GPT Image 2 on Atlas Cloud: OpenAI's Cheapest Image API

GPT Image 2 is OpenAI's text-to-image model on the GPT-5 line. It runs at USD0.008 per image on Atlas Cloud. That makes it the cheapest paid image API in the catalog.

Two model IDs exist. `openai/gpt-image-2/text-to-image` generates from a prompt. `openai/gpt-image-2/edit` edits an existing image with a prompt. Both share the same flat price.

This guide covers pricing, the exact API call, an edit example, and a comparison against Nano Banana 2, Imagen 4 Ultra, and Flux 2 Pro.

 

*Last Updated: May 4, 2026*

Here are real outputs from `openai/gpt-image-2/text-to-image` on Atlas Cloud:

 

GPT Image 2 at a Glance  

 

Why GPT Image 2 Matters

The price is the headline. At USD0.008 per image, GPT Image 2 undercuts every other paid image model on Atlas Cloud. Nano Banana 2 sits at USD0.08. Imagen 4 Ultra sits at USD0.054. Flux 2 Pro sits at USD0.03 to USD0.05. GPT Image 2 is roughly 4x cheaper than the next paid option (Z-Image Turbo at USD0.01) and 10x cheaper than Nano Banana 2.

The edit model uses the same price. That changes the math for image-editing pipelines. A workflow that generates a base image then runs three or four edit passes still costs less than a single Nano Banana 2 generation.

It is built on the GPT-5 image stack. That means strong instruction-following on long prompts and good text rendering inside images. It is not the best for stylized 3D figurines or hyper-photoreal portraits. For those, see the comparison below.

 

GPT Image 2 Pricing

Atlas Cloud charges USD0.008 per image. The price is flat across all three quality tiers and all three sizes. There is no separate cost for the edit model.

The list price on the model page is USD0.01. Atlas Cloud applies a 20% discount, so the billed rate is USD0.008.

The USD1 signup credit covers about 125 images. Enough to run real tests across all three quality tiers and a few edits before deciding whether to top up.

Try GPT Image 2 on Atlas Cloud -- USD1 Free Credit

 

How to Call GPT Image 2

Step 1: Sign up and get an API key

Sign up on atlascloud.ai. The USD1 credit is added automatically. Open the dashboard and create an API key.

 

Step 2: Run the text-to-image call

plaintextCopy
1```python
2import requests
3import time
4
5
6API_KEY = "your-atlas-cloud-api-key"
7BASE_URL = "https://api.atlascloud.ai/api/v1"
8
9
10resp = requests.post(
11    f"{BASE_URL}/model/generateImage",
12    headers={
13        "Authorization": f"Bearer {API_KEY}",
14        "Content-Type": "application/json",
15    },
16    json={
17        "model": "openai/gpt-image-2/text-to-image",
18        "prompt": (
19            "Modern minimalist product photo of a matte black ceramic coffee "
20            "mug on a wooden tabletop, soft window light, neutral background, "
21            "professional product photography"
22        ),
23        "size": "1024x1024",
24        "quality": "medium",
25    },
26    timeout=30,
27)
28
29
30data = resp.json()["data"]
31poll_url = data["urls"]["get"]
32
33
34while True:
35    time.sleep(3)
36    status = requests.get(
37        poll_url,
38        headers={"Authorization": f"Bearer {API_KEY}"},
39    ).json()["data"]
40    if status["status"] == "completed":
41        print(status["outputs"][0])
42        break
43    if status["status"] == "failed":
44        raise RuntimeError(status.get("error"))
45```

 

Step 3: Run an edit call

The edit model takes the same prompt plus an `images` array of source URLs. Field name is `images` (plural). One source image is enough. The model preserves composition and lighting unless the prompt says otherwise.

 

plaintextCopy
1```python
2SOURCE_IMAGE = "https://your-cdn.example.com/source.png"
3
4
5resp = requests.post(
6    f"{BASE_URL}/model/generateImage",
7    headers={
8        "Authorization": f"Bearer {API_KEY}",
9        "Content-Type": "application/json",
10    },
11    json={
12        "model": "openai/gpt-image-2/edit",
13        "prompt": (
14            "Change the mug color to deep forest green and add a small "
15            "spoon resting beside it"
16        ),
17        "images": [SOURCE_IMAGE],
18        "size": "1024x1024",
19        "quality": "medium",
20    },
21)
22```

The poll loop is identical to the text-to-image case.

 

Edit example: before and after

Source image generated with `openai/gpt-image-2/text-to-image`:

 

Same image after a one-line edit prompt through `openai/gpt-image-2/edit`:

Total cost for both calls: USD0.016. The edit kept the wooden surface, lighting, and framing.

