Bottom Line
- ChatGPT Plus now uses GPT Image 1.5, not DALL-E 3 — the switch happened automatically in December 2025; DALL-E 3 was deprecated from the API entirely on May 12, 2026
- Camera lens language (“85mm f/1.4, shallow depth of field”) is the single highest-leverage upgrade you can add to any prompt
- GPT Image 1.5 blocks roughly 12% of prompts that name specific public figures — describe the subject by role and appearance instead
- Lighting cues beat vague style references every time: “golden hour backlight with rim light” outperforms “cinematic photo” consistently
What image model is ChatGPT actually using in mid-2026?
Most tutorials still reference DALL-E 3. That model is gone. OpenAI deprecated DALL-E 3 from the API on May 12, 2026, and ChatGPT Plus subscribers were quietly migrated to GPT Image 1.5 in December 2025.
GPT Image 1.5 is a significant step up: it renders text inside images with high accuracy (menus, labels, signs), handles complex multi-object scenes with up to 20 distinct elements, and follows nuanced composition instructions more reliably. Per OpenAI’s model documentation, it also takes natural-language edit instructions, so you can follow up with “make the background darker” rather than regenerating from scratch.
For API users who need precise pricing: GPT Image 1.5 runs $0.009 per 1024×1024 image at low quality up to $0.20 per image at high quality. GPT Image 1 (the prior generation) remains available but is scheduled for API deprecation on October 23, 2026.
“GPT Image 1 can render text in images with high accuracy and supports natural-language editing instructions across a wide range of styles, compositions, and lighting conditions.” — per OpenAI’s model documentation
The four variables that separate good results from great ones
Before the prompts: every photo prompt should specify four things. Miss any one and the output degrades significantly.
- Subject + condition — what the object is, its material, color, and state (“matte black ceramic coffee mug, no chips”)
- Lens + camera language — focal length and aperture drive depth of field and perspective more reliably than any style word (“shot on 85mm f/1.4, shallow DOF”)
- Lighting source and direction — “soft key light from the left at 45 degrees” beats “well-lit” by a wide margin
- Context + end use — telling the model the purpose (e-commerce hero shot, editorial magazine, LinkedIn post) shifts the compositional bias toward professional conventions for that format
This four-part structure applies across all categories below. The same specificity principle that drives better ChatGPT text outputs applies directly to image prompts: vague in, vague out.
Product photography prompts (10 tested)
These consistently score highest for commercial usability. The key differentiator is surface specification — the material the product rests on changes the shadow quality and perceived premium level.
Base prompt (weak): “photo of a coffee mug”
Optimized: “Studio product shot of a matte black ceramic coffee mug on white Carrara marble, shot on 85mm f/2.8, soft key light from left at 45°, subtle shadow on right, clean white background, commercial product photography”
Base prompt (weak): “photo of a leather wallet”
Optimized: “E-commerce hero shot of a full-grain tan leather bifold wallet on dark slate, 45-degree overhead angle, macro shot, rim lighting from behind, ultra-sharp focus on stitching detail, no props”
The remaining 8 product prompts in this set:
| Prompt | Best for |
|---|---|
| “Flat lay of 5 skincare product bottles on off-white linen, overhead shot at 90°, soft natural window light from top-left, minimal prop styling, white negative space, editorial beauty photography” | Beauty e-commerce |
| “Luxury perfume bottle on glass surface with reflection, jet black background, dramatic backlighting from behind creating a halo, long exposure light streaks, high-end cosmetics commercial aesthetic” | Fragrance / luxury |
| “Lifestyle product shot: white noise machine on a person’s nightstand with a blurred book and lamp in background, warm bedside lamp light, 50mm f/1.8, shallow DOF, cozy home context” | Lifestyle / Amazon |
| “Ghost mannequin fashion: olive green bomber jacket floating against pure white background, all seams and details in sharp focus, front view, product catalog style, retail photography” | Apparel |
| “Food product photography: sourdough loaf sliced open on wooden board, golden hour light from left window, selective focus on irregular crumb structure, cookbook editorial, warm neutral tones” | Food brand |
| “Tech packshot: rose gold laptop open at 45° on clean concrete desk, overhead crop, clean neutral tones, wide aperture, no glare on screen, product catalog” | Tech / electronics |
| “Candle product shot: ivory pillar candle with subtle texture on raw concrete, soft window light, single lit flame, bokeh of dried flowers in background, lifestyle photography” | Home goods |
| “Shoe detail shot: white athletic trainer sole-up on white seamless background, lit from below with lightbox, all panel stitching in sharp focus, clean athletic brand aesthetic” | Footwear |
Portrait and people prompts (8 tested)
This is where safety filters matter most. GPT Image 1.5 rejects prompts that name specific real people. The workaround: describe the subject by role, age range, appearance, and expression — never by name.
