GPT-Image 1.5 is OpenAI's image model — strong prompt adherence, clean text rendering, three quality tiers, and a transparent-background mode that's rare among image models. Best when you need the prompt followed literally instead of reinterpreted.
GPT-Image 1.5 is OpenAI's image generation model. Its defining trait is prompt adherence — what you write is what you get, not what the model thinks you meant. That makes it the go-to choice when you need exact compositions, specific text rendering, or unusual subject combinations that other models tend to smooth over.
Three quality tiers (Low / Medium / High) trade speed for fidelity, three frame sizes (square 1:1, landscape 3:2, portrait 2:3), and a transparent-background mode that ships clean PNGs with no extra cutout step. Pricing scales with quality: 1 credit for low and medium, 4 credits for high.
3
Quality tiers
PNG
Transparent backgrounds
1–4 cr
Per image
GPT-Image 1.5 is one of the best general-purpose models for in-image text — chalkboards, signs, packaging, posters. Wrap exact strings in double quotes and the model holds the letterforms. Multi-line text and varied weights work better here than on most photo-first models.
Set background to 'transparent' and GPT-Image returns a clean PNG with no cutout pass needed — ready to drop into a layout or composite. Especially strong for logos, product icons, and isolated subjects you'd otherwise have to mask in post.
GPT-Image 1.5 commits to non-photo styles when you commit to them in the prompt — flat color blocking, watercolor, editorial-magazine illustration, retro print. Other strong general models default photoreal regardless; this one switches genres cleanly.
High-quality tier produces strong photoreal output with crisp texture detail — moss, dew, fabric weave, skin pores. Worth the 4-credit cost for hero shots and material-driven scenes. Low and medium are perfect for prompt iteration.
Each example shows the exact prompt that produced the result. Copy any prompt with one click.
1:1 · High quality · 4 credits
A vintage chalkboard coffee shop sign reading "Daily Roast — Espresso · Pour Over · Open 7am–3pm" in clean hand-lettered serif chalk, faint smudges, photographed straight-on against a warm brick wall
Wrap the exact text in double quotes — GPT-Image 1.5 reads them as literal. Multi-line text holds up better here than on FLUX or Seedream. Add "hand-lettered" or "kerned" to nudge typography quality.
1:1 · High quality · 4 credits
Extreme macro photograph of a single dewdrop balanced on a green moss tip at sunrise, refracted light through the droplet, shallow depth of field, hyper-real detail
High quality tier earns its 4-credit cost on material-driven macros. Name the specific phenomenon ("refracted light", "shallow DOF") instead of generic "photoreal" and the model commits harder to the technique.
2:3 portrait · High · 4 credits
Editorial illustration of a woman in a wool coat reading a paperback on a park bench in autumn, flat color blocking with subtle grain texture, warm rust orange and burgundy palette, soft afternoon light, in the style of a New Yorker cover
Name the medium ("flat color blocking", "grain texture") and the tradition ("New Yorker cover") explicitly. GPT-Image commits to illustrated styles cleanly when the prompt does too — vague "stylized" produces photo-illustration hybrids.
1:1 · High · transparent · 4 credits
Minimalist logo of a wolf head wearing aviator goggles, geometric single-line black ink art, balanced negative space, scalable mark for a coffee brand
Pair background="transparent" with logo-friendly language — 'single-line', 'geometric', 'scalable mark'. The PNG returns with a clean alpha channel ready to composite. Skip the cutout pass entirely.
3:2 landscape · High · 4 credits
Wide shot of a Tokyo backstreet at night, rain-slick pavement reflecting glowing izakaya signs, soft volumetric mist drifting past a noodle vendor, anamorphic lens flare, 35mm cinematic film photography
3:2 landscape rewards cinematic compositions with strong foreground/background depth. Anchor with camera language ("anamorphic lens flare", "35mm film") — GPT-Image reads photographic vocabulary literally.
Medium quality is 1 credit and good enough for 80% of prompt iteration. Switch to high (4 credits) only for the final render where material detail or text fidelity really matters. Low is fine for moodboarding.
Wrap exact strings in double quotes — 'a sign reading "Daily Roast"' beats 'a sign that says Daily Roast'. GPT-Image 1.5 treats the quoted content as a literal target and holds the letterforms.
Background='transparent' returns a clean PNG with a real alpha channel — not a white background pretending to be transparent. Use it for logos, icons, isolated subjects, and anything you'd composite over a different scene.
For photoreal use camera language (lens, lighting, film stock). For illustration name the medium and tradition. GPT-Image 1.5 switches genres cleanly but won't guess — vague prompts produce middle-ground hybrids.
1:1 for product social and logos. 3:2 landscape for cinematic, environmental, scenic. 2:3 portrait for editorial, fashion, single-subject. The model composes differently per aspect — generating at the right ratio beats cropping later.
GPT-Image's prompt adherence shines on unusual subject combinations and specific compositions. If FLUX or Seedream keep smoothing your weird-but-intentional prompt, GPT-Image 1.5 will usually render it as written.
| Setting | Values | Notes |
|---|---|---|
| Quality | Low · Medium · High | Low/Medium = 1 credit, High = 4 credits per image. |
| Image size | 1024x1024 (1:1) · 1536x1024 (3:2) · 1024x1536 (2:3) | Three fixed sizes. Pick the aspect to match the subject. |
| Background | auto · transparent · opaque | Transparent returns clean PNG with alpha channel — no cutout pass needed. |
| Number of images | 1–4 | Batch up to 4 per call. Each output counts as its own credit charge. |
| Output format | PNG (default) · JPEG · WEBP | PNG required when background=transparent. Otherwise pick by downstream use. |
| Editing | Separate endpoint at /gpt-image-1-5-edit | GPT-Image 1.5 also supports image-to-image edits via a sibling tool. |
GPT-Image's defining trait is literal prompt adherence — what you write is what you get. FLUX has stronger photoreal aesthetic defaults. Seedream has the best out-of-the-box composition from sparse prompts. Pick GPT-Image when you need exact text rendering, transparent backgrounds, or unusual compositions that other models smooth over.