What is the Krea 2 Medium text to image API?

Krea 2 Medium's text-to-image endpoint takes a written description and returns one generated image. The request carries a prompt of up to 5,000 characters, an optional aspect_ratio from five shapes and an optional output_format of png or jpeg, and nothing else. Krea describes this tier as the smaller, faster and more cost-efficient model in the Krea 2 line.

How long can a Krea 2 Medium prompt be?

Up to 5,000 characters on Krea 2 Medium, and at least one, since the prompt is the only required field in the request. That budget matches the other Krea 2 tiers in this catalogue, so a long description written here does not have to be trimmed when the same scene is later run on Krea 2 Large or Krea 2 Medium Turbo.

What happens if I omit aspect_ratio or output_format?

Krea 2 Medium applies its defaults: aspect_ratio becomes 1:1 and output_format becomes png, so a body carrying only a prompt is a complete, valid request. Both fields are enums, which means a value outside the five ratios, or outside png and jpeg, is refused as an invalid request before anything is generated.

What resolution do Krea 2 Medium images come back at?

Krea 2 Medium exposes no width, height or resolution field, so aspect_ratio is the only control over the shape of the frame and the model returns that ratio at its own output size. When a placement needs an exact pixel size, generate at the closest of the five ratios and resize the delivered file afterwards.

Can I send a negative prompt or extra parameters to Krea 2 Medium?

No. Krea 2 Medium accepts prompt, aspect_ratio and output_format and rejects every other key, so a negative_prompt, an output count or a reference URL fails validation instead of being ignored. Anything that should stay out of the frame has to be written into the prompt text itself, which is how the sample prompts here handle it.

Is the Krea 2 Medium API synchronous or asynchronous?

Asynchronous. A call to Krea 2 Medium returns a task record straight away, and the finished image is read back by polling that task or by handling the callback sent to the URL you registered with the request. Nothing holds the HTTP connection open while the image renders, which is what makes a long queue of prompts practical to run.

How much does the Krea 2 Medium API cost?

Krea 2 Medium bills per output image at a flat rate, and no field in the request multiplies that charge: the ratio and the file format leave it unchanged. The live price sits at the top of this page, and the estimate shown when a request is accepted is what settles once the image is delivered.

How do I keep a set of Krea 2 Medium images looking like one series?

Repeat the wording that defines the look. Krea 2 Medium has no reference image input and no seed, so continuity across a set comes from carrying the same light, palette, lens and material language into every prompt, and from holding aspect_ratio fixed. Storing those prompts with the images is what makes the next batch match the first.

Who makes Krea 2 Medium and when was it released?

Krea builds it, and it arrived on 12 May 2026 with the rest of the Krea 2 line. Krea calls Krea 2 its first foundation image model, built from scratch and focused on aesthetics, style transfer and creative control. Medium is the cost-efficient tier of that line, and this endpoint exposes the prompt, the ratio and the output format.

Krea 2 Medium API — Volume Stills, PNG or JPEG

krea/2-medium/text-to-image

Krea 2 Medium is the everyday tier of Krea's text-to-image line, made for frames you produce in volume. It takes a prompt of up to 5,000 characters in five aspect ratios and delivers PNG by default, or JPEG on request.

From $0.03/image

Run model View API

Input

Text to image

Prompt424/5000

Aspect ratio

Output format

JPEGPNG

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

Edit

Split into layers

Animate

{
  "model": "krea/2-medium/text-to-image",
  "input": {
    "aspect_ratio": "3:4",
    "output_format": "png",
    "prompt": "Cinematic film still, a bakery after closing. She sits on the flour-dusted marble counter and he stands close in front of her, both still in half-unbuttoned whites; flour is smeared across her collarbone, his hand flat on the marble beside her. Cold blue-hour light from the street window on one side, the last of the oven glowing orange on the other. 35mm, shallow depth of field, heavy film grain, restrained and tasteful."
  }
}

Introduction

Krea 2 Medium is the Krea 2 line's everyday tier: it doesn't reach for Large's finishing-grade material detail and doesn't chase Medium Turbo's speed. It keeps the line's available-light photographic look and is priced and paced for volume.

