GPT Image 2.5 Flare: Create and Edit a Cafe Promo Poster

GPT Image 2.5 Flare: Create and Edit a Cafe Promo Poster

The drink looks right, the colors work, and the schedule is already approved. Then someone asks for a different headline. Regenerating a promotional poster from scratch can change all the parts you wanted to keep.

We tested a more focused workflow with GPT Image 2.5 Flare: create a café poster, then use that image as the reference for a headline-only edit. The revised image retained the café name, schedule, and a visually similar latte scene. It also added an unrequested full stop—a small but useful reminder that text QA includes punctuation.

The poster brief: approve the words before the image

Our fictional business, MOSS CAFE, needed a weekend iced-coffee promotion. We supplied four text lines, a cream-and-green palette, and a clear composition: brand at the top, headline in the upper half, schedule underneath, and a glass of iced latte below.

The initial brief deliberately excluded prices, discounts, addresses, and extra slogans. These details would need real business approval; asking for a complete-looking advertisement should not invite the model to invent them.

The copy was:

• MOSS CAFE
• WEEKEND COFFEE CLUB
• SAT + SUN
• 9 AM - 2 PM

All business details are illustrative. Both images below are AI-generated, not photographs of an actual café's product or evidence of campaign performance.

Our Flare test settings

We ran the initial generation and one edit on October 12, 2026 using the WeShop gpt-image v1.0 agent through its official API. The first request used text only; the edit used the first output as its single reference image.

The delivered dimensions were close to, but not exactly, 4:5. Check the actual export dimensions and resize for your placement before delivery.

OpenAI describes Flare as the speed-oriented option in the GPT Image 2.5 family. Its image prompting guidance recommends specifying the intended image and separating what should change from what must remain. This case evaluates the resulting poster and revision; it is not a latency benchmark or a quality comparison with Sunburst.

WeShop's model guide confirms the Flare model selection. Specify it explicitly: leaving the model unset in this agent defaults to GPT Image 2.

Generate a design with a readable hierarchy

The initial prompt gave the model a delivery format, subject, visual treatment, text hierarchy, and a list of prohibited additions. It asked for a finished poster image rather than a poster photographed on a wall.

Create a polished vertical 4:5 promotional poster for a fictional independent cafe named MOSS CAFE. Warm cream background, deep forest green typography, subtle paper texture, modern editorial design. A realistic clear glass of iced latte with ice cubes and layered espresso and milk sits in the lower center on a simple dark green counter, soft side lighting. Leave generous margins and strong readable hierarchy. Include only these exact text lines, each once: MOSS CAFE; WEEKEND COFFEE CLUB; SAT + SUN; 9 AM - 2 PM. Brand name small at top, headline large in the upper half, schedule smaller below headline. No prices, discount claims, address, additional text, logos, or watermarks. The final image should look like a finished social feed poster, not a poster photographed in a room.

The result placed all four requested text lines correctly, without adding prices or supporting claims. The large headline, smaller schedule, and lower-half drink formed the intended reading order. Inspect this draft before editing: an error in the source can carry forward into the next image.

Replace the headline using the poster as a reference

For the edit, we supplied the initial poster as image 1. We did not start another text-only request. The change was specific: replace WEEKEND COFFEE CLUB with YOUR WEEKEND, ON ICE, while retaining the other text and visual elements.

Edit image 1. Replace only the headline WEEKEND COFFEE CLUB with the exact words YOUR WEEKEND, ON ICE. Show the new headline once in the same upper-half headline area, using the same dark green typography and a balanced line break. Preserve MOSS CAFE, SAT + SUN, and 9 AM - 2 PM exactly. Preserve the latte glass, ice cubes, drink layers, counter, background color, paper texture, lighting, margins, aspect ratio, and overall layout. Do not add any other text, prices, claims, logos, or objects.

EDIT INPUT — INITIAL POSTER

EDIT OUTPUT — NEW HEADLINE

The new headline occupied two balanced lines. The previous headline disappeared, while the brand and schedule remained correct. The glass outline, ice arrangement, coffee-and-milk layers, counter, and overall composition looked highly similar at visual inspection. That is a visual observation, not a claim that the image stayed pixel-identical.

The small error that should stay in your QA checklist

The requested headline ended at ICE; the rendered headline ended at ICE. The words and comma were correct, but the additional full stop means this edit did not meet a strict verbatim-copy requirement.

If punctuation must match approved copy exactly, correct it before use. A follow-up edit can request removal of the final full stop, but that additional correction was not tested here. For locked typography or an urgent final export, replace the text manually in your design file.

Adapt the workflow to a real promotion

Start with the smallest approved brief: business name, headline, dates, and a visual direction. Generate the first draft, then inspect every text line and the subject before accepting it as the next reference.

For subsequent revisions, change one item at a time. State the old content, the replacement, and the elements to preserve. A different headline length may require new line breaks; decide whether that is acceptable before generating. After every edit, recheck both the changed region and the regions meant to remain unchanged.

If you need to show your actual drink, supply an authorized product photo and inspect the glass, ingredients, portions, and appearance against it. That is an adaptation of this workflow, not a reference-product test performed in this case.

The result is a raster image, not a layered layout with editable text objects. Flare is useful here for preparing a visual draft and testing a focused copy revision. Final production still needs accurate business information, the required export dimensions, and a typography check. One successful scene-preservation example does not establish reliability across repeated runs.

Create a poster draft with the WeShop GPT Image tool, then use that draft as the reference for one targeted revision. For the separate model-selection question, see our GPT Image 2.5 Flare vs Sunburst comparison.