GPT Image 2.5 Review: Sunburst vs Flare for Reference-Led Fashion
GPT Image 2.5 is now available in Astria in two variants: Sunburst and Flare. Both accept image references and expose 1K, 2K, and 4K output settings plus low-to-max quality controls. In our first matched fashion test, both produced polished, reference-led campaign frames. Sunburst pushed the diagonal pose and direct-flash tension harder; Flare kept the complete bag more legible.
That is a first production read, not a universal ranking. We ran one controlled brief with one output per model so you can inspect the actual difference without a hand-picked winner.
GPT Image 2.5 at a glance
| Variant | Start here when… | Tested output |
|---|---|---|
| Sunburst | The art direction depends on energetic posing, hard flash, and a stronger editorial crop | 4K request; 3792×2160 returned |
| Flare | The product and accessory need a calmer, more complete campaign frame | 4K request; 3792×2160 returned |
Both variants support references, 1K/2K/4K selection, and low, medium, high, xhigh, and max quality settings in the current Astria catalog. Browse the live Sunburst gallery and Flare gallery.
The fixed reference set
This is a rerun of the semantic brief from Maison Urban prompt 46557199. For publication, the same Sloane portrait, yellow sleeveless belted dress, and circular-detail gold bag were copied into Astria workspace 896 (Articles). The source prompt wording was preserved apart from the new Articles-workspace reference IDs.
The dress and bag references exceed 2K on the long edge. The supplied Sloane portrait is only 679×722, below our preferred source size, and we did not upscale it. Treat the identity result as a practical reference test, not a high-resolution identity benchmark.
Prompt
Dress hero shot. An off-kilter, flash-lit photograph of the referenced woman in a dynamic, almost off-balance leaning pose against a dark wooden bar counter or pillar at night. Her body creates a diagonal line. She wears the reference dress and holds the reference bag. The strong flash casts a sharp shadow. Blurred shelves with bottles and bar lights define the out-of-focus background. Cool, edgy vibe.
The model-specific reference tokens are omitted above for readability. The generation records use the exact Articles-workspace tune IDs.
Sunburst vs Flare
Both outputs pass the article-size visual gate and preserve the central reference relationships: Sloane remains recognizable, the dress stays pale yellow with the sleeveless belted construction and two gold buttons, and the bag retains its gold chain and repeated circular metal texture.
Neither result is a literal product reconstruction. Pleat placement, the belt wrap, individual bag discs, and crop differ from the supplied images. For campaign work, those shifts may be acceptable. For a PDP or catalog asset, compare every construction detail against the product before approval.
What the variants do differently
Sunburst is the more aggressive editorial interpretation in this pair. The body line is strongly diagonal, the flash contrast is harder, and the crop feels closer to a magazine campaign. The tradeoff is that the bag reaches the frame edge.
Flare is more product-complete. It keeps the bag fully visible, gives the dress more breathing room, and still holds the direct-flash nightlife brief. Its pose is less precarious and therefore slightly less surprising.
Our routing rule from this one brief is simple:
- start with Sunburst when shot energy and art direction are the approval risk;
- start with Flare when garment and accessory visibility are the approval risk;
- run both when one extra comparison is cheaper than guessing which interpretation the client meant.
The same brief in three other models
We also ran the identical Articles-workspace references, semantic prompt, 16:9 request, and one-output count through Nano Banana 2, MAI-Image-2.6-Flash, and Muse. The resolution switch stayed at 4K where the endpoint supports it. MAI and Muse do not expose the same 4K selector; that incompatibility is recorded instead of silently changing the GPT Image 2.5 settings.
The wider context does not produce a new universal winner. Nano Banana 2 remains a strong general fashion default. Muse keeps the product relationships coherent but misses the requested canvas ratio. MAI Flash gives a usable campaign frame at lower output resolution. The new GPT Image 2.5 variants are immediately competitive when a polished, flash-led fashion treatment is the goal.
Production checklist
- Use the same references in both GPT Image 2.5 variants.
- Keep aspect ratio, resolution, quality, prompt, and output count fixed.
- Review identity, garment construction, and accessory geometry separately.
- Choose Sunburst or Flare from the job's approval risk, not from the model name.
- Recheck the live Astria pricing page and model selector before building a production budget.
Bottom line
GPT Image 2.5 arrives as two useful reference-led routes rather than one vague upgrade. Sunburst is our first test for energetic campaign direction; Flare is our first test for a steadier frame with clearer product visibility. Both produced sharp, commercially credible 4K fashion outputs from the same modest identity source and the same dress-and-bag references.
See the full fashion model comparison, or start directly in the Sunburst and Flare galleries.