Run models

The best NSFW AI image editors, tested on real edits

By Elowen Thorne, ai video editor and creative consultant · Published by the SpicyAPI team · Updated · Test data as of · How we test

For NSFW image editing, Qwen Image Edit Spicy and Qwen Image 2.1 Pro delivered every edit run we sent, and Seedream 5.0 Pro posted the best blind-test score among editors that delivered most of their runs. Editing has its own failure mode: a model can accept the request and still quietly change the face or pose you wanted to keep, so the notes below say where each one is strong.

Quick picks

#ModelBest forRuns deliveredBlind testSpicy IndexFromLink
1Qwen Image 2.1 ProBest overall editor42/4272.482$0.036/imageRun
2Qwen Image 2.1 LoRABest for edits with a LoRA42/4270.881.9$0.03/image at 1kRun
3Qwen Image 2.1 TurboBest for rapid revisions43/4471.481.1$0.014/imageRun
4Qwen Image 3.0 ProBest quality from Qwen35/3576.578.8$0.04/image at 1kRun
5Seedream 5.0 LiteBest ByteDance editor for range34/3475.675.9$0.035/imageRun
6Qwen Image Edit SpicyEdit-only, no refusals23/2374.675.8$0.038/imageRun
7Seedream 5.0 ProBest blind-test score37/3878.875.6$0.036/image at 1kRun
8FLUX.2 DevCheapest capable editor37/506665.6$0.012/image at 1kRun

Runs delivered counts test runs returned as requested across all five intensity levels; softened and refused runs are not counted. “From” is the lowest listed price on SpicyAPI, checked 2026-11-05; higher resolutions cost more. Sample sizes are small for some models — see how we test.

How we picked

Every model with an image-to-image mode and a completed range test was eligible: 51 of them. Editors that also generate from text count here for their editing behaviour only. The text-to-image ranking covers generation.

The ranking

Qwen Image 2.1 Pro cover1

Best overall editor

Qwen Image 2.1 Pro by Alibaba

  • Highest Spicy Index here
  • No refusals in testing

All 42 runs delivered, and the same model generates from scratch, so a draft and its corrections share one style.

Spicy Index
82
Runs delivered
42/42
Blind test
72.4
From
$0.036/image
  1. Mild 6/6 all delivered
  2. Moderate 7/7 all delivered
  3. Strong 9/9 all delivered
  4. Very strong 7/7 all delivered
  5. Maximum 13/13 all delivered
Inputs
Image edit, Text to image
Resolution
1k, 1.5k, 2k
Input images
Several per job
Test record
42 runs, last tested 2026-10-14
Standard $0.036/image

10% below Alibaba’s own list price at Standard (checked 2026-10-14).

Strong at

  • Combines several source images in a single edit
  • Edits come out at up to 2K

Watch out for

  • Cannot apply a LoRA while editing

Delivered every run at all five levels. Inputs: Image edit, Text to image.

Try Qwen Image 2.1 Pro on SpicyAPI Full test record

Qwen Image 2.1 LoRA sample output
Public sample from the Qwen Image 2.1 LoRA page on SpicyAPI. It shows the model’s look; it is not one of the test runs counted here.
2

Best for edits with a LoRA

Qwen Image 2.1 LoRA by Alibaba

  • No refusals in testing
  • Loads your LoRA

Applies up to three LoRA files while editing, useful when a trained character must survive the change. All 42 runs delivered.

Spicy Index
81.9
Runs delivered
42/42
Blind test
70.8
From
$0.03/image at 1k
  1. Mild 6/6 all delivered
  2. Moderate 6/6 all delivered
  3. Strong 12/12 all delivered
  4. Very strong 6/6 all delivered
  5. Maximum 12/12 all delivered
Inputs
Image edit, Text to image
Resolution
1k, 1.5k, 2k
LoRA
Up to three files by link
Input images
Several per job
Test record
42 runs, last tested 2026-10-05
1k $0.03/image1.5k $0.054/image2k $0.10/image

Strong at

  • Applies your LoRA during the edit
  • Edits from $0.03 at 1K

Watch out for

  • A 2K edit costs over three times a 1K edit
  • The LoRA must be trained for Qwen Image

Delivered every run at all five levels. Inputs: Image edit, Text to image.

