Grok Imagine 2.0: What's New, Access and API (2026)

Grok Imagine 2.0 is the current image model from xAI, the company behind the Grok chatbot, and it is the version most people mean when they talk about Grok’s image generator in late 2026. You give it a text prompt, or a prompt plus up to three source images, and it returns finished pictures in the loose, slightly irreverent style Grok is known for, now with more detail and better control. This post covers what changed, how spicy mode and the free tier work today, how to reach Grok image and video models from code, and what each request costs.

What is Grok Imagine 2.0?

Grok Imagine is xAI’s purpose-built visual product. It is not the same thing as the Grok chat models. Grok 2, the conversational model, could produce images as a side feature of chatting. Grok Imagine was built specifically for visual content: still images first, then video.

Grok Imagine 2.0 is the newest generation of the image side. It is a text-to-image model, meaning a machine learning model that takes a natural-language description and produces a picture matching it. Like most current image models, it also works image-to-image: you can hand it existing pictures and ask it to edit, restyle or combine them.

The headline capabilities of Grok Imagine 2.0 are:

• Multi-reference input. Up to three source images per request, useful for keeping a character, product or style consistent.
• Native editing. Change parts of an existing image from a prompt instead of regenerating it from scratch.
• Typography. Readable text inside the image, such as signs, labels and posters, which older models often garbled.
• Configurable output. Aspect ratio, resolution and quality settings.

Reception has been mixed but mostly positive on image quality. Users tend to praise it for abstract and artistic styles, and some say it beats Midjourney for images that contain text. In September 2026 it was reported to have jumped to fourth place on a widely followed independent text-to-image leaderboard, up 14 spots from the previous generation. Leaderboards shift often, so treat that as a sign of momentum rather than a permanent ranking.

What changed from earlier Grok image models?

Grok’s image story has three stages, and the naming causes most of the confusion people bring to search engines.

Grok 2: images inside a chatbot

Grok 2 was a general conversational model that could also generate images when you asked. It was convenient, but image generation was a feature of the chat model, not its purpose. Control was limited to whatever you could express in a chat message.

Grok Imagine: a dedicated visual model

Grok Imagine split image and video generation into its own product, with its own interface in the Grok app. The first generation returned several variations per prompt, so you could pick the best one instead of re-rolling. It also introduced animating images into short clips, which is where the “Grok Imagine videos” searches come from.

Grok Imagine 2.0: fidelity and control

Grok Imagine 2.0 keeps the same character and the variation-based workflow but raises fidelity, and adds the control features that working designers asked for: editing, multi-reference input and reliable text. If you tried Grok Imagine in 2025 and found it fun but hard to steer, the 2.0 release targets exactly that complaint.

A practical way to think about it: Grok 2 answered “can a chatbot draw?”, the first Grok Imagine answered “can Grok make good-looking pictures fast?”, and Grok Imagine 2.0 answers “can I get the specific picture I need, with my product, my character and my text in it?”

Grok Imagine spicy mode, safety and the free tier

Two topics dominate Grok Imagine Reddit threads: spicy mode and whether the tool is still free.

What spicy mode is

Spicy mode is a setting in xAI’s consumer Grok app that allows more permissive, adult-leaning output than the default. It sits behind account and age settings. In early 2026, Grok became the centre of a scandal after people used its image editing features on X and in the standalone app to produce sexualised edits of real people. xAI responded by restricting parts of the Imagine feature set, and the rules have kept moving since.

If you are a developer, the takeaway is simple: do not build anything that depends on spicy mode. API access is a separate surface with its own content rules, and permissive consumer settings do not carry over. Editing images of real, identifiable people also carries legal risk in many places, whatever the tool allows.

Is Grok Imagine no longer free?

The free allowance in the Grok app has changed several times. Many users report tighter daily limits, slower queues or features that now need a paid subscription. Exact limits depend on your account type and region, so check the app’s current terms rather than trusting a months-old thread.

Some third-party sites advertise “Grok Imagine 2.0 free” access. Usually that means a small trial allowance on someone else’s account, not unlimited use. If you need predictable access for a product or a client project, pay-per-request API access is the more stable route: each image has a known price and no daily cap resets on you mid-project.

How to access Grok Imagine 2.0: app, X and API

There are four common ways to use Grok Imagine 2.0 or its siblings:

1. The Grok app and website. The simplest option for personal use. Type a prompt, pick from the variations, optionally animate one.
2. X. Grok features, including image editing, are built into X for subscribers.
3. Third-party web tools. Various sites wrap the model with their own interface and trial credits.
4. An API. For apps, batch jobs, automation and anything you want to repeat reliably.