 

Parameter Reference

`size`

Three options: `1024x1024`, `1024x1536`, `1536x1024`, `2048x2048`, and `3072x3072`. Square is the default. Use `1024x1536` for portraits and `1536x1024` for banners or hero images. Price is the same for all three.

 

`quality`

Three tiers: `low`, `medium`, `high`. Price is the same for all three. Higher quality takes longer (closer to 30s) and produces cleaner detail. Use `low` for prompt iteration. Move to `medium` or `high` for the final pass.

 

`images` (edit only)

Array of URLs. JPG and PNG accepted. The URL must be publicly fetchable. Atlas Cloud also exposes `POST /api/v1/model/uploadMedia` for uploading local files and getting a `download_url` to pass back in.

 

`input_fidelity` (edit only)

Boolean-like flag that pushes the model to preserve fine detail from the source. Useful for faces, logos, and small text that should survive the edit unchanged.

 

GPT Image 2 vs Other Image Models

 

Where GPT Image 2 wins

Volume jobs. Marketing mockups. UI design exploration. Concept art batches. Anything where the cost of being wrong on a generation is the cost of regenerating it.

Edit pipelines. Generate once, edit five times, total spend is still under five cents.

Long, detailed prompts. The GPT-5 base handles paragraph-length prompts without losing the early instructions.

 

Where GPT Image 2 falls short

Figurines and stylized 3D characters. Use Nano Banana 2 for this. The figurine-specific material rendering and packaging detail is not GPT Image 2's strength.

Premium photoreal portraits at print size. Use Imagen 4 Ultra when the image will sit on a homepage or in print.

Reference-image style transfer for team headshots. Use Flux 2 Pro with `reference_image_url`.

 

Who Should Use GPT Image 2

Choose GPT Image 2 if:

• The use case is high-volume image generation (1K+ images per month).
• You need both text-to-image and edit in one pipeline.
• You are building a SaaS feature where image cost flows through to the unit economics.
• You want a single model that handles marketing visuals, UI mockups, and concept art at one price.
• Edit-and-iterate workflows are core to the product.

 

Skip GPT Image 2 if:

• Output style is the only thing that matters and that style is "Nano Banana 2 figurines."
• The image is a hero asset and photorealism trumps cost.
• You need 4K output. GPT Image 2 caps at 1536 on the long edge.

 

Frequently Asked Questions

How much does GPT Image 2 cost per image?

USD0.008 on Atlas Cloud. Flat across all sizes and quality tiers. The USD1 free signup credit covers about 125 images.

 

Is the edit model priced separately?

No. Both `openai/gpt-image-2/text-to-image` and `openai/gpt-image-2/edit` charge USD0.008 per output image.

 

What sizes does GPT Image 2 support?

Three: `1024x1024`, `1024x1536`, `1536x1024`. No 4K option.

 

How do I pass an image to the edit model?

Pass an `images` array of public URLs. JPG or PNG. One image is enough. Multiple images can be passed for style or composition reference.

 

How long does a generation take?

15-30 seconds typical. `low` quality finishes faster. `high` quality takes longer.

 

Can I use GPT Image 2 commercially?

Yes. Atlas Cloud passes through OpenAI's standard commercial rights for the model. Check the OpenAI usage policy for excluded categories.

 

Does GPT Image 2 render text inside images?

Yes, and well. It is one of the stronger options for posters, signage, UI mockups, and any layout that needs a few words rendered inside the image.

 

Why is the response shape different from the OpenAI API directly?

Atlas Cloud uses an async pattern. The POST returns a `request_id`. The poll URL returns `outputs` (array of image URLs) once the status flips to `completed`. The poll URL is in `data.urls.get` from the initial response.

 

Verdict

GPT Image 2 is the new default for cost-sensitive image generation on Atlas Cloud. The price puts it in a different category from every other paid model in the catalog. The edit model carries the same price, which makes iterate-and-refine pipelines genuinely cheap.

It is not the best at every style. Use Nano Banana 2 for figurines, Imagen 4 Ultra for premium portraits, Flux 2 Pro for reference-driven team work. But for everything else, GPT Image 2 is the call.

Get USD1 Free Credit on Atlas Cloud -- Try GPT Image 2 and 300+ Models

 

────────────────────────────────────────────────────────────

Related Articles

• Best GPT Image 2 Prompts: Marketing, Mockups, and Edits
• Best AI Image Generation Models 2026
• Best AI Image Editing Models 2026
• Cheapest AI Image Generation API 2026
• Nano Banana 2 API Guide
• Imagen 4 Ultra API Guide
• Flux 2 Pro Deep Dive