Tool Introduction

OpenAI was founded in December 2015 and is headquartered in San Francisco, California, with founders including Sam Altman, Elon Musk, Ilya Sutskever, and Greg Brockman; Sam Altman has served as chief executive officer since 2019. OpenAI first revealed DALL-E on January 5, 2021, followed by DALL-E 2 in 2022 and DALL-E 3, released inside ChatGPT for Plus and Enterprise subscribers.

Both DALL-E 2 and DALL-E 3 are being retired from the API on May 12, 2026, with the GPT Image model family positioned as the successor line under the same product umbrella.

Demo Available?

ChatGPT Free includes limited image generation at no cost, so anyone can try the feature before upgrading; Enterprise and Business use a Contact Sales flow.

Aggregate Rating

4.2/5 stars (based on 70+ verified reviews across the internet)

Detailed Overview

What OpenAI DALL-E / GPT Image Does

A user describes a desired image in plain language and the model returns one or more finished images, which can then be refined by asking for revisions in the same ChatGPT conversation. Beyond fresh generation, the model edits existing images: marking a region to replace lets it fill that area from the prompt while leaving the rest intact.

Who OpenAI DALL-E / GPT Image Is For

Individual users reach the tool through the free or paid ChatGPT app for personal and creative image-making, while professional creatives such as artists, creative directors, authors, and architects use it for concept art and visual ideation. Software teams call the Image API or the Responses API image generation tool to add generation and editing features inside their own products.

Top Differentiators

Compared with earlier DALL-E versions, GPT Image renders intricate details such as text, hands, and faces more reliably and supports conversational multi-turn editing rather than re-prompting from scratch. Output reaches presets up to 3840x2160 with custom dimensions, and every image generated through ChatGPT and the API carries a C2PA Content Credentials manifest plus an invisible SynthID watermark.

Founding Year

2015

Target Industry

B2C, B2B, SMB, Enterprise, creative and design, publishing and editorial, architecture, software and product development, healthcare, education

Available Integrations

ChatGPT: the native consumer surface on web, iOS, and Android for conversational generation and editing; free-tier resolution and quality are capped relative to API access.
OpenAI API: the Images and Responses endpoints callable via Python, JavaScript, or HTTP for programmatic generation and editing; rate limits follow the caller's usage-tier tokens-per-minute allowance.
Azure OpenAI Service / Microsoft Foundry: enterprise hosting of gpt-image-1, gpt-image-1-mini, and gpt-image-1.5; the gpt-image-1 series is in limited-access preview requiring approval, and dall-e-3 was retired March 4, 2026.
Microsoft Designer / Bing Image Creator: a consumer creative app built on DALL-E 3 with background removal and generative fill; output runs at 1024x1024 with weekly boost caps.
Microsoft 365 Copilot: calls DALL-E 3 to generate up to four 1024x1024 images per prompt in Word and PowerPoint; available to Copilot Pro but not yet Commercial Copilot users.
Canva: a native DALL-E by OpenAI app inside Canva's image-generation suite, letting users generate without leaving the canvas; Canva was also a ChatGPT Apps pilot partner.
OpenAI Apps SDK: an open standard on the Model Context Protocol for interactive in-chat apps; developers can surface image or design actions reachable by name-mention, in developer preview.
Zapier: a no-code integration offering Generate an Image and Create Image Edit actions; it is bring-your-own-key using a personal OpenAI API key with prepaid billing.

Image generation is sold two ways: bundled into ChatGPT subscription tiers, and as a metered API product (the Images endpoint and gpt-image family). DALL-E 2 and DALL-E 3 were deprecated on November 14, 2025, with API shutdown on May 12, 2026, and replaced by the gpt-image line.

ChatGPT consumer tiers (image generation bundled, not separately priced):

Free: $0, image creation available with the strictest rate limits of any tier.
Go: $8/month in the US, with more messages, uploads, and image creation than Free.
Plus: $20/month, a higher usage ceiling and full access to the current native image model.
Pro: $100/month at roughly 5x Plus usage, or $200/month at roughly 20x Plus usage, switchable anytime from Settings.
Business: $25/user/month billed monthly or $20/user/month billed annually, minimum 2 seats.
Enterprise: custom contact-sales pricing with no published rate card.

API pricing per 1M tokens (image input/output, with a 50% Batch API discount):

gpt-image-2: $8.00 input / $30.00 output; per-image low quality from $0.005.
gpt-image-1.5: $8.00 input / $32.00 output.
gpt-image-1-mini: $2.50 input / $8.00 output.