Try Qwen Image 2.1 LoRA on SpicyAPI Full test record

Qwen Image 2.1 Turbo sample output
Public sample from the Qwen Image 2.1 Turbo page on SpicyAPI. It shows the model’s look; it is not one of the test runs counted here.
3

Best for rapid revisions

Qwen Image 2.1 Turbo by Alibaba

Delivered 43 of 44 runs at less than half the per-image price of Pro, which makes it the editor to iterate with.

Spicy Index
81.1
Runs delivered
43/44
Blind test
71.4
From
$0.014/image
  1. Mild 6/6 all delivered
  2. Moderate 8/8 all delivered
  3. Strong 10/10 all delivered
  4. Very strong 7/8 some held back
  5. Maximum 12/12 all delivered
Inputs
Image edit, Text to image
Resolution
1k, 1.5k, 2k
Input images
Several per job
Test record
44 runs, last tested 2026-10-13
Standard $0.014/image

10% below Alibaba’s own list price at Standard (checked 2026-10-14).

Strong at

  • $0.0144 per edit
  • The same input options as the Pro tier

Watch out for

  • Softened one of its 44 runs

Delivered every run except Very strong (7 of 8). Inputs: Image edit, Text to image.

Try Qwen Image 2.1 Turbo on SpicyAPI Full test record

Qwen Image 3.0 Pro sample output
Public sample from the Qwen Image 3.0 Pro page on SpicyAPI. It shows the model’s look; it is not one of the test runs counted here.
4

Best quality from Qwen

Qwen Image 3.0 Pro by Alibaba

A higher blind-test score than 2.1 Pro and all 35 runs delivered. Use it for the final pass.

Spicy Index
78.8
Runs delivered
35/35
Blind test
76.5
From
$0.04/image at 1k
  1. Mild 6/6 all delivered
  2. Moderate 6/6 all delivered
  3. Strong 5/5 all delivered
  4. Very strong 6/6 all delivered
  5. Maximum 12/12 all delivered
Inputs
Image edit, Text to image
Resolution
1k, 2k
Input images
Several per job
Test record
35 runs, last tested 2026-09-24
1k $0.04/image2k $0.075/image

Strong at

  • The strongest Qwen editor on blind-test quality
  • Delivered all 35 runs

Watch out for

  • Fewer aspect ratios than the 2.1 editors

Delivered every run at all five levels. Inputs: Image edit, Text to image.

Try Qwen Image 3.0 Pro on SpicyAPI Full test record

Seedream 5.0 Lite sample output
Public sample from the Seedream 5.0 Lite page on SpicyAPI. It shows the model’s look; it is not one of the test runs counted here.
5

Best ByteDance editor for range

Seedream 5.0 Lite by ByteDance

All 34 runs delivered, unlike the Pro and Flash tiers of the same family.

Spicy Index
75.9
Runs delivered
34/34
Blind test
75.6
From
$0.035/image
  1. Mild 6/6 all delivered
  2. Moderate 6/6 all delivered
  3. Strong 4/4 all delivered
  4. Very strong 6/6 all delivered
  5. Maximum 12/12 all delivered
Inputs
Image edit, Text to image
Input images
Several per job
Test record
34 runs, last tested 2026-09-24
Standard $0.035/image

Strong at

  • Delivered every edit run
  • Several source images per edit

Watch out for

  • No layer output

Delivered every run at all five levels. Inputs: Image edit, Text to image.

Try Seedream 5.0 Lite on SpicyAPI Full test record

Qwen Image Edit Spicy cover6

Edit-only, no refusals

Qwen Image Edit Spicy by Alibaba

  • No refusals in testing
  • Small sample: 3 runs or fewer at some levels

A dedicated editor with no text-to-image mode. All 23 runs delivered at every level.

