GPT Image 2.5 Flare: AI Image Generator Online

Everything below comes from VdoBloom's GPT Image 2.5 service and credit table. The rule to know before you pick a ratio: 27:16, 16:27, 9:8 and 8:9 are 1K-only. VdoBloom lowers the resolution to 1K for those ratios instead of failing the job.

The dollar columns use the Lite plan rate ($15 a month for 300 credits, $0.05 a credit) and the one-time Lite pack ($19.50 for 300 credits, $0.065 a credit). Pack credits never expire; plan credits reset each billing cycle. The free 10 signup credits run three 1K images (9 credits) or two 2K images (10 credits).

OpenAI shipped GPT Image 2.5 in two variants, and VdoBloom lists each as its own text-to-image and image-to-image model, four in all. In VdoBloom's service code, Flare is the fast, default variant and Sunburst trades latency for edit precision and fidelity. Both accept the same inputs: the same 13 aspect ratios, the same 1K/2K/4K tiers, the same 20,000-character prompt cap.

They also cost the same: 3, 5 or 8 credits by resolution, for both generation and editing. Sunburst is not a premium tier. The practical way to choose is by the job: Flare for everyday generation and fast iteration, Sunburst when you want the precision-focused pass. Because the price is identical, running one prompt on both costs no more than running it twice on one.

All four carry the STRICT label. GPT Image 2.5 Flare has the same price ladder as GPT Image 2 but fewer resolution rules: GPT Image 2 forces 1K on the auto ratio and cannot render 1:1 at 4K, while 2.5 has neither rule and limits only four uncommon ratios to 1K.

The table maps common text-to-image jobs to a ratio and resolution the model accepts, and the credit cost of that combination. Rows with a 1K-only ratio are priced at 1K, because that is the tier to select for them.

Product poster (2:3, 4K, 8 credits): "Vertical poster. Top third: the headline 'SUMMER DROP' in bold condensed sans-serif, reproduce exactly. Middle: a pair of white sneakers floating above a pastel orange gradient, soft shadow below. Bottom: small text 'Available June 1', reproduce exactly. Clean, minimal, no other text."

Blog header (16:9, 2K, 5 credits): "Wide illustration for a blog header about budgeting. A tidy desk seen from above: a notebook with a hand-drawn bar chart, a calculator, a coffee cup at the top right. Flat vector style, three colours: navy, cream and coral. Leave the left third empty for a title."

Infographic panel (3:4, 2K, 5 credits): "Infographic titled 'How a seed grows', reproduce the title exactly. Four numbered steps from top to bottom, each with a simple icon and a two-word label: Seed, Root, Sprout, Plant. Even spacing, white background, green accents."

Each one states the output type and frame first, places every element, quotes every word of text, and limits the palette. That pattern is what the long prompt cap is for.

Flare is the model many of VdoBloom's image tools are built on. Inside those tools the prompt, ratio and resolution are set for you; most run at 2K (5 credits an image).

Flare follows literal instructions, so write the brief as a specification. Start with the output type and frame ("A 16:9 YouTube banner"), then the layout (what sits left, centre, right), then the style, then a short list of things to avoid. Give exact counts and positions rather than adjectives.

Put any text that must appear in quotation marks and say "reproduce exactly"; keep it to a few words per line. The 20,000-character cap is generous, but one clear paragraph usually beats a page of mood words.

Pick the ratio yourself when the output must be 2K or 4K. If you choose 27:16, 16:27, 9:8 or 8:9, select 1K: those ratios are always rendered at 1K, and the job is billed at the resolution you select. Draft at 1K for 3 credits, then rerun the winner at the size you need.

The 1K-only ratios are the main trap. VdoBloom lowers the resolution to 1K for 27:16, 16:27, 9:8 and 8:9 rather than rejecting the job, and you are charged the 1K price (3 credits) for it, so a 2K or 4K selection on those ratios costs the same as 1K. Pick a standard ratio such as 16:9 when you need 2K or 4K.

This model only generates. To change an existing photo, use the Flare image-to-image model. It does not offer the 1:4 to 8:1 strip ratios of the Nano Banana 2 family or 5:4 and 4:5, which GPT Image 2 has.

The STRICT label means OpenAI's filters refuse restricted subjects; a refused or failed job is refunded. Output quality depends heavily on prompt clarity, so a vague one-liner can look soft or generic. GPT Image 2.5 is proprietary and reached through VdoBloom's hosted route, with no downloadable weights.