The rest of this post focuses on the API route, because that is where you get control over cost, volume and integration.

Grok models on Unicorn API

Unicorn API gives you one API key and one bill for image, video, audio and chat models from many providers. The catalog lists xAI’s image model as Grok Imagine (model id grok-imagine), with a listing date of 2025-11-03. Check the model page for the exact version being served before you assume it matches the newest Grok Imagine 2.0 release in the app. Alongside it are three Grok video models: Grok Imagine Video 1.5, Grok Imagine Video 1.0 and Grok Imagine Upscale.

What the image model does through the API:

• Returns 6 variations per text prompt, or 2 when you attach reference images.
• Accepts prompt, ref_images, aspect_ratio and n as input fields.
• Supports aspect ratios 1:1, 2:3, 3:2, 16:9 and 9:16.
• Is listed at 4 credits, multiplied by n when you set it, rounded up to whole credits.

Credits on monthly plans cost $0.01 each (Starter $25, Creator $50, Studio $100, Pro $200 per month). Top-up credits cost $0.015 each and never expire. So 4 credits is about $0.04 on a plan or about $0.06 from top-ups. Because the model returns several variations per call, use the price endpoint (POST /v1/models/price) to see the exact cost of your specific request before you run it. Full plan details are on the pricing page.

If you want to compare outputs before writing any code, the image playground runs the same models in the browser.

How to call the Grok Imagine API from code

Image and video models run as asynchronous jobs. You submit a job, get back an id, poll until the status is succeeded, then read the output URL. Here is the full flow.

Step 1: get an API key

Create a key on the API keys page and export it in your shell:

export UNICORN_API_KEY="your-key-here"

Step 2: submit a job with curl

curl -X POST https://api.unicornapi.net/v1/jobs \
  -H "Authorization: Bearer $UNICORN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "grok-imagine",
    "input": {
      "prompt": "A hand-painted poster of a lighthouse in a storm, bold title text reading NORTH LIGHT, teal and orange palette",
      "aspect_ratio": "2:3"
    }
  }'

The response contains a job id. Poll GET /v1/jobs/{id} with the same header until it finishes.

Step 3: a complete Python script

This script submits a Grok Imagine job, polls until it finishes and prints every output URL, since the model returns several variations.

import os
import time
import requests

API = "https://api.unicornapi.net"
HEADERS = {
    "Authorization": f"Bearer {os.environ['UNICORN_API_KEY']}",
    "Content-Type": "application/json",
}

def run_job(model, input_fields, poll_seconds=3, timeout=600):
    resp = requests.post(
        f"{API}/v1/jobs",
        headers=HEADERS,
        json={"model": model, "input": input_fields},
    )
    resp.raise_for_status()
    job = resp.json()
    job_id = job["id"]

    deadline = time.time() + timeout
    while time.time() < deadline:
        job = requests.get(f"{API}/v1/jobs/{job_id}", headers=HEADERS).json()
        status = job.get("status")
        if status == "succeeded":
            return job
        if status in ("failed", "canceled"):
            raise RuntimeError(f"Job {job_id} ended with status {status}: {job}")
        time.sleep(poll_seconds)
    raise TimeoutError(f"Job {job_id} did not finish in {timeout} seconds")

if __name__ == "__main__":
    job = run_job(
        "grok-imagine",
        {
            "prompt": (
                "A hand-painted poster of a lighthouse in a storm, "
                "bold title text reading NORTH LIGHT, teal and orange palette"
            ),
            "aspect_ratio": "2:3",
        },
    )
    print("First image:", job["outputs"][0]["url"])
    for i, output in enumerate(job["outputs"]):
        print(f"Variation {i + 1}: {output['url']}")

Output URLs are where you download the finished files. Save the images you want to keep to your own storage. The API documentation covers the job object, reference image formats and error codes in full.

If you work inside Cursor, Claude Code or another coding agent, you can also call these models through MCP, which is described on the agents page.

Grok Imagine videos: animating images and upscaling

Grok Imagine videos come from dedicated video models, not the image model. They pair well with Grok Imagine 2.0 style stills: generate a frame you like, then animate it.

Model	What it does	Durations	Resolutions	Example price
Grok Imagine Video 1.5	Image-to-video only: animates a start frame, strong motion quality	1-15 s	480p, 720p, 1080p	5 s at 720p: 22.5 credits (billed as 23)
Grok Imagine Video 1.0	Text-to-video or image-to-video, same per-second rate	6-30 s	480p, 720p, 1080p	10 s at 720p: 45 credits
Grok Imagine Upscale	Upscales an earlier Grok video job by its task id	n/a	720p, 1080p	10 credits at 720p, 20 at 1080p

The per-second rate for both video models is 2.4 credits at 480p, 4.5 at 720p and 8 at 1080p. Video 1.5 has a lower floor (1 second) and is the better choice for short loops and social clips. Video 1.0 is the one to pick when you need longer clips, up to 30 seconds, or when you want to start from text alone. Video 1.5 can be busy at peak times, so build retries into anything time-sensitive.