Spicy Index
75.8
Runs delivered
23/23
Blind test
74.6
From
$0.038/image
  1. Mild 7/7 all delivered
  2. Moderate 3/3 all delivered
  3. Strong 3/3 all delivered
  4. Very strong 3/3 all delivered
  5. Maximum 7/7 all delivered
Inputs
Image edit
Test record
23 runs, last tested 2026-10-05
Standard $0.038/image

Strong at

  • All 23 runs delivered at every level

Watch out for

  • Accepts one source image per edit
  • Cannot generate from text

Delivered every run at all five levels. Inputs: Image edit.

Try Qwen Image Edit Spicy on SpicyAPI Full test record

Seedream 5.0 Pro cover: a mirrored hotel hall where chandeliers reflect into infinity down opposing mirrored walls over polished marble7

Best blind-test score

Seedream 5.0 Pro by ByteDance

  • Best blind-test score here

The highest blind-test score of the eight picks. It delivered 37 of 38 runs, softening one.

Spicy Index
75.6
Runs delivered
37/38
Blind test
78.8
From
$0.036/image at 1k
  1. Mild 7/7 all delivered
  2. Moderate 7/7 all delivered
  3. Strong 5/5 all delivered
  4. Very strong 5/6 some held back
  5. Maximum 13/13 all delivered
Inputs
Image edit, Layer split, Text to image
Resolution
1k, 1.5k, 2k
Input images
Several per job
Test record
38 runs, last tested 2026-09-30
1k $0.036/image1.5k $0.04/image2k $0.072/imageauto $0.084/image

20% below ByteDance’s own list price at 1k (checked 2026-09-15).

Strong at

  • Splits an image into separate layers
  • 20% below ByteDance’s list price at 1K

Watch out for

  • 2K edits cost twice as much as 1K

Delivered every run except Very strong (5 of 6). Inputs: Image edit, Layer split, Text to image.

Try Seedream 5.0 Pro on SpicyAPI Full test record

FLUX.2 Dev sample output
Public sample from the FLUX.2 Dev page on SpicyAPI. It shows the model’s look; it is not one of the test runs counted here.
8

Cheapest capable editor

FLUX.2 Dev by Black Forest Labs

The lowest per-image price here. 37 of 50 runs delivered and the rest softened rather than refused.

Spicy Index
65.6
Runs delivered
37/50
Blind test
66
From
$0.012/image at 1k
  1. Mild 8/8 all delivered
  2. Moderate 4/8 some held back
  3. Strong 9/10 some held back
  4. Very strong 0/8 blocked
  5. Maximum 16/16 all delivered
Inputs
Image edit, Text to image
Resolution
1k, 2k
Input images
Several per job
Test record
50 runs, last tested 2026-10-09
1k $0.012/image2k $0.012/imageStandard $0.023/image

Strong at

  • The lowest price per edit on this page

Watch out for

  • Softened 13 of its 50 runs

Delivered every run except Moderate (4 of 8) and Strong (9 of 10) and Very strong (0 of 8). Inputs: Image edit, Text to image.

Try FLUX.2 Dev on SpicyAPI Full test record

What makes an edit succeed

An edit has two jobs: change what you asked for and leave everything else alone. Our range test measures the first job, whether the model carries out the request. The blind-test score covers prompt adherence and visual quality, which is the closest proxy we have for the second job. A model that scores well on range but lower on the blind test will usually do the edit and add some drift elsewhere.

Upload images you created or have the rights to use. Every model here refuses photos of real, identifiable people, and that refusal is not a bug to work around.

Well-known models that held back

GPT Image and Nano Banana are the editors most people try first. On this test set they delivered only a small share of edit runs, as did the Max and Pro tiers of FLUX.2.

  • GPT Image 2 by OpenAI

    2 of 12 runs delivered. 2 of 12 tested runs delivered; Maximum not tested yet. Image edit, Text to image.

    1. 1/4
    2. 1/2
    3. 0/5
    4. 0/1
    5. —
  • Nano Banana 2.1 by Google

    9 of 52 runs delivered. Held back at every level, from Mild up. Image edit, Text to image.