Krea 2 Medium is the volume tier of the Krea 2 line: a single text-to-image endpoint, 5,000-character prompts, five aspect ratios and a choice of PNG or JPEG. Built for the frames you make many of.

Key features & highlights

A photographic direction for volume work

Krea 2 Medium is suited to generating multiple candidate frames while a team is still selecting a composition and visual direction. Compare candidates against a single creative brief instead of mixing unrelated ideas in one batch.

Preserve the approved brief for the next stage

Once a direction passes review, keep the wording and the selected frame together. That record is more useful to a finishing workflow than a vague request to recreate the same look later.

Use the model for a new visual, not a source edit

This mode begins with written direction. When an existing image must retain its identity or layout, choose an editing mode instead of asking text-only generation to approximate the source.

Request parameters

Request parameters for Krea 2 Medium · Text to Image
Parameter	Required	Description
`prompt`	Yes	Up to 5,000 characters. Lead with the subject, then the composition, the light and the material; leave the rest unsaid rather than padding it with adjectives. at least 1 characters · at most 5000 characters
`aspect_ratio`	No	Square by default, with the five shapes most placements ask for, from 16:9 landscape to 9:16 portrait. `1:1` · `4:3` · `3:4` · `16:9` · `9:16` · Default: `1:1`
`output_format`	No	png by default, or jpeg. png is lossless and the safer handoff into further editing or compositing; jpeg is a smaller file and enough for review and web placement. `jpeg` · `png` · Default: `png`

Pricing

$0.03 / image

Use cases

Start from a written brief

Use this endpoint when one brief should produce many candidate frames: 5,000-character prompts, five aspect ratios, and PNG by default so the keepers go straight to retouching. There is no image input, so the whole direction has to be in the words.

Internal preview and pitch drafts

Enough quality to carry a concept in a deck without spending a finishing tier's time on a frame that will be replaced. Move the wording that survives review to Krea 2 Large unchanged.

Editorial and portrait batches

Available-light portraiture, interiors and styling frames produced fifty at a time through a single endpoint. The png default is what makes that practical: the keepers go straight into a retouch instead of through a re-export, and nothing is thrown away by a lossy pass before anyone has chosen.

Social and feed assets

Medium offers 1:1, 4:3, 3:4, 16:9 and 9:16 — no widescreen or anamorphic strip — so choose the placement ratio first and write the prompt for it. Pick JPEG when frames go straight to a feed and PNG when they will be retouched.

Explore key use-case tutorials

Prompt guide & pro tips

Write the light, not the adjectives

Describe where the light comes from and what it falls on — a bedside lamp, a window at dusk, one bare bulb — and name the lens and the film stock. Precise light does more for this line's photographic look than words such as "stunning" or "hyper-real".

Keep the prompt with the keeper

With no seed, a good frame cannot be regenerated. Save the prompt beside the asset, and when a wording survives review, carry it to Krea 2 Large unchanged for the finishing pass.

Adapt the included example

Start with the published example using aspect_ratio: 3:4, output_format: png. Replace the subject and creative direction while keeping the input contract for this mode intact.


Available-light photograph seen through a half-open hotel room door, an adult woman sitting on the edge of a made bed wearing only an oversized cream knit sweater, bare legs crossed, absorbed in her phone, one warm bedside lamp, deep falling shadows, quiet and unposed, muted amber palette, 50mm, fine film grain.

The prompt behind the reference frame for this endpoint — it runs unchanged here, though the image shown was produced on the Turbo tier of the same line as a style reference.

