Skip to main content
Filtered — This model refuses requests beyond its own limits outright. Measured, not assumed — models we have not measured carry no label at all.

Krea 3 Review: Edits Up to Eight Images, and Where the Filter Starts

Pick Krea 3 for instruction-following photo edits and film-look stills that stay inside a tame content range; look elsewhere if your requests go past Mild.

Open the Playground

Spicy Index

48.2/ 100

Freedom Score

0.0/ 100

Range if the rest pass: 0.0–60.0

Content levels

Tested 3 of 6 probe types · 4 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
$0.036 per image
Median time
26 s
Success rate
Collecting data

Krea 3 is Krea's own image model, built first for editing from plain-language instructions. SpicyAPI serves it as text-to-image and as an edit endpoint that takes one to eight source images. Both produce about one megapixel in ten aspect ratios, with a high-quality and a fast variant at the same price. We tested both endpoints on 16 October 2026 and read every result.

What Krea 3 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 woman in her early thirties with tousled auburn hair reclines in a sunken conversation pit of a 1970s canyon house, wearing a rust-coloured silk halter jumpsuit, one knee raised, a smouldering half-smile at the camera, low afternoon sun through eucalyptus leaves striping the orange shag rug. Shot on 35mm Kodak film, visible grain, warm amber and teal grade, soft halation, 1970s Los Angeles.

    Try this model

  • Text to image

    A man in his late thirties with a short dark beard works at a pottery wheel in a redwood studio, linen shirt unbuttoned to the chest, sleeves rolled, wet clay on his forearms, he glances up with a confident, alluring look, dusty sunbeams, shelves of glazed bowls. Shot on 35mm Kodak film, visible grain, warm amber and teal grade, soft halation, 1970s Los Angeles.

    Try this model

  • Text to image

    A Latina woman in her thirties leans against the tailgate of a faded turquoise 1970s pickup truck on a winding canyon road, high-waisted denim flares and a cropped knit top, wind lifting her dark hair, the city haze glowing amber far below. Shot on 35mm Kodak film, visible grain, warm amber and teal grade, soft halation, 1970s Los Angeles.

    Try this model

  • Image editing

    Dress the woman from image 1 in the emerald paisley wrap dress from image 2 and place her barefoot on a sunlit redwood deck of a 1970s canyon house, one bell sleeve slipping off her shoulder, hair loose in the warm breeze, leaning on the railing with a slow alluring glance over her shoulder, eucalyptus and dry hills behind her. Shot on 35mm Kodak film, visible grain, warm amber and teal grade, soft halation, 1970s Los Angeles.

    Try this model

  • Image editing

    The same man from the photo, keeping his exact face, age (early forties), greying temples and moustache, now lounging in a hanging rattan swing chair on a shaded canyon-house porch at golden hour, cream knit polo with the collar open, aviator sunglasses pushed up into his hair, a confident easy smile, bougainvillea and a hazy valley behind him. Shot on 35mm Kodak film, visible grain, warm amber and teal grade, soft halation, 1970s Los Angeles.

    Try this model

  • Image editing

    Same woman, same framing, now in a 1970s canyon-house den with wood panelling and an amber lamp, wearing a cream satin blouse with a wide collar instead of the blazer. Shot on 35mm film, visible grain, warm amber and teal grade, 1970s Los Angeles.

    Try this model

Make your own with Krea 3

Open the Playground

The verdict

It is strong where editing is the job: put the person from one photo in the clothes from another, move someone into a new place or decade, keep the face and the framing. It is a poor fit if you want an unfiltered model. Krea 3 screens the prompt and the source images before it draws, refused our requests from Moderate up within about two seconds, and did not charge for them. Mild fashion work rendered normally. This is the opposite of the unfiltered families on this site, and the page label says so.

Pick Krea 3 for try-on, restaging and film-photograph stills where the content stays tame. Skip it when you need Moderate or stronger requests to land: FLUX 3 Klein and the other open-weights families refused nothing in our tests.

We also ran our blind output test on Krea 3 on 23 October 2026, with text-to-image and edit prompts scored by three judges. The result is on this page's score card with the Spicy Index and Freedom Score; this text prints no number and makes no ranking claim.

Try Krea 3 →

Key facts

  • Content limits: a filter screens prompts and source images and refuses from Moderate up. In our runs the label on the model page is policy filtered. Refused requests cost nothing.
  • Text-to-image: prompt (up to 5,000 characters), aspect_ratio (10 options), quality (high or fast), enable_thinking, negative_prompt (high quality only), output_format (PNG or JPEG) and seed.
  • Edit: the same, plus image_urls with 1 to 8 source images. Name them in the prompt as image 1, image 2 and so on, in upload order.
  • Size: about one megapixel in every aspect ratio, generated at that size. One image per request.
  • Speed: most of our test images took 7 to 30 seconds.
  • Price: one flat price per image, the same for high quality, fast, thinking and every aspect ratio. An edit adds the same amount again for every source image. Live prices are on the model page.
  • Seeds: a repeated seed gets close, not identical, because the model expands the prompt before drawing.

What we liked and what held it back

What we liked

  • Edits that keep the person. The published edit examples change clothes, place and light while the face, hair and framing stay.
  • Up to eight sources in one edit. A portrait plus a garment photo becomes that person wearing it; the dress swap example uses exactly that.
  • Prompts that name sources by number. In our check, a prompt that named only image 2 drew only that subject, so the numbering is followed.
  • A film-photograph finish from one line. A closing sentence about film stock, grain and colour grade sets the period look of a whole text-to-image prompt.
  • Fast or high quality at the same price. Find the framing on fast, then rerun the keeper on high quality with the same seed.
  • Refusals are free. A request stopped by the filter returns in about two seconds with no charge.

