GPT Image 2: AI Image Editor Online

These limits are the ones VdoBloom's service enforces. The two rules in the 'Resolution rules' row are the source of most 'why is my image small?' questions.

You are charged for the resolution you select. Because the auto ratio always renders at 1K, selecting 2K or 4K with auto pays for a size you will not get; set an explicit ratio when you want 2K or 4K. The dollar figures use the Lite plan rate of $0.05 a credit; a one-time 300-credit pack is $19.50 and never expires.

All of these sit in the Image Edit dropdown. GPT Image 1.5 is a flat 10 credits with three aspect ratios and a 3,000-character prompt. GPT Image 2.5 (Flare and Sunburst) keeps GPT Image 2's 3 / 5 / 8 price, raises references to 16 and drops the square-at-4K rule, but has a different ratio list (no 4:5 or 5:4). Nano Banana 2.1 is 3 / 4 / 6 with 14 references. VdoBloom does not rank them on quality; the edit you care about, run on two of them, answers that faster.

Same model generation, same 3 / 5 / 8 credit ladder, same 11 aspect ratios and the same two resolution rules. The difference is the input. The text-to-image model starts from a prompt alone; the image-to-image model needs at least one uploaded photo and accepts up to ten. If you are starting from a blank canvas, use GPT Image 2 text-to-image, then bring its result here for the next change.

The model and its prices are the same everywhere it appears; some tools pick it for you.

Write the edit as numbered changes, then a 'keep' list. For example: '1. Replace the background with a white seamless sweep. 2. Add a soft shadow under the shoe. 3. Set the words SPRING DROP in bold sans-serif top left. Keep the shoe, its laces and its colour exactly as photographed.' Separate clauses are less likely to be dropped than one long sentence.

With several photos, say which is which: 'Image 1 is the product, image 2 is the background plate, image 3 is the lighting reference.' Put text that must appear on the image in quotes and spell it exactly. Choose the ratio first: auto renders at 1K, so for a 2K or 4K deliverable pick 16:9, 4:5 or whichever explicit ratio you need. For a square 4K, there is no option on this model; GPT Image 2.5 allows 1:1 at 4K for the same 8 credits.

GPT Image 2's provider has two rules that VdoBloom passes through: the auto ratio renders at 1K only, and a 1:1 square cannot be 4K. Every other combination is available. The table shows what you get for the settings that cause questions. The charge follows the resolution you select, which is why the auto row says to pick 1K.

Adding or changing lettering is one of the main reasons people pick a GPT Image model for editing. Three habits help. Put the exact words in quotes and spell them as they must appear, including capital letters. Say where the text goes and how big it is relative to the frame ('across the top third, bold sans-serif, white'). And keep the line short: one headline and one small line survive better than a paragraph. If a word comes back misspelled, re-run at 1K with the same prompt before trying 2K or 4K; the cheap draft tells you whether the wording works.

The auto ratio caps output at 1K, and the price follows the resolution you selected, so selecting 2K or 4K with auto costs more for a 1K result. A 1:1 square at 4K is refused. Ten references is the ceiling; GPT Image 2.5 takes 16 and Nano Banana 2.1 takes 14. The ratio list has no 8:1 banner strips and no 27:16 or 9:8.

The STRICT label reflects OpenAI's provider rules: edits involving real people, suggestive content or violence are declined more often than on models labelled HOT, such as Seedream 4.5 Edit. If an edit is refused or fails, VdoBloom refunds the credits automatically. GPT Image 2 is proprietary; you use it through VdoBloom's hosted route, not a ChatGPT login or downloadable weights.