    1. 6/10
    2. 0/8
    3. 0/14
    4. 0/8
    5. 3/12
  • FLUX.2 Max by Black Forest Labs

    6 of 50 runs delivered. Held back at every level, from Mild up. Image edit, Text to image.

    1. 4/8
    2. 0/8
    3. 0/10
    4. 0/8
    5. 2/16
  • FLUX.2 Pro by Black Forest Labs

    12 of 50 runs delivered. Held back at every level, from Mild up. Image edit, Text to image.

    1. 4/8
    2. 2/8
    3. 1/10
    4. 0/8
    5. 5/16

Every scored model is in the filter tracker.

Which one for your job

If you need…PickWhy
Edit and regenerate with one modelQwen Image 2.1 ProBoth modes, all runs delivered.
Keep a trained character through editsQwen Image 2.1 LoRALoRA loading during edits.
Many quick variationsQwen Image 2.1 TurboLow price per edit, 43 of 44 delivered.
The cleanest final imageSeedream 5.0 ProTop blind-test score among the eight picks.
The lowest price per editFLUX.2 DevCheapest listed rate of the eight.

All 51 image editing models we tested

The full field behind this ranking, ordered by Spicy Index. Models outside the eight picks are here so you can check any name you have heard of.

#ModelMild → Maximumdelivered / runsBlind testSpicy Index100 edits fromLink
1Qwen Image 2.1 ProAlibaba
  1. 6/6
  2. 7/7
  3. 9/9
  4. 7/7
  5. 13/13
72.482$3.60 / 100 imagesRun
2Qwen Image 2.1 LoRAAlibaba
  1. 6/6
  2. 6/6
  3. 12/12
  4. 6/6
  5. 12/12
70.881.9$3.00 / 100 imagesRun
3Qwen Image 2.1 TurboAlibaba
  1. 6/6
  2. 8/8
  3. 10/10
  4. 7/8
  5. 12/12
71.481.1$1.44 / 100 imagesRun
4Qwen Image 3.0 ProAlibaba
  1. 6/6
  2. 6/6
  3. 5/5
  4. 6/6
  5. 12/12
76.578.8$4.00 / 100 imagesRun
5Qwen Image 2.1Alibaba
  1. 6/6
  2. 6/6
  3. 5/5
  4. 5/6
  5. 12/12
72.478$2.40 / 100 imagesRun
6Qwen Image 3.0Alibaba
  1. 6/6
  2. 6/6
  3. 4/4
  4. 6/6
  5. 12/12
75.377.7$3.00 / 100 imagesRun
7MiniMax H3 Image LoRAMiniMax
  1. 6/6
  2. 6/6
  3. 6/6
  4. 8/8
  5. 12/12
63.476.8$4.20 / 100 imagesRun
8Seedream 5.0 LiteByteDance
  1. 6/6
  2. 6/6
  3. 4/4
  4. 6/6
  5. 12/12
75.675.9$3.45 / 100 imagesRun
9Qwen Image 2512 LoRAAlibaba
  1. 6/6
  2. 6/6
  3. 6/6
  4. 2/8
  5. 12/12
71.975.8$3.00 / 100 imagesRun
10Qwen Image Edit SpicyAlibaba
  1. 7/7
  2. 3/3
  3. 3/3
  4. 3/3
  5. 7/7
74.675.8$3.80 / 100 imagesRun
11Seedream 5.0 ProByteDance
  1. 7/7
  2. 7/7
  3. 5/5
  4. 5/6
  5. 13/13
78.875.6$3.60 / 100 imagesRun
12Seedream 5.0 FlashByteDance
