How to Use GPT Image 1.5 Image-to-Image: Specs, Credits and Prompting

GPT Image 1.5 image-to-image is VdoBloom’s instruction-following photo editor: you attach up to 10 reference images, write an editing instruction of up to 3,000 characters, pick 1:1, 2:3 or 3:2 as the output frame and medium or high quality, and every run costs a flat 10 credits regardless of the settings you choose. It is the model to reach for when the edit is a list of conditions that must all survive — change this, keep that, match the other — rather than a vague restyle.

What GPT Image 1.5 I2I is actually best at

GPT Image 1.5 comes out of OpenAI’s multimodal stack, and that heritage shows in one specific place: it reads editing instructions the way a language model reads text. A prompt like “replace the grey hoodie with a denim jacket, keep the chest logo legible and unchanged, and keep the lighting coming from camera left” tends to arrive intact on the other side. Most diffusion-first editors treat that sentence as a mood board and give you a denim-ish image where the logo has quietly turned into a smudge.

That comprehension advantage is worth the most in three places. Text-bearing edits — packaging, signage, UI mockups, price tags — where a wrong glyph ruins the asset. Conditional edits, where half the instruction is a list of things that must not change. And multi-source composites, because all 10 attached images are read as joint context for one instruction: subject from image one, jacket texture from image two, background from image three.

What it is not best at is size. There is no 4K here and no ultrawide frame. GPT Image 1.5 I2I spends its budget on executing the instruction faithfully at sensible dimensions, and the quality toggle changes rendering effort, not pixel count.

The real specs and credit cost

Everything below is read straight from VdoBloom’s model config and credit table, not from a marketing page.

Spec	GPT Image 1.5 image-to-image
Model family	OpenAI GPT Image
Capability	Image-to-image editing (reference image required)
Reference images per job	Up to 10
Prompt length limit	3,000 characters
Aspect ratios	1:1, 2:3, 3:2 (default 3:2)
Quality settings	Medium (default), High
Resolution tiers	None — no 1K/2K/4K selector
Credit cost	10 credits per image, flat
Dollar cost at Lite rates	$0.50 per image ($15 = 300 credits, so 1 credit = $0.05)
Content filter tier	Strict — OpenAI’s own provider filters
User rating on VdoBloom	4.4

The flat price is the part to internalise. Unlike the resolution-tiered editors on the platform, medium and high cost exactly the same 10 credits, which means there is no financial reason to draft at medium. Draft at high and spend your credits on iterations instead.

How to prompt it well

Write instructions, not descriptions. The single biggest gain with this model comes from switching your prompt from “a woman in a denim jacket in golden hour light” to “change the jacket to denim. Do not change her face, hair, pose or the background.” The first is a generation prompt; the second is an edit instruction, and this model was built for the second.

• Name the preservation list explicitly. Anything you do not pin down is fair game for the model. Logos, text, jewellery, hands and background signage are the usual casualties.
• Order your steps. Numbered or comma-chained steps survive better here than in almost any other editor on VdoBloom. Two or three changes in one job is realistic; eight is not.
• Label your attachments in the prompt. With multiple references, say which is which: “use the second image only as the fabric texture reference.”
• Say where the light comes from. Compositing failures read as lighting failures far more often than as shape failures.
• Match the frame to the source. With only three ratios and no auto mode, pick 3:2 for landscape sources and 2:3 for portrait ones so the model is not forced to recompose.

One rule that has nothing to do with quality: if your source photo shows a real, identifiable person, you need that person’s consent before you edit and publish their likeness. That is on you, not the model.

A worked example

Say you have a product photo of a skincare bottle on a white sweep and you want a lifestyle variant. Attach three images: the product shot, a marble bathroom counter, and a reference photo with the warm window light you like. Set the frame to 3:2, quality to high, and write:

“Place the bottle from image 1 on the marble counter from image 2. Keep the label text, logo and cap colour exactly as they are in image 1. Match the warm side lighting from image 3, with a soft shadow falling to the right. Do not add any other products or text.”

That is one job, 10 credits, $0.50. Budget three attempts and you have a usable lifestyle asset for $1.50 — and 10 free credits on a new account covers the first one outright.

When to pick a different model instead

GPT Image 1.5 I2I is a specialist, and three situations should send you elsewhere on VdoBloom:

• You need 4K or an unusual frame. Go to GPT Image 2 image-to-image, which adds 1K/2K/4K output and eleven aspect ratios including auto — and starts at 3 credits, cheaper than 1.5 at every tier.
• Your subject is swimwear, dance or fitness content. GPT Image 1.5 sits in the strict content tier because it inherits OpenAI’s filters. VdoBloom’s flexible-tier editors handle that material; explicit and illegal content stays blocked everywhere.
• You are doing bulk variations. At a flat 10 credits, a 50-image run is 500 credits. Cheaper editors on the platform cost a fraction of that per image for work that does not need literal instruction-following.

Run it from the AI image editor, where the credit cost is shown before every generation. If you want to build a source image first, start in text-to-image and bring the result across. Full plan and credit-pack details are on the pricing page.

Frequently asked questions

How much does one GPT Image 1.5 image-to-image edit cost?

10 credits per image, flat, whether you choose medium or high quality. At the Lite plan rate of $15 for 300 credits, that is $0.50 per edit. One-time credit packs start at $2.49 for 75 credits and never expire.

Does the high quality setting cost more credits than medium?

No. Both settings bill at the same 10 credits, because this model is priced per image rather than per resolution tier. Use high by default.

How many reference images can I attach?

Up to 10 per job, and they are treated as joint context for a single instruction rather than a batch queue. Name them in the prompt so the model knows what each one is for.

Why are there only three aspect ratios?

1:1, 2:3 and 3:2 are the frames this model natively supports, with 3:2 as the default. If you need 16:9, 21:9 or a source-matching auto frame, GPT Image 2 image-to-image offers eleven ratios including auto.

Will it refuse my prompt?

It can. GPT Image 1.5 carries VdoBloom’s strictest content label because it runs behind OpenAI’s provider filters, so edits involving real people, violence or suggestive material are declined more often than on the platform’s flexible-tier models. If your work is fashion, swimwear or dance, pick a flexible-tier editor instead.

Can I download the result without a watermark?

Paid plans download watermark-free with commercial rights. Free accounts get 10 credits with no card, enough for one GPT Image 1.5 edit, to test output quality before subscribing.