GPT Image 1.5 I2I vs GPT Image 2 I2I: Credits, Ratios and Limits

GPT Image 2 image-to-image beats GPT Image 1.5 image-to-image on almost every axis that matters: it costs 3 credits at 1K against a flat 10 credits for 1.5, it offers 11 aspect ratios instead of 3, it accepts prompts up to 20,000 characters instead of 3,000, and it adds real 1K/2K/4K resolution control. Pick GPT Image 1.5 only when you are matching the look of edits you already shipped with it, or when its medium/high quality switch fits your pipeline better than a resolution picker. For everything else on VdoBloom, GPT Image 2 is the default.

Spec against spec

Spec	GPT Image 1.5 I2I	GPT Image 2 I2I
Model id	gpt-image/1.5-image-to-image	gpt-image-2-image-to-image
Credits per image	10 flat	3 at 1K, 5 at 2K, 8 at 4K
Dollar cost at Lite rates	$0.50	$0.15 / $0.25 / $0.40
Quality or resolution control	Quality: medium or high	Resolution: 1K, 2K, 4K
Aspect ratios	1:1, 2:3, 3:2	auto, 1:1, 5:4, 9:16, 21:9, 16:9, 4:3, 3:2, 4:5, 3:4, 2:3
Input images per request	Up to 10	Up to 10
Maximum prompt length	3,000 characters	20,000 characters
Maximum file size per input	10 MB	30 MB
Model rating	4.4	4.6
Content filter	Strict	Strict

Credits convert at one rate across the whole platform: the Lite plan is $15 for 300 credits, so a credit is $0.05. That makes a single 1.5 edit $0.50 and a 1K GPT Image 2 edit $0.15 — the same job for less than a third of the price. Even a 4K GPT Image 2 render at 8 credits undercuts 1.5 by two credits. The 10 free credits a new account receives, with no card, buy one 1.5 edit or three 1K GPT Image 2 edits. Plan tiers are on the pricing page.

The two constraints you must know about GPT Image 2

Its flexibility comes with two rules that are enforced before any credits leave your balance. First, the auto aspect ratio forces 1K — if you ask for auto plus 2K or 4K, the request is clamped down to 1K rather than failing, so you pay the 1K price and get the 1K image. Second, a 1:1 square cannot be rendered at 4K; that combination is rejected outright with a message telling you to use 2K or a non-square ratio. Neither rule exists on GPT Image 1.5 because it has no resolution concept at all. If you want a big square, pick 2K square, or render 4K in 4:3 and crop.

Where GPT Image 1.5 still earns its 10 credits

Honestly, the list is short, but it is not empty.

• Look consistency. If a campaign, catalogue or series was already edited with 1.5, continuing on 1.5 keeps the rendering style and colour handling consistent. Switching mid-series is visible.
• Three ratios, no decisions. Some teams prefer a model that only offers 1:1, 2:3 and 3:2, because it removes the aspect-ratio argument from the brief entirely.
• The high quality switch. 1.5 exposes a medium/high quality toggle at a flat price, which means high never costs more. On GPT Image 2 the equivalent step up is a resolution change that does cost more.

What it does not win on is price, prompt headroom, framing options or file-size tolerance. A 3,000-character prompt cap is genuinely limiting on multi-image composites where you have to describe each reference and what to preserve from it.

Where GPT Image 2 wins outright

Multi-image composition is the big one. Both models accept up to 10 input images, but only GPT Image 2 gives you 20,000 characters to explain them — enough to say which image supplies the subject, which supplies the environment, which supplies the lighting, and exactly which details must survive untouched. Vertical and ultrawide work is the other: 9:16 for stories and reels and 21:9 for site banners simply do not exist on 1.5, so producing them means rendering 3:2 and cropping, which throws away resolution you paid for. And the 30 MB per-file limit means you can hand it camera-original JPEGs instead of pre-compressing them.

A worked example: the same job, both models

You have a product photo, a studio backdrop photo and a model photo, and you need six channel assets. On GPT Image 1.5 that is six passes at 10 credits, 60 credits total, $3.00 — and two of the six assets have to be cropped down from 3:2 because there is no 9:16. On GPT Image 2 at 2K, that is six passes at 5 credits, 30 credits, $1.50, every asset rendered natively in its target ratio, with a prompt long enough to lock the label text. Same inputs, half the cost, better framing. If 1K is enough for social, the same six passes cost 18 credits, $0.90. Both models run from Image Edit, so switching between them is a dropdown, not a new workflow.

Both are strict-filter models, which makes them good at commercial product, brand and portrait work and unwilling to produce the swimwear, dance and fitness material our flexible-tier models handle. If any input photo shows a real person, you must have that person’s consent before editing it.

When to pick neither

GPT Image 2.5 arrived as two variants, Flare and Sunburst, and both are priced identically to GPT Image 2 — 3 credits at 1K, 5 at 2K, 8 at 4K — while accepting up to 16 input images instead of 10. Flare is the fast variant, Sunburst the precision one; Sunburst trades latency, not price. They rate 4.7 and 4.8 against GPT Image 2’s 4.6. If you are starting something new today and want the most input images per pass, start there. For a cheaper multi-reference edit outside the GPT family, Seedream 5 Lite image-to-image is 6 credits with up to 14 references, and Nano Banana Edit is 5 credits with up to 10.

The verdict by use case

Social and e-commerce volume: GPT Image 2 at 1K, 3 credits. Hero and print assets: GPT Image 2 at 4K, 8 credits, in a non-square ratio. Long multi-image composites: GPT Image 2, purely for the 20,000-character prompt. Continuing a series already shipped on 1.5: GPT Image 1.5, 10 credits, for consistency. Anything square at maximum size: GPT Image 2 at 2K, because 4K square is blocked. Full spec pages sit at GPT Image 1.5 I2I and GPT Image 2 I2I, and there is a task-shaped walkthrough at AI image editor from photo.

Frequently asked questions

Is GPT Image 2 image-to-image really cheaper than GPT Image 1.5?

Yes, at every tier. GPT Image 1.5 is a flat 10 credits per image. GPT Image 2 is 3 credits at 1K, 5 at 2K and 8 at 4K, so even its most expensive setting costs less than 1.5.

Why did my GPT Image 2 request come back at 1K when I asked for 4K?

You almost certainly had the aspect ratio set to auto. Auto forces 1K, and the request is clamped rather than rejected. Choose an explicit ratio such as 16:9 or 3:2 to get 2K or 4K.

Can I get a 4K square image from GPT Image 2?

No. A 1:1 ratio at 4K is rejected before generation starts. Use 2K for square output, or render 4K in 4:3 or 5:4 and crop to square afterwards.

How many images can I upload to each model?

Both accept up to 10 inputs per request. GPT Image 1.5 caps each file at 10 MB, GPT Image 2 at 30 MB. GPT Image 2.5 Flare and Sunburst raise the input count to 16 at the same price as GPT Image 2.

Does GPT Image 1.5 high quality cost extra?

No. Its medium and high quality settings both bill at the same flat 10 credits, so there is no reason to run it on medium.

Are the outputs watermarked?

Free-plan downloads carry a watermark. Every paid plan, starting at Lite for $15 a month with 300 credits, downloads watermark-free with commercial rights, and one-time credit packs from $2.49 for 75 credits never expire.