  1. 3/3
  2. 6/6
  3. 4/5
  4. 1/3
  5. 7/9
77.873.9$2.60 / 100 imagesRun
13P-Image LoRAPruna AI
  1. 2/2
  2. 2/2
  3. 2/2
  4. —
  5. 2/2
—71.3$0.60 / 100 imagesRun
14FLUX 3 Klein LoRABlack Forest Labs
  1. 2/2
  2. 2/2
  3. 2/2
  4. 0/2
  5. 2/2
—69.3$1.80 / 100 imagesRun
15Wan 2.7Alibaba
  1. 6/6
  2. 6/6
  3. 6/7
  4. 1/6
  5. 7/12
73.669.3$3.00 / 100 imagesRun
16Wan 2.7 ProAlibaba
  1. 6/6
  2. 6/6
  3. 5/7
  4. 0/6
  5. 8/12
74.668.3$7.50 / 100 imagesRun
17FLUX.2 Klein 9B LoRABlack Forest Labs
  1. 2/2
  2. 2/2
  3. 2/2
  4. 2/2
  5. 2/2
—67.6$1.80 / 100 imagesRun
18FLUX.2 Klein Base 9B LoRABlack Forest Labs
  1. 2/2
  2. 2/2
  3. 2/2
  4. 2/2
  5. 2/2
—67.6$2.40 / 100 imagesRun
19Z-Image Base LoRAAlibaba
  1. 8/8
  2. 7/8
  3. 13/14
  4. 0/10
  5. 14/14
55.567.5$1.44 / 100 imagesRun
20Qwen Image 2Alibaba
  1. 6/6
  2. 6/6
  3. 6/6
  4. 4/6
  5. 12/12
72.667.2$3.50 / 100 imagesRun
21FLUX 3 KleinBlack Forest Labs
  1. 2/2
  2. 2/2
  3. 2/2
  4. 0/2
  5. 2/2
72.666.5$1.20 / 100 imagesRun
22Seedream 4.0ByteDance
  1. 6/6
  2. 6/6
  3. 8/8
  4. 4/6
  5. —
72.566.5$3.00 / 100 imagesRun
23FLUX.2 DevBlack Forest Labs
  1. 8/8
  2. 4/8
  3. 9/10
  4. 0/8
  5. 16/16
6665.6$1.16 / 100 imagesRun
24Z-Image Turbo LoRAAlibaba
  1. 6/6
  2. 6/6
  3. 7/7
  4. 4/7
  5. 12/12
48.465.3$1.20 / 100 imagesRun
25FLUX.2 Klein 9BBlack Forest Labs
  1. 8/8
  2. 8/8
  3. 4/10
  4. 0/8
  5. 16/16
67.765$1.10 / 100 imagesRun
26FLUX.1 Krea Dev LoRABlack Forest Labs
  1. 2/2
  2. 2/2
  3. 2/2
  4. —
  5. 2/2
—64.8$2.40 / 100 imagesRun
27FLUX.1 Schnell LoRABlack Forest Labs
  1. 2/2
  2. 2/2
  3. 2/2
  4. —
  5. 2/2
—64.8$0.60 / 100 imagesRun
28Krea 2 Turbo LoRAKrea
  1. 2/2
  2. 2/2
  3. 2/2
  4. 2/2
  5. 2/2
—64.8$2.16 / 100 imagesRun
29Wan 2.1 Image LoRAAlibaba
  1. 2/2
  2. 2/2
  3. 2/2
  4. —
  5. 2/2
—64.8$3.00 / 100 imagesRun
30Anima Turbo LoRACircleStone Labs
  1. 6/6
  2. 6/6
  3. 9/9
  4. 11/11
  5. 9/12
46.864.6$1.80 / 100 imagesRun
31Nano Banana 2.1Google
  1. 6/10
  2. 0/8
  3. 0/14
  4. 0/8
  5. 3/12
75.664.1$2.17 / 100 imagesRun
32Seedream 4.5ByteDance
  1. 2/2
  2. 2/2
  3. 1/2
  4. 0/2
  5. 2/2
70.964.1$4.00 / 100 imagesRun
33FLUX.2 Dev LoRABlack Forest Labs
  1. 2/2
  2. 2/2
  3. 2/2
  4. —
  5. 2/2
—64$2.16 / 100 imagesRun
34FLUX.2 Klein 4B LoRABlack Forest Labs
  1. 2/2
  2. 2/2
  3. 2/2
  4. —
  5. 2/2
—64$1.56 / 100 imagesRun
