Qwen-Image 2.1 vs Grok Imagine vs Nano Banana 2.1

Short answer: Qwen-Image 2.1 Turbo costs less for reference edits

LLMTR is a gateway that gives access to AI models from different providers through one OpenAI-compatible API. Qwen-Image 2.1 Turbo (alibaba/qwen-image-2.1-turbo), Qwen-Image 2.1 Pro (alibaba/qwen-image-2.1-pro), Grok Imagine (xai/grok-imagine-image and xai/grok-imagine-image-quality) and Nano Banana 2.1 (google/gemini-nano-banana-2.1) are called here at https://llmtr.com/v1 with the same API key from your llmtr.com account. Qwen-Image 2.1 and Nano Banana 2.1 take the reference image in the image_urls field of a /v1/images/generations request; editing with Grok Imagine goes through the /v1/images/edits endpoint.

In a single-reference edit, such as replacing the background of a product photo, Qwen-Image 2.1 Turbo costs $0.016 per generated image and the reference image you send is not billed. On Grok Imagine an edit counts both the images you send and the images it generates. If you need 4K output, the option is Nano Banana 2.1; Qwen-Image 2.1 tops out at a 2048x2048 pixel area and Grok Imagine at 2k.

• Alibaba Cloud, Model Studio pricing · 2026-10-19
• Google, Gemini API pricing · 2026-10-19

Image editing API: Qwen-Image 2.1, Grok Imagine or Nano Banana 2.1?

Four questions decide it: what you want to pay per image, how many reference images you send, which resolution you need, and which endpoint your edit request will go to. Qwen-Image 2.1 and Grok Imagine bill a fixed price per image. Nano Banana 2.1 bills by token; the figures in the table are Google's per-image equivalents, and the prompt and reference images count separately as input tokens.

The table shows the three model families with the values in the LLMTR catalog and docs. Turbo and Pro use the same request shape, so moving between them is only a change of the model id. The same holds for the two Grok Imagine models.

What does the same edit cost on each model?

Picture 100 product photos, each edited with one reference into one new image. On Qwen-Image 2.1 Turbo the total is $1.60 and on Qwen-Image 2.1 Pro $4.00; neither counts the reference images. On Grok Imagine each edit counts as two images, which at catalog prices comes to $4.00 on the standard model and $10.00 on the Quality model; when xAI reports a request's real cost, that cost is what you pay. On Nano Banana 2.1 the images come to $3.36 at 1K output, plus the input tokens of the prompt and the reference images.

The Python example below sends the same reference and the same prompt to the five models and writes the returned images to files. Qwen-Image 2.1 and Nano Banana 2.1 go to /v1/images/generations with image_urls, and Grok Imagine goes to /v1/images/edits with an image object. Put the results side by side and choose the model that fits your work by looking at your own images.

Python example that edits the same product photo with Qwen-Image 2.1 Turbo, Qwen-Image 2.1 Pro, Nano Banana 2.1 and the two Grok Imagine models through https://llmtr.com/v1 and saves the results.

import base64
import os

import requests

BASE = "https://llmtr.com/v1"
HEADERS = {"Authorization": "Bearer " + os.environ["LLMTR_API_KEY"]}
PROMPT = "Keep the product unchanged and replace the background with a light beige studio wall."
REFERENCE = "https://example.com/product.png"


def save(model, response):
    response.raise_for_status()
    for i, item in enumerate(response.json()["data"]):
        ext = item.get("mime_type", "image/png").split("/")[1]
        name = model.split("/")[1] + "-" + str(i) + "." + ext
        if item.get("b64_json"):
            data = base64.b64decode(item["b64_json"])
        else:
            data = requests.get(item["url"], timeout=60).content
        with open(name, "wb") as f:
            f.write(data)
        print(model, name)


# Qwen-Image 2.1 and Nano Banana 2.1: the reference goes in image_urls on /v1/images/generations.
for model in ["alibaba/qwen-image-2.1-turbo", "alibaba/qwen-image-2.1-pro", "google/gemini-nano-banana-2.1"]:
    body = {"model": model, "prompt": PROMPT, "image_urls": [REFERENCE]}
    save(model, requests.post(BASE + "/images/generations", headers=HEADERS, json=body, timeout=600))