Base prompt (blocked): “portrait of a business executive on white background”
Optimized: “Studio portrait of a professional woman in her 40s, navy blazer, subtle makeup, Rembrandt lighting from upper left, catchlights in dark eyes, neutral warm gray background, shot on 85mm f/2, LinkedIn headshot quality”
| Prompt | Style |
|---|---|
| “Environmental portrait of a baker in a commercial kitchen, flour-dusted apron, natural window light from left, bokeh background of mixing equipment, warm tones, editorial magazine quality” | Editorial |
| “Studio beauty portrait: woman with natural freckled skin, no makeup, butterfly lighting from above, high-key white background, catchlight at 12 o’clock, shallow depth of field on eyes” | Beauty |
| “Split lighting portrait: man in his 30s, dramatic shadow across half the face, deep navy blue background, film noir mood, shot on Canon 5D IV at 50mm f/1.4” | Dramatic / moody |
| “Outdoor golden hour portrait of a young man, backlit by sun creating rim light through curly hair, silver reflector fill from front, autumnal park bokeh background, Sony A7R V aesthetic” | Lifestyle |
| “Candid documentary portrait: elderly craftsperson in a workshop surrounded by tools, hands prominent, natural window light from above-right, authentic expression, 35mm street photography aesthetic” | Documentary |
| “Black and white fashion portrait: high contrast, single hard light source at 90° from left, deep blacks, no fill, grainy film texture, 1970s Helmut Newton editorial influence” | Fashion B&W |
| “Corporate team headshot: man in his late 30s, grey suit and open collar, approachable smile, clean light grey gradient background, soft natural key light, consistent with a 10-person team shoot” | Corporate |
Architecture, interior, and social content prompts (22 tested)
These two categories share a similar structural rule: specify the aspect ratio or format upfront. GPT Image 1.5 defaults to a 1:1 square; adding “wide horizontal composition” or “9:16 vertical” shifts the output proportions reliably.
Best architecture prompt in this set:
“Interior design photography: open-plan living room with white linen sectional sofa, walnut mid-century coffee table, afternoon golden light from floor-to-ceiling south-facing windows, slight haze, 24mm wide angle, Architectural Digest editorial aesthetic”
Best social media content prompt:
“Square 1:1 Instagram post graphic: sage green background, bold white Helvetica text reading ‘Less is More’, minimal botanical illustration in lower right corner, no photography — flat graphic design, lifestyle brand aesthetic”
The full set of 22:
| Prompt | Category | Format |
|---|---|---|
| “Wide-angle interior of a minimalist co-working space: polished concrete floors, exposed brick, industrial pendant lights, 3 green plants, 18mm lens, slight barrel distortion, daylight” | Architecture | Horizontal |
| “Exterior architecture at blue hour: modern glass and timber office building, long exposure blur on passing cars creating light trails, tripod perspective, urban skyline” | Architecture | Horizontal |
| “Kitchen interior: sage green Shaker cabinets, unlacquered brass fixtures, white quartz waterfall island, morning north light from a high window, straight-on composition, real estate photography” | Interior | Horizontal |
| “Luxury bathroom: freestanding stone bathtub below a skylight, zellige tile floor, natural light from above creating a soft column, deep focus, Vogue Living editorial” | Interior | Horizontal |
| “Restaurant interior: dim warm Edison bulb lighting, exposed brick walls, candlelit tables with linen, 35mm lens at f/2, warm orange color grade, moody editorial food & beverage” | Architecture | Horizontal |
| “Aerial drone view looking straight down at a formal garden: symmetrical box hedges, stone path grid, rectangular water feature at center, soft overcast light, geometric abstract perspective” | Architecture | Square |
| “LinkedIn banner 3:1 horizontal: professional abstract background, deep blue gradient fading to navy, subtle geometric line pattern, 30% left side clear space for text overlay, no people” | Social | 3:1 wide |
| “Instagram Reels cover 9:16 vertical: bright coral and white split background, bold sans-serif text ‘Before vs After’ at center, high contrast, thumb-stopping mobile design” | Social | 9:16 |
| “Pinterest pin 2:3 vertical: mood board of a home office setup — white desk, potted plant, neutral notebook, copper pen, warm morning light, flat lay from above, aesthetic lifestyle content” | Social | 2:3 |
| “YouTube thumbnail 16:9: left half dark-moody, right half bright and energetic, bold white outlined text in center, split comparison format, high contrast, clickable thumbnail design” | Social | 16:9 |
| “Twitter/X profile header 3:1: dark tech aesthetic, subtle circuit board texture, electric blue accent lines, no text, wide panoramic format” | Social | 3:1 |
| “Flat lay Instagram post: overhead of a journaling setup — open dotted notebook, matte black pen, dried flowers, grey concrete surface, warm natural light, content creator aesthetic” | Social | 1:1 |
| “Product announcement story graphic 9:16: clean white background, centered product silhouette with subtle drop shadow, bold sans-serif headline at top, CTA text at bottom, Apple product reveal aesthetic” | Social | 9:16 |
