena. Two gladiators stand forehead to forehead in a stare-down, oiled torsos catching torchlight, breath visible, sand and dust in the air. Three lines of clean typography set across the lower third: the words MARCUS VS DRAVEN on one line, XIV OCTOBER on the next, ARENA VETUS on the last. 65mm, shallow depth of field, heavy grain, torch orange against black, restrained and tasteful.

    Try this model

  • Text to image

    First-person point of view photograph looking down the bed on a grey morning, soft window light, shallow depth of field. In the foreground on a rumpled white duvet lie a woman's smooth bare crossed ankles in black over-the-knee socks, slim feminine calves, and a black lace slip draped over the corner of the bed. On the nightstand just beyond, a handwritten note is propped against a coffee cup, the handwriting large and perfectly legible in blue ink: "GONE FOR COFFEE — DON'T MOVE. BACK IN TEN. — J". Beside it a brass key and a paperback face-down. Muted grey-green palette, real 35mm grain, razor-sharp handwriting, no logos, no brand names.

    Try this model

  • Image editing

    Keep the face, hair and pose from Image 1. Dress her in a deep indigo silk kimono worn loose off both shoulders over a matching camisole, and place her kneeling beside the low table in the machiya room from Image 2. Relight her with the warm paper-lantern glow from the screens behind, keeping the green-cyan shadow cast of Image 2 so both read as one photograph.

    Try this model

  • Image editing

    Keep the watch, the ice block, the stone and the lighting exactly as they are. Turn the frame into a print advertisement: add the headline 'Hold the Moment' in a refined serif across the empty space above, and a small caps line 'Water resistant to 300 metres' along the bottom edge. Change nothing else.

    Try this model

Make your own with Qwen Image 3.0 Pro

Create in StudioOpen the Playground

The verdict

We ran 35 test requests across text-to-image and editing on 23 and 24 September: L1 Suggestive, L2 Mature themes, L3 Strong, L4 Intense, and both L5 Extreme probe types (restraint/impact and gore). Every one came back exactly as asked, nothing softened and nothing refused. The same requests also show what the model is built for: a magazine masthead, a perfume campaign's single line of type, a private eye's gold-leaf office door — all legible, all spelled correctly, without a separate lettering pass.

Pick it when the shot needs to carry real words: a cover line, a sign, a tagline. Pick it too when one edit has to pull from up to three source images at once. Skip it if a lower price per draft is what matters most; the base Qwen Image 3.0 tier and Qwen Image 2.1 both cost less per image.

How it scores. Most of the Spicy Index is output quality — public blind leaderboards such as arena.ai and Artificial Analysis, plus our own blind output test — with specs contributing 15% of the quality-and-capability base before the Freedom Score is included; the live figure, its current standing on our image leaderboard and the full breakdown are on the score card above and in the test report. Freedom Score 98.0 of 100, covering all 6 probe types across those 35 runs on 24 September. Median response time is about 73 seconds.

Try text-to-image → · Try the editor →

Key facts

  • Freedom Score: 98.0/100 on this page's score card, computed 26 September from 35 runs across both endpoints on 24 September; every rung passed.
  • Pass rate: every run we scored came back as asked, at every rung from L1 Suggestive to both L5 Extreme probe types.
  • Spicy Index: see the score card above, and its live standing on our image leaderboard.
  • Price: $0.04 per image at 1K, $0.075 at 2K (the default), on both text-to-image and editing.
  • Limits: prompts up to 5,000 characters, up to three reference images per edit (10 MB each), one image per request, eight aspect ratios, median response about 73 seconds.
  • Sources: every scored run is in the Qwen Image 3.0 Pro test report, the scoring rules are on our review methodology page, and the full parameter list is on the Qwen Image 3.0 Pro model page.

What we liked and what held it back

What we liked

  • It spells things correctly. Mastheads, cover lines, motel signage, a handwritten note — every string we tested came back legible and correctly spelled, on the first try, without a retouch pass.
  • Uncensored on both endpoints. All 6 probe types passed on text-to-image and on editing on 24 September: nothing softened, nothing refused.
  • Three references in one edit. The wardrobe-and-set example pulls a face and pose from one photo and a room's lighting from a second, and keeps both consistent in the result.
  • Cheap relative to the frontier. Community pricing checks put it well under the leading closed model per image at comparable settings; see What the community says.
  • Eight aspect ratios, including 21:9, cover most social and print formats without a crop step afterward.
  • A seed field, so a composition worth keeping can be re-run with small prompt edits instead of re-rolling blind.

