GPT Image 1.5 was tested here for the jobs that make image generation useful in production: product photography, short readable text, packaging layouts, reflective materials, illustration style, and a constrained image edit. Each GPT Image 1.5 test used one output at 3:2, high quality, JPEG with compression 90, low moderation, and one sample. The edit used a supplied watch image and a direct instruction to preserve the product. The model accepts text prompts, optional source images and masks, 1:1, 3:2, or 2:3 output sizes, three quality settings, PNG/JPEG/WEBP output, up to 10 samples, and high or low input fidelity for edits.
Model and setup
GPT Image 1.5 on Wiro covers both text-to-image generation and image editing. The six prompts were deliberately specific. That makes the results more useful than a broad request for a pretty image: each one asks the model to bind materials, light, text, placement, or preservation rules to a visible outcome.
| Setting | Value used |
|---|---|
| Canvas | 3:2 (1536 x 1024) |
| Quality | High |
| Format | JPEG, compression 90 |
| Outputs per prompt | 1 |
| Moderation | Low |
| Edit fidelity | High for the watch edit |
The recorded completion times range from 42 to 55 seconds. These are observations from these six Wiro runs, not a service guarantee. No per-output price was recorded with the original runs, and the model documentation reviewed for this update does not state a Wiro per-image price. Rather than infer a cost from another provider or a different setting, the table marks it as unavailable.
What the six GPT Image 1.5 tests show
1. Macro product shot: materials and lighting
The coffee-mug prompt asks for wet slate, an espresso drip, small ripples, a left softbox, a right rim light, sharp specular highlights, and an 85mm look. The output completed in 51 seconds. It holds the studio-light direction together: the mug remains matte while the puddle and wet slate carry the strongest highlights. The useful result is not just a dramatic coffee photo. It shows that GPT Image 1.5 can separate materials when the prompt names the lighting role and surface behavior.
2. Chalkboard menu: short text under texture
The menu test asks for four exact short strings: DRIP COFFEE 3, OAT LATTE 5, CROISSANT 4, and OPEN 7AM. It returned in 45 seconds. The lettering is readable at article size and sits convincingly on a chalkboard in a real-looking cafe scene. That is the right scope for the result: short lines, strong contrast, and a loose handwritten style. It should not be treated as final artwork for a dense menu, legal copy, or a price list that cannot tolerate a single character error. Check every character before shipping.
3. Packaging mockup: text and straight edges
The protein-bar wrapper puts three pieces of copy on a small surface: MOUNTAIN FUEL, Cocoa + Sea Salt, and 12g protein. It completed in 45 seconds. The output reads as a plausible studio mockup and keeps the wrapper edges straighter than a busy collage prompt usually would. The lesson is practical. Ask for a limited hierarchy, give the exact strings, and leave generous empty space. For regulated packaging or final brand files, use this as a concept image and rebuild approved copy in a design tool.
4. Glass and mirror: reflection consistency
The iced-tea scene combines a transparent glass, condensation, lemon, a round mirror, afternoon sunlight, table reflections, and a blurred street. It was the slowest generation at 55 seconds. The output keeps a coherent light direction and gives the glass, mirror, and wood distinct highlight behavior. It is a good prompt pattern for lifestyle product concepts. It is less suitable when a campaign needs exact product geometry, a specific label, or physically measured reflections.
5. Children’s illustration: style control
The raccoon astronaut test asks for clean ink lines, watercolor fills, warm light, readable shapes, and a simple background. It completed in 54 seconds. The output shows why style instructions work best when they describe both the mark-making and the scene complexity. The restrained background gives the character and pancakes room to read. This is a solid choice for a first visual direction, story pitch, or social graphic. It still needs a human art pass if a series must maintain exact characters across many images.
6. Watch edit: preserve the subject, replace the scene
The final test starts with a watch photo. The instruction changes only the background, removes water droplets, adds a soft shadow, and explicitly protects the watch shape, text, hands, and bezel. It finished in 42 seconds, the fastest result in the set. The edited output makes the watch presentation cleaner without turning it into a different object. That is the strongest use case in this group: give the model a clear source image, say what must remain untouched, then list the limited changes.
 |
 |
Timing and cost record
| Test | What it checks | Recorded time | Recorded cost |
|---|---|---|---|
| 1 | Product lighting and wet materials | 51 seconds | Not recorded |
| 2 | Short chalkboard text | 45 seconds | Not recorded |
| 3 | Packaging hierarchy and text | 45 seconds | Not recorded |
| 4 | Glass, mirror, and reflections | 55 seconds | Not recorded |
| 5 | Illustration style | 54 seconds | Not recorded |
| 6 | Constrained product edit | 42 seconds | Not recorded |
When to choose GPT Image 1.5
Choose GPT Image 1.5 when the brief benefits from natural-language control and the output can be reviewed before publication. It fits product concepts, promotional scenes, short display text, early packaging directions, and image edits where the subject must survive a background change. Use high quality when finish matters and leave time for a roughly one-minute result. Use a source image plus explicit preservation rules for edits. Keep exact text short, high contrast, and separate from busy imagery.
Pick another workflow when pixel-perfect typography, approved product dimensions, or repeatable character identity is non-negotiable. Those cases need a design system, source assets, or manual compositing after generation. For more Wiro tests, see GPT Image 1.5: 5 Prompts for Clean Layouts, 8 Prompts for Product Photography with GPT Image 1.5, and FireRed Image Edit vs Seedream V5 Lite.
Sources
OpenAI describes image generation and editing controls in its official image generation documentation. Its image generation announcement also discusses text rendering, instruction following, multi-turn work, and known limitations.
Run GPT Image 1.5
Run GPT Image 1.5 on Wiro for a text-to-image prompt or a source-image edit.
