Text-to-image
Start without references. Describe subject, setting, composition, light, materials, and delivery intent; the adapter sends that prompt with size and quality.
Create a new visual from text or prepare a controlled edit with up to ten reference images in the GPT Image 1.5 workspace.
Loading generator…
Model overview
The GPT Image 1.5 AI Image Generator in ZMS AI prepares text creation or reference-led edits with prompt, framing, quality, and fidelity controls in one reviewable request.

The local driver and estimate are configured, but ZMS has not verified a signed-in task, returned media, or final debit for this route.
OpenAI now marks GPT Image 1.5 deprecated and calls it the previous image model. This page documents the independent ZMS controls; compare the GPT Image family before choosing a route.
Current controls
Use only controls visible in this workspace; adjacent API features may not exist on this route.
Start without references. Describe subject, setting, composition, light, materials, and delivery intent; the adapter sends that prompt with size and quality.
Add 1–10 images for editing. Name each reference role and protected feature; input fidelity applies only on this path.
Choose 1:1 at 1024×1024, 2:3 at 1024×1536, or 3:2 at 1536×1024 before composing.
Choose low, medium, or high quality; reference edits add low or high input fidelity. Current output and debit differences remain unverified.
Creation and edit recipes
These original suggested briefs cover two text tasks and two reference edits. Try copies the complete prompt without changing mode; supply authorized references yourself.
Two creation briefs and two reference-led edit briefs
4 starter recipesProtect product geometry, negative space, and placement-specific light.
Create a 3:2 editorial product photograph of one unbranded brushed-aluminum desk radio on a pale limestone plinth. Preserve a compact rectangular silhouette, two round controls, evenly spaced front grille slots, and four small rubber feet. Place the radio in the left third with wide clean negative space on the right for future copy. Use soft north-window light from camera left, a restrained warm-gray palette, subtle metal reflections, and a waist-height three-quarter camera angle. No hands, duplicate product, readable logo, extra controls, warped grille, floating shadow, text, or decorative props. Keep every edge straight and the complete product inside frame.
Check silhouette, controls, grille, feet, reflections, copy space, edges, and crop.
Keep one subject and action readable without crowding.
Create a vertical 2:3 editorial photograph of an adult ceramic artist glazing one small bowl at a daylight studio table. Show both hands clearly: the left hand steadies the bowl while the right holds a broad brush above the rim. Keep the same bowl, brush, apron, table edge, and window geometry throughout the composition. Use diffused morning light from a tall window on the right, muted clay and indigo colors, realistic skin and fabric texture, and shallow depth of field that leaves the hands and bowl sharp. Reserve calm space above the subject for a magazine title. No extra fingers, duplicate tools, illegible labels, dramatic smoke, cropped hands, or background people.
Inspect hands, brush contact, bowl, apron, window lines, title space, and object count.
Separate identity and environment references, then name protected facts.
Edit the supplied images into one 2:3 editorial portrait. Use reference one only for the consenting adult subject's face, hair, skin tone, and navy linen shirt. Use reference two only for the pale library interior, tall window proportions, and soft afternoon light. Place the subject seated at the reading table, three-quarter view, looking toward the window while holding a closed notebook with both hands. Preserve facial proportions, hairstyle, shirt collar, hand count, notebook shape, window grid, and the direction of light. Do not blend identities, copy text from either reference, add jewelry, change age, alter ethnicity, redesign the room, or invent other people. Keep the pose anatomically natural and the background subordinate.
Compare identity, clothing, hands, notebook, window, and light with their assigned sources.
Give geometry, material, and layout one approved source each.
Create one 3:2 campaign still from the supplied authorized references. Reference one controls the exact compact table-lamp silhouette, switch, cable exit, shade angle, and proportions. Reference two controls the brushed brass material and fine directional grain. Reference three controls the pale oak desk, cream wall, and open negative space on the right. Keep the lamp in the left third at eye level and turn it on with a warm pool of light while preserving the product geometry exactly. Do not copy logos, add hands, duplicate the lamp, change the cable route, bend the stem, alter the switch, introduce extra props, or place text inside the image. Use clean editorial lighting and straight architectural lines.
Audit silhouette, switch, cable, shade, brass, layout, copy space, and lamp count.
Interface boundary
This workspace omits output format, transparency, masking, compression, and exact output count. Plan only with controls users can set here.
| Control | Workspace status | Planning impact | Alternative |
|---|---|---|---|
| Output format | Not exposed | Format cannot be promised before review | Inspect the returned file first |
| Transparency | Not exposed | Transparency cannot be requested | Add it in a later editing step |
| Mask editor | Not exposed | Masked replacement is unavailable | Use full-frame reference edits |
| Compression and output count | Not exposed | Compression and count are not settings | Use visible ratio and quality controls |

Editorial planning illustration
Protect identity and geometry, then vary only crop, environment, light, or material.
Reference-role planning artwork for ZMS AI; not GPT Image 1.5 output or proof of live delivery.Practical workflow
Keep the first attempt narrow and settings stable.
Name placement, subject, purpose, negative space, and ratio before composing.
Use text for new work; add and label references for controlled edits.
Set quality and, for edits, fidelity; keep both stable during prompt revisions.
Inspect full-size media for anatomy, identity, text, geometry, dimensions, and protected regions.
Lifecycle and service
Model lifecycle, workspace controls, and delivery are separate decisions. Use this ledger to confirm which route, controls, account test, media review, rights check, and debit evidence are required before approving a production batch on ZMS AI.
| Decision | Current fact | Review action |
|---|---|---|
| Choose the model route | OpenAI marks GPT Image 1.5 deprecated and calls it the previous image model. ZMS documents an independent configured route, not a direct OpenAI API connection. | Compare lifecycle and controls in the GPT Image family or AI image workspace before selecting another route. |
| Prepare the request | ZMS exposes text, 1–10 references, three ratios, three quality levels, and edit fidelity, but not masks, transparency, compression, or output format. | Assign one role per authorized reference, protect identity and geometry, and select the final crop before submission. |
| Approve production use | Signed-in delivery, returned media, current service behavior, and final debit remain unverified; the illustration is decor, not model evidence. | Test one account task, retain prompt and settings, inspect the full-size file, confirm rights, and check ZMS AI pricing. |
FAQ
Answers about inputs, dimensions, fidelity, lifecycle, and delivery status.
It prepares text creation or edits with 1–10 references, three ratios, three quality levels, and edit-only fidelity.
ZMS maps 1:1 to 1024×1024, 2:3 to 1024×1536, and 3:2 to 1536×1024; live output remains unverified.
Add 1–10 references, give each one a role, and remove conflicts between preserve and replace instructions.
Low or high fidelity appears only with references. Compare returned media with every protected source feature.
No. OpenAI calls it the previous image model and marks it deprecated; this page documents the current ZMS route.
The client estimates by ratio and quality for creation, and quality for edits. The signed-in debit remains authoritative.