Animating a Grok image

Using the run_job helper from the script above, pass the image URL from your first job as the start frame:

video = run_job(
    "grok-imagine-video-1-5-preview",
    {
        "image": job["outputs"][0]["url"],
        "prompt": "Waves crash against the lighthouse, the beam sweeps slowly through rain",
        "duration": 5,
        "resolution": "720p",
        "aspect_ratio": "2:3",
    },
)
print("Video:", video["outputs"][0]["url"])

Worked example: the cost of one finished clip

Say you want a poster image, a 5-second animated version and a 1080p final:

• Grok Imagine image job: 4 credits
• Video 1.5, 5 seconds at 720p: 22.5 credits, rounded up to 23
• Upscale to 1080p: 20 credits

Total: 47 credits, about $0.47 on a monthly plan or about $0.71 from top-ups. Rendering the same 5 seconds directly at 1080p on Video 1.5 costs 40 credits instead of 23 plus 20, so upscaling makes most sense when you preview several clips cheaply at 720p and only upscale the keeper.

Grok Imagine 2.0 vs other image models

Grok Imagine 2.0 is strong for artistic styles and text in images, but it is not the only good option at this price. Here is how the Grok image model on Unicorn API compares with two common alternatives in the same catalog:

	Grok Imagine	Nano Banana 2 Lite	GPT Image 1.5
Provider	xAI	Google	OpenAI
Price	4 credits	4 credits per image	3 low, 4 medium, 22 high per image
Outputs per request	6 variations (2 with references)	Per image, set with n	Per image, set with n
Reference images	Yes	Up to 10	Up to 4
Aspect ratios	1:1, 2:3, 3:2, 16:9, 9:16	16:9, 1:1, 9:16, 21:9, 4:3, 3:2, 2:3	1:1, 3:2, 2:3
Stand-out strength	Many variations, artistic range	Fast, cheap 1K images and edits	Prompt adherence, legible text, transparent background

How to choose:

• Pick Grok Imagine when you are exploring a look and want several takes per prompt, or when you plan to animate the result with Grok’s video models.
• Pick Nano Banana 2 Lite for high-volume edits, wide formats like 21:9, or when you need many reference images to lock in a product or character.
• Pick GPT Image 1.5 when strict prompt adherence matters most, or when you need a transparent background for logos, stickers and product cut-outs.

Since all three sit behind the same key, you can run one prompt through each and keep the winner. For a deeper look at Google’s image line, see Imagen 4 vs Nano Banana. The full model catalog lists every other option.

Common mistakes with Grok Imagine 2.0

Most wasted money and time with Grok Imagine 2.0 and its API siblings comes from a handful of avoidable errors.

1. Assuming the API version matches the app. Apps and APIs update on different schedules. Check the model page and run a test prompt before you promise a client “Grok Imagine 2.0 quality”.
2. Re-rolling instead of reviewing. The image model already returns several variations per call. Look at all of them before paying for another round. Many people only open the first URL.
3. Mismatched aspect ratios. If you will animate the image, generate the still at the aspect ratio your video needs (16:9 for landscape, 9:16 for vertical). Cropping later throws away detail and can confuse the motion.
4. Rendering everything at 1080p. Preview at 480p or 720p, then upscale or re-render only the clip you will publish.
5. Ignoring rounding. Video prices are rounded up per output, so a 22.5-credit clip bills as 23. It adds up across hundreds of jobs; the price endpoint shows the real figure.
6. Polling without a timeout. Video jobs can queue at busy times. Poll every few seconds, cap the wait, and handle failed jobs instead of looping forever.
7. Building on spicy mode or real faces. Permissive consumer settings do not carry over to APIs, and editing identifiable people invites legal and platform trouble. Use original characters or people who have given consent.
8. Vague text prompts. Grok Imagine 2.0 renders text well, but only when you spell it out. Put the exact words in quotes or capitals and say where they go, such as “title text at the top reading NORTH LIGHT”.

Get those right and Grok Imagine 2.0 is one of the more enjoyable image models to work with: fast to explore, good with style and text, and a natural starting point for short Grok videos. Start in the playground to find your prompt, then move the same request into code when you are ready to repeat it.