What held it back

  • One image per request. num_outputs is fixed at 1; batching means separate calls.
  • 2K tops out modestly. The 2K tier is a sharp frame but not a large one; models built for big prints go further.
  • Only three references on the editor. Seedream 5.0 Lite takes ten; if a composite needs more source photos, that ceiling arrives first.
  • Community testing puts dense UI and infographic layouts behind the leading closed model, even though photographic frames and lettering hold up well. See What the community says.
  • 2K is the billed default. Leave resolution unset and a "1K job" comes back at 2K pricing; set resolution: 1k explicitly to draft cheap.

What the images look like

The gallery holds published, safe-for-work examples from the Qwen Image 3.0 Pro model page, shown as rendered. Our ladder test images aren't shown; we report the results only.

A cover that reads as a cover. The magazine-cover example sets a masthead, two cover lines and small-caps type across a studio portrait, all sharp enough to read at thumbnail size.

A sign in the rain. The motel-vacancy frame puts "VACANCY" in pink neon script and a second board underneath, both legible through wet asphalt reflections and neon glare.

One line changed, everything else held. The cover-line-swap edit keeps a portrait's pose, wardrobe, lighting and grade untouched and only swaps the cover line's text — the kind of single, surgical edit a magazine layout actually needs.

Two photos into one room. The wardrobe-and-set edit takes a face from one reference and a room's paper-lantern lighting from a second, producing one photograph where the shadow cast and glow match across both source images.

Prompts that worked

Qwen Image 3.0 Pro takes literal instructions for wardrobe, pose, light source and on-image text. Say exactly what the type should read and where it sits, and the editor takes "keep X, change only Y" instructions well.

1. A cover with working type (text-to-image, 2K). The prompt behind the magazine-cover example.

text
A glossy fashion magazine cover, full-bleed portrait of an adult woman in a black lace bodysuit under a sheer open kimono, arms raised adjusting her hair, studio rim light on a deep charcoal background; masthead text 'SPICY' in large condensed type across the top, cover lines reading 'THE LATE SHIFT' and 'ISSUE 07' in small caps, high-end print typography, razor-sharp 2K detail.

Run this prompt on Qwen Image 3.0 Pro →

2. Change one line, keep everything else (edit). Upload the cover, then paste this.

text
Keep the portrait, wardrobe, lighting and grade exactly as they are. Change the cover line from 'THE LATE SHIFT' to 'AFTER HOURS' in the same small caps and the same position, and leave the masthead untouched.

Run this edit on Qwen Image 3.0 Pro →

3. Two references, one room (edit). The prompt behind the wardrobe-and-set composite. Upload the face photo first, the room second.

text
Keep the face, hair and pose from the first image. Dress her in a deep indigo silk kimono worn loose off both shoulders over a matching camisole, and place her kneeling beside the low table in the second image's machiya room. Relight her with the warm paper-lantern glow from the screens behind, keeping the second image's green-cyan shadow cast so both read as one photograph.

Run this edit on Qwen Image 3.0 Pro →

Weak: "Woman on a magazine cover." No masthead text, no cover lines — Qwen Image 3.0 Pro fills in generic type or none at all. Strong: prompt 1. Name every string you want set, in quotes, and it holds.

How far it goes

Qwen Image 3.0 Pro passed every rung we ran on both endpoints. On 23 and 24 September we ran L1 Suggestive, L2 Mature themes, L3 Strong, L4 Intense and both L5 Extreme probe types on text-to-image and on editing, 35 runs in all: every one rendered as asked, none softened, none refused. The page label is Uncensored: we add no content filter on our side.

Freedom Score: 98.0 of 100, shown on this page's score card and recalculated as we add runs; all 6 probe types passed across those 35 runs.

Not tested yet: three-reference edits past our two-reference composite; scenes with two women or two men above L1 Suggestive; batches larger than one image per call.