35Anima TurboCircleStone Labs
  1. 7/7
  2. 7/7
  3. 11/11
  4. 11/11
  5. 9/12
44.762.6$1.20 / 100 imagesRun
36FLUX.2 Klein Base 4B LoRABlack Forest Labs
  1. 2/2
  2. 1/1
  3. 2/2
  4. —
  5. 2/2
—62.2$1.80 / 100 imagesRun
37GPT Image 2.5 SunburstOpenAI
  1. 3/10
  2. 0/1
  3. 0/5
  4. 0/3
  5. 0/1
79.562.1$0.18 / 100 imagesRun
38GPT Image 2OpenAI
  1. 1/4
  2. 1/2
  3. 0/5
  4. 0/1
  5. —
81.261$0.18 / 100 imagesRun
39GPT Image 2.5 FlareOpenAI
  1. 1/5
  2. 0/1
  3. 0/3
  4. —
  5. —
81.160.8$0.18 / 100 imagesRun
40FLUX.2 Klein 4BBlack Forest Labs
  1. 8/8
  2. 5/8
  3. 3/10
  4. 0/8
  5. 15/16
67.259.6$0.93 / 100 imagesRun
41Nano Banana ProGoogle
  1. 5/6
  2. 0/1
  3. 0/1
  4. 0/1
  5. 0/1
74.457.9$6.68 / 100 imagesRun
42Nano Banana 2Google
  1. 3/4
  2. 1/1
  3. 0/1
  4. —
  5. —
70.357.6$3.60 / 100 imagesRun
43FLUX.1 Dev LoRABlack Forest Labs
  1. 3/3
  2. 3/3
  3. 1/4
  4. 0/6
  5. 3/6
59.753.4$1.80 / 100 imagesRun
44FLUX.2 ProBlack Forest Labs
  1. 4/8
  2. 2/8
  3. 1/10
  4. 0/8
  5. 5/16
69.452.3$3.00 / 100 imagesRun
45FLUX 3 Klein BaseBlack Forest Labs
  1. —
  2. —
  3. 0/2
  4. 0/2
  5. 0/1
67.952.1$3.60 / 100 imagesRun
46FLUX.2 FlexBlack Forest Labs
  1. 2/8
  2. 0/8
  3. 1/10
  4. 0/8
  5. 5/16
72.151.8$5.78 / 100 imagesRun
47FLUX.2 MaxBlack Forest Labs
  1. 4/8
  2. 0/8
  3. 0/10
  4. 0/8
  5. 2/16
71.851$7.00 / 100 imagesRun
48FLUX.1 Fill Dev LoRABlack Forest Labs
  1. 1/1
  2. 1/1
  3. 1/1
  4. —
  5. 1/1
—50.8$4.20 / 100 imagesRun
49Krea 3Krea
  1. —
  2. 0/2
  3. 0/1
  4. 0/1
  5. —
72.448.2$3.60 / 100 imagesRun
50Qwen Image 1.0 LoRAAlibaba
  1. 2/2
  2. 2/2
  3. 2/2
  4. —
  5. 2/2
—43.8$3.00 / 100 imagesRun
51Qwen Image Edit 2509 LoRAAlibaba
  1. 1/1
  2. 1/1
  3. 1/1
  4. —
  5. 1/1
—39.5$3.00 / 100 imagesRun

Questions

Which AI image editor allows NSFW edits?

In our tests Qwen Image 2.1 Pro, Qwen Image 2.1 LoRA, Qwen Image 3.0 Pro, Seedream 5.0 Lite and Qwen Image Edit Spicy delivered every NSFW edit run.

Why does an editor change the face?

Editing models regenerate the whole image, not only the area you mention. Models with higher blind-test scores keep more of the original; a LoRA build keeps a trained character more reliably.

Can I edit a photo of a real person?

No. Images of real, identifiable people are refused by every model here and are outside our tests.

Open the model picker →