# Grok Imagine: edits go to /v1/images/edits with an image object.
for model in ["xai/grok-imagine-image", "xai/grok-imagine-image-quality"]:
    body = {"model": model, "prompt": PROMPT, "image": {"type": "image_url", "url": REFERENCE}, "response_format": "b64_json"}
    save(model, requests.post(BASE + "/images/edits", headers=HEADERS, json=body, timeout=600))

Reference images, resolution and the result address

On Qwen-Image 2.1 the references go in the image_urls field as https addresses or base64 data URLs, with up to 8 images in a request. You give the size as width and height, such as 1024x1024, and n returns between 1 and 6 images in one request. negative_prompt is not supported in this series; write what you do not want into the prompt. If the prompt asks for a transparent background, the model can return a transparent PNG. The result address is valid for 24 hours, so save the image within that time.

In a Grok Imagine edit the image goes in the image field, several images go in the images array, and the array takes up to 5 images. Resolution is 1k or 2k. The returned url is short-lived; if you will show the image in your interface for good, ask for b64_json as the response_format.

Nano Banana 2.1 takes references through image_urls or file_ids from the Files API, and the two share a limit of 8 images. If you reuse the same reference again and again, upload it once and send the same file_id for 7 days. Output is chosen as 1K, 2K or 4K. The model can think before it generates the image, and those tokens bill at the text output price.

• Alibaba Cloud, Qwen-Image generation and editing API reference · 2026-10-19

Which one when?

The choices below follow from the differences in price, limits and request shape; compare image quality with your own samples.

• High-volume edits such as replacing the background of catalog photos or placing a product in a new scene: Qwen-Image 2.1 Turbo. It costs $0.016 per image and the reference images are not billed.
• Images in the same job where detail matters: Qwen-Image 2.1 Pro. The card in the LLMTR catalog recommends Turbo for high-volume work and drafts and Pro when fine detail matters; the request shape is the same.
• Several variations in one request: Qwen-Image 2.1, up to 6 images with n. On Nano Banana 2.1 each request returns one image.
• 4K output, or reference files reused over 7 days: Nano Banana 2.1.
• Your application already works with the /v1/images/edits shape and 2k is enough: Grok Imagine. Count the image you send in the edit in your cost estimate.

Trying all five with the same key

Open an account on llmtr.com, top up your balance under Dashboard > Billing and create a key under Dashboard > API Keys. The same key works on all five models, and you do not need a separate account with Alibaba, xAI or Google. Each image is charged to your balance at the catalog price in force when it is generated.

Run the trial with 10 to 20 samples from your own product photos, send the same prompt to each model and put the results side by side with the cost per image. Setting a spending limit on your API key caps the cost during the trial.

Comparing image editing models with the same reference

Edit the same product photo with Qwen-Image 2.1, Grok Imagine and Nano Banana 2.1 using one LLMTR key, then compare the result and the cost per image.

1. Key. Top up your balance on llmtr.com and create a key under Dashboard > API Keys.
2. Same reference. Send the same image to Qwen-Image 2.1 and Nano Banana 2.1 through /v1/images/generations and to Grok Imagine through /v1/images/edits.
3. Choice. Compare the results with the cost per image, then choose the model id.

Frequently asked questions

What is the difference between Qwen-Image 2.1 Turbo and Pro?

Both generate images from text and edit images with the same request shape. Turbo costs $0.016 per generated image and Pro $0.04. Send the same prompt and reference to both models and compare the result on your own work.

How is an edit with Grok Imagine billed?

An edit counts both the images sent and the images generated; the standard model costs $0.02 per image and the Quality model $0.05. When xAI reports a request's real cost, that cost is what you pay.

Which model offers 4K images?

Nano Banana 2.1; Google's per-image equivalent at 4K is $0.113. Qwen-Image 2.1 produces up to a 2048x2048 pixel area and Grok Imagine up to 2k resolution.

Related posts

• Editing with reference images, multi-reference and try-on in Nano Banana 2.1
• Grok Imagine Image: why the standard tier is enough for most product work

• Qwen-Image 2.1 Turbo model
• Qwen-Image 2.1 Pro model
• Grok Imagine Image model
• Grok Imagine Image Quality model
• Gemini Nano Banana 2.1 model
• Image generation guide
• Google image generation documentation