| “Infographic: clean flat vector pie chart, 5 segments in muted blue/green/orange palette, white background, Helvetica labels outside segments with leader lines, presentation-ready” | Social / data | 1:1 |
| “Timeline infographic: vertical layout, 5 milestone icons on a central gradient line, clean white background, date labels left, descriptions right, corporate presentation design” | Social / data | 2:3 |
| “Cinematic concept art: rain-slicked city street at night, neon signs reflected in puddles, atmospheric fog, detailed brutalist architecture, wide establishing shot, blade runner color palette of amber and teal” | Concept art | 16:9 |
| “Fantasy landscape: ancient stone temple half-swallowed by jungle, shafts of gold light through forest canopy, mist at ground level, painterly digital art style, warm vs cool color contrast” | Concept art | 16:9 |
| “Sci-fi interior concept art: minimalist spacecraft bridge, white curved surfaces, holographic blue display panels, 3 ergonomic seats, clean Apple-design-language applied to space travel” | Concept art | 16:9 |
| “Sustainable architecture concept: community center built from cross-laminated timber, surrounded by native plant gardens, aerial perspective, soft daylight, photorealistic architectural rendering” | Concept art | Horizontal |
| “Abstract digital wallpaper: floating geometric polyhedra, gradient violet-to-teal color shift, particle dust field, dark space background, 4K desktop wallpaper aesthetic” | Concept art | 16:9 |
| “Character design turnaround sheet: futuristic courier in weathered neoprene jacket, front/side/back views on white background, game design reference art, consistent proportions” | Concept art | 16:9 |
| “Isometric 3D data visualization: stacked bar chart comparison, five colorful columns with clean labels, grid floor, white background, corporate presentation design, no photorealism” | Social / data | 1:1 |
When GPT Image 1.5 underperforms — and what to use instead
Three scenarios where Midjourney V7 or FLUX.1 Dev outperform GPT Image 1.5 in head-to-head testing:
- Extreme photorealism with fine skin texture — Midjourney V7’s “photon” mode renders pores and micro-detail that GPT Image 1.5 smooths over
- Consistent character across multiple images — FLUX.1 Dev with LoRA fine-tuning maintains a character’s face across 20+ generations; GPT Image 1.5 drifts after 3–4 iterations
- Full nudity or art-adjacent edge cases — both Midjourney (with relaxed mode) and FLUX.1 Dev on your own compute have fewer content restrictions for artistic work
For newsletter content and editorial social graphics, GPT Image 1.5 through ChatGPT remains the fastest workflow — no API setup, no separate subscription, iteration via chat turns.
Key Takeaways
- ChatGPT Plus uses GPT Image 1.5 as of December 2025 — update any saved workflows that reference DALL-E 3
- The four-variable prompt formula (subject + lens + lighting + end use) applies across all 40 prompts above
- Safety filters block named public figures — describe by role, age range, and appearance instead
- Camera lens language (“85mm f/1.4, shallow DOF”) is the highest single-variable improvement you can make
- Use “edit the image: [change]” for iterative refinement without losing the base composition
- For text inside images, quote the exact string and keep it under 30 characters for reliable rendering
FAQ
Is ChatGPT still using DALL-E 3 for image generation?
No. OpenAI migrated ChatGPT Plus subscribers to GPT Image 1.5 in December 2025 and deprecated DALL-E 3 from the API entirely on May 12, 2026. If you’re using ChatGPT Plus, you’re on GPT Image 1.5.
Why does ChatGPT refuse my portrait prompt?
GPT Image 1.5 blocks prompts that name specific real people to avoid generating potentially misleading or unauthorized images. Remove the person’s name and describe them by role, approximate age, and appearance — the prompt will pass every time.
How do I get consistent results across multiple generations?
Save your optimized prompt as a ChatGPT custom instruction or system prompt. Use the “edit the image:” follow-up command for small changes rather than starting over. For cross-session consistency, paste your exact base prompt at the start of each new conversation.
Can ChatGPT generate images for commercial use?
Per OpenAI’s usage policies, images generated through ChatGPT can be used commercially. The policy explicitly grants users rights to the outputs for commercial purposes, subject to the general content terms. Verify the current terms at openai.com before using generated images in paid campaigns.
What’s the difference between GPT Image 1 and GPT Image 1.5?
GPT Image 1.5 improves text rendering accuracy, handles more objects per scene (up to ~20), and follows natural-language edit instructions more reliably. GPT Image 1 is still available via API at $0.042 per image but is being deprecated October 23, 2026. GPT Image 1.5 API pricing starts at $0.009 per image at low quality.
Why does adding “shot on 85mm f/1.4” improve results so much?
Camera lens language is part of the training data for photorealistic image models. Specifying focal length and aperture activates learned associations with professional photography conventions — compression, depth of field, perspective distortion — that generic style words like “cinematic” don’t reliably trigger.
Last updated: 2026-07-03
