Choose Flare for everyday image work or explore Sunburst for detailed edits. Bring references, describe the composition, and keep refining in the same chat.
Opens Clico Web in a new tab. Review your prompt and settings before sending.
A convincing picture is only the starting point. For product work, the harder question is whether you can change the setting while keeping the bottle, label and camera angle recognizable. That is the useful way to evaluate Image 2.5.
Compare the settings you can actually select in Clico.
Flare is positioned for faster everyday generation. Sunburst is positioned for detailed creative work. Both expose the same controls in Clico. Image 2 remains available with three quality choices; 2.5 adds Extra high and Max. Extra settings alone do not establish better results.
| Model | Output size | Quality | Image inputs | Aspect ratios |
|---|---|---|---|---|
| GPT Image 2.5 Flare | Model managed | Low, Medium, High, Extra high, Max | Up to 5 references | 1:1, 3:2, 2:3, 4:3, 3:4, 16:9, 9:16, 21:9 |
| GPT Image 2.5 Sunburst | Model managed | Low, Medium, High, Extra high, Max | Up to 5 references | 1:1, 3:2, 2:3, 4:3, 3:4, 16:9, 9:16, 21:9 |
| GPT Image 2 | Model managed | Low, Medium, High | Up to 5 references | 1:1, 3:2, 2:3, 4:3, 3:4, 16:9, 9:16, 21:9 |
These are Clico controls. A provider may expose additional features on its own platform. Availability depends on your account and the generation provider.
The reported upgrade focuses on more controlled edits and faster iteration. Spatial precision still needs to be judged in the image.
| Question | Reported change in 2.5 | What that means for your work |
|---|---|---|
| Will it follow a targeted edit? | The launch report describes better preservation of composition, lighting and subjects outside the requested change. | Name the detail to change and the details to preserve. Our Flare examples keep the label recognizable, but reflections and textures still change. |
| Will revisions stay consistent? | The report describes improved consistency through repeated edits. | Keep the version you want as the reference. This page does not provide a controlled multi turn comparison with Image 2. |
| Is spatial layout more controllable? | The report shows sketch guided composition as a way to communicate relative positions and proportions. | Use a reference to specify layout. Inspect perspective, object overlap and contact shadows; a sketch example does not establish a spatial benchmark score. |
| Is it faster? | The report relays OpenAI’s claim of up to 50% lower latency than Images 2.0. | That is an attributed launch claim, not a Clico timing guarantee. Flare prioritizes iteration; Sunburst prioritizes detailed edits. |
Source: New Intelligence’s Image 2.5 launch report on 36Kr. These qualitative changes are reported by the source. The settings comparison above is verified against Clico’s model catalog.
Three recorded edits of the same product photograph. Look at the label, contact shadow and reflected light as the setting changes.
The desert scene was created and FIELD 07 remained readable. The light changed the glass highlights and warmed the cap. This is a creative edit, not a pixel preserving cutout.
Put this perfume bottle in a dramatic sunlit desert scene with pink sand and a blue sky.
The terracotta setting replaced the gray studio. The bottle, label and flower remained recognizable. Reflections and surface texture still changed.
Edit the supplied photograph. Replace only the gray background and stone surface with warm terracotta plaster and soft palm leaf shadows. Keep the bottle in the exact same position, size and three quarter angle. Preserve its emerald green glass, silver cylindrical cap, cream label and the exact black text FIELD and 07. Keep the orange flower beside the bottle. Do not redesign, reshape, rotate or relabel the bottle. Keep the original landscape framing.
The flower and stem disappeared, and the stone surface filled the space. The bottle label stayed readable. Inspect the repaired texture before using it.
Edit this image by removing only the orange flower and its green stem. Fill that area with the matching stone surface. Preserve the perfume bottle, FIELD 07 label, glass color, cap, light, shadows and background. Keep the same camera framing.
These are existing local provider API experiments using the Clico request shape. They are not an authenticated Clico test or a controlled comparison with Image 2. View prompts, settings and output records.
Describe the foreground, the surface that supports the subject, and what sits behind it. Then inspect contact shadows, reflections and overlap. These are practical checks for spatial coherence, not a numerical claim that 2.5 beats 2.0.
These works illustrate composition and creative direction. They are not verified outputs from GPT Image 2.5.
Create a landscape product photograph. A cobalt glass vase stands on a pale stone table in the foreground. One peach blossom rises from the vase. A window is behind the table on the left. Its reflection follows the curved glass. Keep the table edge continuous behind the vase. Leave the right third empty. No text.
The model, references and actual output usage determine the charge. Quality is not a fixed price per picture.
Image generation uses the selected model’s metering rates. The same rate can still produce a different bill when the request uses more tokens. Reference images can add input usage, and generating multiple images adds output usage.
The table shows the Clico metering rates reviewed on September 10, 2026, in USD per million billing tokens. It is not a price per image or per second. Your account balance and current plan determine access.
| Model | Text input | Image input | Image output |
|---|---|---|---|
| GPT Image 2.5 Flare | $5.00 | $8.00 | $39.00 |
| GPT Image 2.5 Sunburst | $5.00 | $8.00 | $39.00 |
| GPT Image 2 | $5.00 | $8.00 | $39.00 |
Image 2, Flare and Sunburst each use $39 per million image output tokens in the reviewed rate table. This example excludes text, reference inputs and cache adjustments. It does not estimate tokens for a requested image.
Compare Clico plans and credits. For upstream pricing, consult the linked model sources below; upstream list prices and Clico usage charges are different.
Model positioning comes from the linked provider listings. Clico’s exposed controls are listed above.
New Intelligence’s September 9 report, republished by 36Kr, describes improved targeted edits, multi turn consistency and sketch guided composition. It relays partner feedback about preserving the parts of an image that should not change. Those are reported observations, not Clico benchmark results.
For spatial control, the report’s room sketch example is relevant: a sketch communicates relative positions and proportions. In Clico, use a supported uploaded image as a reference. This page does not claim that ChatGPT’s Sketch tool is part of Clico.
In the indexed excerpt of a Reddit launch discussion, a commenter reports a checkerboard pattern behind text. That is a useful inspection target: zoom in on lettering and broad, even areas before using an image in a layout. A single comment does not establish the cause or how often it occurs.
Access scope: publicly indexed post and comment excerpts, not the full discussion or its original image files. We have not verified the commenter’s explanation.OpenRouter positions Flare for everyday generation, creator content and rapid prototyping, with speed as its main distinction within Image 2.5.
Provider listing read September 10, 2026. Provider features and pricing can differ from Clico.OpenRouter positions Sunburst for detailed creative work where editing accuracy matters more than generation speed. The listing does not supply a Clico spatial reasoning benchmark.
Provider listing read September 10, 2026. Provider features and pricing can differ from Clico.No. Prepare your prompt and settings, continue to Clico Web, sign in if needed, then review and explicitly send the request.
Use supported PNG, JPEG or WebP references within the selected model’s limits. Each image can be up to 8 MB. State what to preserve and what to change. An instruction does not guarantee that every unmentioned pixel remains identical.
The prompt, references, model and settings all affect the output. Gallery references are not verified model tests. The recorded experiments on this site describe their own settings and scope.