Check the result before iterating

Three fields, no more: prompt, aspect_ratio and output_format. There is no seed, so nothing can be pinned and a frame worth keeping has to be stored with its prompt. This endpoint takes no image input — no image-to-image and no reference images — and exposes no creativity control. For style references use Krea 2 Medium Turbo or Krea 2 Large; for the highest detail ceiling, Krea 2 Large. Medium also has no widescreen or anamorphic ratio.

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": "krea/2-medium/text-to-image",
  "input": {
    "prompt": "A ceramic teapot on a linen cloth, soft window light, muted palette",
    "aspect_ratio": "1:1",
    "output_format": "png"
  }
}'

# 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.

The plain-prompt tier for high-volume Krea 2 batches

Krea 2 Medium is Krea's cost-efficient text-to-image endpoint: a prompt of up to 5,000 characters returns one image in one of five aspect ratios, delivered as lossless PNG unless you ask for JPEG. The whole request is three fields, with no seed, no reference image and no creativity setting, so the prompt is the only description the model works from. Krea describes this tier as its smaller, faster model, particularly strong on illustration, anime and painting.

Open the playground

The plain-prompt tier for high-volume Krea 2 batches

Aspect ratios: 1:1, 4:3, 3:4, 16:9, 9:16 (default 1:1)
Prompt length: 1 to 5,000 characters, the only required field
Output format: png (default) or jpeg
Images per call: One image, billed per output image
Not on this tier: No seed, no reference image, no creativity control
Released: 12 May 2026, with the Krea 2 line

Positioning: The smaller, faster tier in Krea's own lineupKrea describes Medium as a smaller, faster and more cost-efficient model with extensive post-training applied, and as particularly strong on illustration, anime and painting. Photorealism and raw looks like motion blur or grain are what Krea points to Krea 2 Large for, a model more than twice the size with softer post-training.
Interface: Three fields, and the schema refuses the restprompt, aspect_ratio and output_format are the entire request body. Any other key fails validation instead of being quietly dropped, so a call that comes back with an image is a call that used every value you sent.
Delivery: Lossless PNG unless you ask otherwisepng is the default here, which keeps every pixel for a crop or a retouch pass. jpeg returns a smaller file for review threads and web placement, and the switch changes the delivered file only, not the image that was generated and not the charge.

What the Krea 2 Medium API does well

A minimal, predictable contract

One string in, one image out: prompt carries the intent, aspect_ratio and output_format are both optional enums with defaults. There is no parameter sheet to keep in sync as the integration grows, and no combination of settings that can silently change what the model is asked to do.

Five aspect ratios, picked before generation

1:1, 4:3, 3:4, 16:9 and 9:16 cover square, landscape and both portrait orientations, with 1:1 applied when the field is left out. The 2.35:1 anamorphic shape is not part of this tier; in this catalogue it belongs to Krea 2 Large and Krea 2 Medium Turbo, which carry eight ratios each.

Prompt wording carries across the Krea 2 line
The 5,000-character budget is the same on Krea 2 Large and Krea 2 Medium Turbo, so a description drafted here runs unchanged when the same scene is tried on another tier. Only the fields around the prompt differ between them, which keeps a prompt library portable instead of tier-specific.
PNG by default, JPEG when transfer size matters
output_format defaults to png, the safer handoff into compositing or retouching. Setting it to jpeg returns a smaller file, enough for a review thread or a direct web placement. The switch changes the delivered file alone: the generated image and the per-image charge stay exactly the same.
One image and one charge per call
Every request produces exactly one image, and the schema has no batch count or variant field that could multiply the cost inside a single call. Volume comes from repeating the request, which makes the cost of a set the number of calls times the per-image rate shown on this page.
Asynchronous, so nothing blocks on the render
A call returns a task record immediately; the finished image arrives when you poll that task or when the callback you registered fires. A queue of a few hundred prompts can therefore be submitted from one worker and collected as results land, without holding connections open.