Adults only, in every image. A real, identifiable person needs to have consented, and you need the rights to every photo you edit.

What the community says

Qwen Image 3.0 launched 21 July 2026 and its Hacker News thread turned into a running comparison against the leading closed image model. On price, one commenter did the math: cost is significantly lower in this case. A Pro image here is $0.04, a gpt-2-image high is $0.21 and lower resolution.

Quality opinion split by task. On raw output, one reader was blunt: Looks like this model is meaningfully less good than gpt-image-2. Arena.ai score is 1263 vs 1380. A developer who ran side-by-side design tests landed in between: It's a big step up over their prior model, but it's nowhere near gpt-image-2 at least for high density UI design. That matches what we saw: photographic frames and on-image lettering hold up well in our testing, and the community's complaint is specifically about dense infographic and interface layouts, a different job than most of our examples.

Alternatives

How far Qwen Image 3.0 Pro 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
LevelImage editingText to image
MildThe lightest prompts in the test
3/3 delivered20.0 / 20 pts
3/3 delivered20.0 / 20 pts
ModerateWhere filtered models start to refuse
3/3 delivered20.0 / 20 pts
3/3 delivered20.0 / 20 pts
StrongMid-scale; must be tested before a model is ranked
3/3 delivered20.0 / 20 pts
2/2 delivered16.0 / 20 pts
Very strongWhere most models start to hold back
3/3 delivered20.0 / 20 pts
3/3 delivered20.0 / 20 pts
MaximumThe top of the scale, scored in two halves: class A and class B
Class A: 3/3 deliveredClass B: 3/3 delivered20.0 / 20 pts
Class A: 3/3 deliveredClass B: 3/3 delivered20.0 / 20 pts
Task score100.0 / 100.0Tested Sep 24, 202696.0 / 100.0Lowest: pulls the average downTested Sep 24, 2026

Tested Sep 23, 2026 to Sep 24, 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

Create in StudioOpen the Playground

How Qwen Image 3.0 Pro 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
66.8
Blind output test 35% of quality
76.5
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
68.0
Freedom Score 30% weight
98.0
Open the full test report →

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
TasksImage editing, Text to image
20.0 / 20
Input images per request3 images
8.0 / 20
Shape control8 shapes
10.0 / 10
Images per request1 image per request
0.0 / 5
Output format choiceFixed format
0.0 / 5
Customization (LoRA)No LoRA support
0.0 / 10

Arena rating: 1039 (95% interval 923–1151); votes are still coming in.

What Qwen Image 3.0 Pro 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
FeatureImage editingText to image
Resolutions1k · 2k1k · 2k
Aspect ratios1:1 · 16:9 · 9:16 · 4:3 · 3:4 · 3:2 · 2:3 · 21:91:1 · 16:9 · 9:16 · 4:3 · 3:4 · 3:2 · 2:3 · 21:9
Reference inputs3—
Seed controlYesYes
Negative promptYesYes
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
66 s
90th percentile
2.9 min
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.

What people say about Qwen Image 3.0 Pro

1 positive · 1 mixed · 1 negative

  • cost is significantly lower in this case. A Pro image here is $0.04, a gpt-2-image high is $0.21 and lower resolution.

    hacker-news· Aug 5, 2026· PositiveSource

  • Looks like this model is meaningfully less good than gpt-image-2. Arena.ai score is 1263 vs 1380.

    hacker-news· Aug 5, 2026· NegativeSource

  • It's a big step up over their prior model, but it's nowhere near gpt-image-2 at least for high density UI design.

    hacker-news· Aug 5, 2026· MixedSource

Quotes are verbatim from public posts, linked to the original. They are not ratings and don’t feed any score.

Models to consider instead of Qwen Image 3.0 Pro

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.

Qwen Image 3.0 Pro FAQ

Try Qwen Image 3.0 Pro yourself

Generate in Studio in a few clicks, or set every parameter in the Playground and see the exact price before you run it.

$0.04 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
  • A pale sculptural ring and a curved red strip

    Seedream 5.0 Flash

    Choose Flash for creation and reference edits; its independent quality evidence remains preliminary.

    Spicy Index
    73.9
    Freedom Score
    62.2

    Tested Sep 30, 2026

    Read the review