What held it back

  • A filter you cannot switch off. From Moderate up, our requests were refused. Mild fashion renders.
  • The edit boundary is not fixed. One Moderate edit passed once and was refused on a retest with another seed, so do not build a workflow that needs it to land.
  • The edit does not keep the source shape. Pick the aspect ratio yourself; match it to your main photo for the same framing.
  • About one megapixel is the ceiling. There is no 2k tier on this endpoint.
  • The negative prompt works on high quality only. The fast variant ignores it.
  • Every source image adds to the price, and a link that cannot be downloaded fails the request.
  • Seeds do not repeat exactly. Expect close, not identical.

What the images look like

The gallery shows the six published examples from the Krea 3 model page: three text-to-image and three edits. Everything in it is clothed and safe for work. The 1970s Los Angeles look is the set's running theme and a good picture of what Krea 3 does with a style line at the end of the prompt.

Film-photograph stills. A conversation pit in a canyon house, a pottery studio with dusty sunbeams, and a turquoise pickup on a canyon road at 21:9. Grain, halation and warm amber light come from one closing sentence.

A person and a garment, one picture. The woman from image 1 in the paisley wrap dress from image 2, moved to a sunlit canyon deck. Her face, hair and build carry over and the dress keeps its print and sleeves.

Restaging. A man from a casual photo moved onto a porch swing, and an office portrait moved into a wood-panelled den in a satin blouse, on the fast variant. Faces and framing stay.

Prompts that worked

For edits, say what should change first, then close with one sentence on what stays the same. Refer to sources as image 1, image 2 in upload order. For text-to-image, write subject, clothing, place and light in plain sentences, then one line of film stock and grade.

1. Edit, two sources. aspect_ratio 2:3.

text
Dress the woman from image 1 in the emerald paisley wrap dress from image 2 and place her on a sunlit redwood deck of a 1970s canyon house, hair loose in a warm breeze, hills behind her. Keep her face and hair exactly. Shot on 35mm film, visible grain, warm amber and teal grade.

2. Edit, one source, fast variant. aspect_ratio 4:5, quality fast.

text
Same woman, same framing, now in a 1970s den with wood panelling and an amber lamp, wearing a cream satin blouse with a wide collar instead of the blazer. Shot on 35mm film, visible grain.

3. Text-to-image, wide scene. aspect_ratio 21:9.

text
A woman in her thirties leans against the tailgate of a faded turquoise 1970s pickup truck on a winding canyon road, denim flares and a knit top, wind lifting her dark hair, city haze glowing amber far below. Shot on 35mm film, visible grain, warm amber and teal grade, soft halation.

Weak: "make her look better". Strong: the change first, the sources by number, and one sentence on what must not change.

How far it goes

Krea 3 has a content filter in front of the model: it screens prompts and source images before drawing. We tested with fictional adults only, at five intensity levels (Mild, Moderate, Strong, Very strong, Maximum; the test prompts are not published), and report every result as delivered, toned down or refused.

  • Mild: delivered on text-to-image, together with more than 20 clothed fashion probes for size and variant checks, and on the fashion edits.
  • Moderate: refused on text-to-image. On edit, one run was delivered and a retest with another seed was refused.
  • Strong: refused on edit.
  • Very strong: refused on text-to-image.
  • Maximum: not tested on this model.

We ran the stronger levels once or twice each, so read them as "the filter is on and it is not fully consistent", not as exact rates. Refusals came back in about two seconds and were not charged. If you need an image model that delivers these levels, use an unfiltered family; the alternatives below list some.

Adults only, in every prompt and every source photo. Never upload a real person's likeness without consent, and never depict anyone who could read as a minor.

Alternatives

  • Unfiltered edits? FLUX 3 Klein takes up to ten reference images and refused nothing in our tests.
  • The previous Krea? Krea 2 Large is the text-to-image generation before Krea 3.
  • Another edit-first model? Nano Banana 2.1 and Seedream 4.5 both edit from instructions.

How far Krea 3 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 testNot tested yetNot tested yet
ModerateWhere filtered models start to refuse
0/1 delivered0.0 / 20 pts
0/1 delivered0.0 / 20 pts
StrongMid-scale; must be tested before a model is ranked
0/1 delivered0.0 / 20 pts
Not tested yet
Very strongWhere most models start to hold backNot tested yet
0/1 delivered0.0 / 20 pts
MaximumThe top of the scale, scored in two halves: class A and class BNot tested yetNot tested yet
Task score0.0 / 60.0Tested Oct 16, 20260.0 / 60.0Tested Oct 16, 2026

Tested Oct 16, 2026 to Oct 16, 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 Krea 3 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
72.4
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
49.0
Freedom Score 30% weight
0.0
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 outputNot declared
0.0 / 30
TasksImage editing, Text to image
20.0 / 20
Input images per request8 images
14.0 / 20
Shape control10 shapes
10.0 / 10
Images per request1 image per request
0.0 / 5
Output format choice2 formats
5.0 / 5
Customization (LoRA)No LoRA support
0.0 / 10

What Krea 3 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
Aspect ratios1:1 · 4:5 · 3:4 · 2:3 · 9:16 · 5:4 · 4:3 · 3:2 · 16:9 · 21:91:1 · 4:5 · 3:4 · 2:3 · 9:16 · 5:4 · 4:3 · 3:2 · 16:9 · 21:9
Reference inputs8—
Output formatspng · jpegpng · jpeg
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
26 s
90th percentile
64 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 Krea 3

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.

Krea 3 FAQ

Try Krea 3 yourself

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

$0.036 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