Who Krea 2 Medium is for

Illustration and concept teams
Painterly and anime-adjacent work without extra setup
Krea points this tier at illustration, anime and painting, and the endpoint puts all of that direction in one place. A prompt naming the medium, the palette and the composition is the whole interface, with no style-reference field to assemble and no strength value to calibrate first.
Batch content production
Running one prompt shape across a long list
A three-field request is straightforward to loop over a spreadsheet of prompt variations, and the flat per-image price makes the cost of a run the row count times the unit price. The response carries a task id per row, so a failed row can be retried on its own.
Prompt drafting before a bigger render
Cheap iteration ahead of a higher-tier pass
Medium shares its prompt budget with the rest of the Krea 2 line, so wording settled here reads the same when the scene is later run on Krea 2 Large. What changes at that point is the fields around the prompt: Large's endpoint carries eight ratios and a style-reference input where this one carries five and output_format.
Product teams shipping a generate button
An interface an end user can actually fill in
Three fields map cleanly onto a product surface: a text box, an optional ratio picker and a format toggle. There is nothing else for someone to misconfigure, and no advanced panel to hide, which keeps the feature explainable in a single line of help text.

How to call the Krea 2 Medium API

Pick aspect_ratio before writing the prompt
Composition reads differently at 9:16 and 16:9, and the model builds the frame for the shape you asked for. Choosing the placement first avoids a crop that throws away staging you already paid for, since the endpoint has no second pass to recover it.
Put everything about the frame into one prompt
The prompt is the only place this endpoint reads intent from, so exclusions belong inside it. That is why two of the sample prompts on this page end by banning text, a signature or a logo in the frame: there is no negative_prompt field to hold that separately.
Set output_format for where the file lands
png when the image feeds compositing, retouching or a print check, because it is lossless. jpeg when the frame is going to a review thread or straight to a web placement and the smaller file matters more than edge fidelity.
Submit the task, then poll or take the callback
The endpoint is asynchronous: the create call returns a task, and the image is read back from that task once it reaches a terminal state. Registering a callback URL saves the polling loop when the job is running inside a queue worker.
Save the prompt with anything you keep
There is no seed on this endpoint, so an identical request draws a fresh frame rather than the same one. The prompt text is the only record that gets you back to the same kind of result, and it belongs next to the asset, not in a scratch file.
Repeat the call for a set
One request returns one image, so a batch is a loop over prompt variations with the same ratio held fixed. Each call is billed on its own, which also means a single bad row can be rerun without regenerating the whole set.

Krea 2 Medium worked examples

An illustrated character study

A single descriptive prompt with medium, palette and pose named up front, aimed at the illustration and painting work Krea points this tier at. aspect_ratio stays at the 1:1 default.

Gouache illustration of a young woman reading in a window seat, warm afternoon light, loose visible brushwork, muted terracotta and sage palette, three-quarter view, soft paper texture, no text and no signature anywhere in the frame

A product frame at a fixed ratio
aspect_ratio set to 4:3 in the request, not described in the prompt, so the composition arrives at the placement's shape and nothing has to be cropped out of it afterward.
```
Studio product photograph of a matte ceramic pour-over coffee dripper on a pale oak counter, single soft overhead light, shallow shadow, clean negative space to the right of frame, no text and no logo
```

A portrait built for a vertical feed

aspect_ratio set to 9:16, with the light source and the mood stated in the prompt, since this endpoint has no setting that adjusts either one after the fact.

Available-light portrait of a man in his forties leaning against a rain-streaked window at dusk, city lights softly blurred behind him, contemplative expression, cool blue-grey palette, fine film grain

Krea 2 Medium API questions

Describe one scene per call and keep the description concrete: a named material, a stated light source and a fixed point of view survive the render better than stacked adjectives. Wording that works here transfers unchanged to the other Krea 2 tiers, so a prompt worth keeping is worth storing wherever your team keeps briefs. A miss is answered with clearer wording and a new call, because this endpoint has no setting to nudge and no seed to hold a result still.

Workbench