Tech & AI, Explained

openai-image

5 min read

🇺🇸 USA · OpenAI Image (gpt-image-1 / “Image 2.0”)

Status: 🟩 COMPLETE 🟦 LIVING Last updated: 2026-06-26 Plain-English tagline: OpenAI’s flagship image-generation model — the one inside ChatGPT, replaced DALL-E 3 in 2025. Best in class for instruction-following and text in images.

Front-matter facts

FieldValue
VendorOpenAI (San Francisco, USA)
Country / origin🇺🇸 USA
Recommended for Australian users?✅ Yes — accessible via ChatGPT or OpenAI API
Privacy summaryChatGPT tier privacy applies; API: never trains on inputs
Free tierLimited via ChatGPT free; meaningful access via Plus
Paid tiersChatGPT Plus US200/mo (heavy quota); OpenAI API pay-per-image
First releasedgpt-image-1 March 2025 (replaced DALL-E 3)
Last reviewed2026-06-26
Official sitehttps://openai.com (accessed via chatgpt.com / API)

What it is

gpt-image-1 is OpenAI’s current flagship image-generation model. It replaced DALL-E 3 (the previous flagship) in March 2025. Marketing sometimes refers to it as just “Image” inside ChatGPT or informally as “Image 2.0” (because it succeeded DALL-E 3 as a major architecture step).

Key distinguishing strengths:

  • Best-in-class instruction-following — does what you ask faithfully, vs Midjourney’s tendency to reinterpret freely
  • Best-in-class text rendering in images — long a weak spot for image AI; gpt-image-1 mostly nails text in posters, signs, screenshots
  • Built into ChatGPT — no separate app; just ask in chat
  • Iteration via chat — refine outputs conversationally (“now make the sky stormy”)
  • Photorealistic AND illustrative styles
  • Editing — upload an image, ask for edits
  • Vision + generation — same conversation can analyse and create

Available in:

  • ChatGPT consumer (Plus / Pro / Free with limits) — the most-common access path
  • OpenAI API (gpt-image-1 model) — for building image gen into your own apps
  • Microsoft Designer + Copilot — Microsoft licenses gpt-image-1 via OpenAI partnership
  • DuckDuckGo AI Chat and other intermediaries

The product is DALL-E’s successor; DALL-E branding is being phased out by OpenAI.

What you’d use it for

  • Posters / flyers with embedded text — gpt-image-1’s text rendering is excellent
  • Diagrams / charts / labelled images — text-in-image accuracy matters
  • Photo edits via conversation — “remove the person on the left,” “add a sunset behind”
  • Product mockups — labels, packaging, signage
  • Conversation-driven iteration — refining via chat is gpt-image-1’s UX win
  • Inside ChatGPT — convenient one-app workflow
  • API integration in your own apps

How to use it

Via ChatGPT (easiest)

  1. Sign in to chatgpt.com
  2. Type a prompt — gpt-image-1 generates automatically for image requests
  3. Iterate via chat
  4. Download

Via OpenAI API

  1. Sign up at platform.openai.com
  2. Use gpt-image-1 model in API calls
  3. Pay per image generated (varies by resolution / quality)

Via Microsoft Designer / Copilot

  1. Use Microsoft Designer at designer.microsoft.com or Copilot’s image features — gpt-image-1 powers them

What it costs

ChatGPT tier (most common access)

  • Free: very limited daily images
  • Plus US$20/mo: ~40+ images/day at standard quality
  • Pro US$200/mo: essentially unlimited, highest quality

OpenAI API

  • ~US$0.02-0.08 per image depending on resolution and quality tier
  • Standard / HD modes
  • Pay per generated image (no subscription)

Via Microsoft Designer

  • Free with Microsoft account, limited boosts/day
  • Microsoft Copilot Pro for higher quotas

How it compares to alternatives

CapabilityOpenAI Image / gpt-image-1Imagen 4 (Google)MidjourneyAdobe Firefly
Instruction-followingBestExcellentLooser interpretationGood
Text in imagesBestExcellentImprovingGood
PhotorealismExcellentBestStylisedExcellent
Distinctive aestheticPolished generalPhotorealMost distinctiveLess distinctive
In-chat iterationBest (via ChatGPT)Yes (via Gemini chat)Discord-basedInside CC apps
Editing existing imagesStrongGoodLimitedBest (Photoshop integration)
Commercial-safety positioningStrongStrongSome controversyBest (Firefly indemnification)
Built into a major chat productYes (ChatGPT)Yes (Gemini)NoNo

For text-in-image, instruction-following, and chat-driven iteration, gpt-image-1 wins. For pure photorealism, Imagen edges ahead. For commercial safety + Creative Cloud, Firefly. For distinctive style, Midjourney.

Privacy / data handling

  • ChatGPT tier privacy applies to image generation inputs (training defaults apply per tier; opt-out available)
  • API doesn’t train on inputs
  • Generated images include C2PA invisible provenance metadata
  • For sensitive content (personal photos, confidential mockups), use ChatGPT Team / Enterprise tier

Recent changes

  • March 2025: gpt-image-1 launched as DALL-E successor
  • 2025-26: Quality refinements and capability expansion
  • Pre-2025: DALL-E 3 was the flagship; gpt-image-1 succeeded it
  • DALL-E 2 / 1: earlier OpenAI image gens; now historical

Gotchas

  • Real-person generation is restricted — gpt-image-1 refuses likenesses of celebrities, politicians, etc.
  • Copyrighted character generation is restricted — Disney characters, Marvel, etc. blocked
  • Style transfers (“in the style of [living artist]”) are restricted
  • Quality vs speed tradeoff — higher-quality settings take longer and cost more API tokens
  • API uses tokens for both prompt and image bits — high-res images = more output cost
  • Generated images carry C2PA invisible provenance metadata — detectable by C2PA-compliant tools; not a problem for most uses but worth knowing
  • In ChatGPT, images count against your Plus / Pro daily limits — heavy image generation can hit limits
  • Editing existing images can produce surprising changes — review carefully

See also

Sources

Graph View

Backlinks