OpenAI for Nonprofits offers up to a 75% discount on ChatGPT Business or Enterprise. gpt-image-1-mini, gpt-image-1.5, and chatgpt-image-latest are scheduled for shutdown on December 1, 2026 in favor of gpt-image-2.

Pricing last verified: 07/07/2026
Disclaimer: Pricing is subject to change. Please confirm current pricing on the vendor site.

Features & Specs

Model lineup: gpt-image-2 (flagship, snapshot gpt-image-2-2026-04-21), gpt-image-1.5, gpt-image-1, and gpt-image-1-mini.
Dimensions: maximum edge of 3840px, edges as multiples of 16px, aspect ratio up to 3:1, and total pixels from 655,360 to 8,294,400; presets include 1024x1024 up to 3840x2160.
Quality levels: low, medium, high, and auto.
Output formats: PNG (default), JPEG, and WebP, with 0-100% compression via the output_compression parameter.
API surfaces: the Image API (/images/generations and /images/edits, with an n parameter for image count) and the Responses API image_generation tool (action auto, generate, or edit).
Editing: mask-based inpainting with an alpha channel under 50MB, plus up to 4 reference images for style and subject consistency.
Streaming: a partial_images parameter returns 0 to 3 interim renders on eligible models.
Moderation: a strictness parameter of auto or low, with structured error feedback naming the stage and violation category.
Performance: a 1024x1024 low-quality image samples at 196 output tokens; complex prompts can take up to roughly 2 minutes.

Security & Compliance

Data use: data sent to the OpenAI API is not used to train or improve OpenAI's models unless a customer explicitly opts in, in effect since March 1, 2023.
Log retention: abuse-monitoring logs are retained up to 30 days by default unless longer retention is legally required.
Elevated options: Modified Abuse Monitoring and Zero Data Retention are offered to eligible customers; Zero Data Retention covers gpt-image-2, gpt-image-1.5, gpt-image-1, and gpt-image-1-mini, but not dall-e-3 or dall-e-2.
Certifications: a SOC 2 Type II examination covering Security, Availability, Confidentiality, and Privacy for the period January 1, 2025 to June 30, 2025, plus ISO/IEC 27001:2022 and ISO/IEC 27701:2019.
Regulatory support: GDPR, CCPA, HIPAA, and FERPA, with a Data Processing Addendum and a Business Associate Agreement for healthcare customers.
Provenance: images from ChatGPT, the API, and Codex carry a C2PA Content Credentials manifest and an invisible SynthID watermark developed with Google DeepMind, verifiable through OpenAI's public tool.

Tool's Alternatives

Midjourney: Version 7 delivers sharper fine detail in eyes, reflections, and small background elements plus closer prompt adherence, and its Omni Reference feature keeps a character or subject consistent across generations. Access runs mainly through Discord or a web app rather than a public developer API. Its Style Reference system and Style Creator tool build reusable custom house styles aimed at illustrative aesthetic quality.

Adobe Firefly: Its commercial models are trained only on Adobe Stock, openly licensed, and public-domain content, and enterprise customers can buy contractual IP indemnification for generated output. That full indemnification is gated behind Firefly for Enterprise and Creative Cloud for enterprise entitlements rather than included for everyone. Firefly is built directly inside Photoshop, Illustrator, Adobe Express, and Firefly Services APIs as an embedded step in an existing production pipeline.

Google Imagen (Imagen 4): Imagen 4 improves text rendering with better spelling, longer strings, and new layouts suited to comics and packaging, and adds an ultra-fast mode running up to 10x faster than the prior model at up to 2K resolution. Weaknesses remain in factual accuracy for complex compositions and precise object centering. Distribution sits inside the Gemini app and API, Vertex AI, and Google Workspace apps such as Docs and Slides.

Stable Diffusion (SD 3.5): The release shows strong prompt adherence across 3D, photography, painting, and line art with diverse representation without extensive prompting, using Query-Key Normalization for customizability. That intentional diversity means greater variation from identical prompts, and vague prompts can yield unpredictable quality. Its fully open-weight release downloads from Hugging Face and GitHub under the Stability AI Community License, allowing self-hosting, fine-tuning, and offline use.

Frequently Asked Questions

What is OpenAI DALL-E and GPT Image?

DALL-E and GPT Image are OpenAI's family of text-to-image models that generate and edit original images from natural-language prompts, available inside ChatGPT and through the OpenAI API.

How much does ChatGPT image generation cost per month?

Image generation is bundled into ChatGPT tiers: Free at $0, Go at $8/month, Plus at $20/month, Pro at $100 or $200/month, and Business at $25/user/month billed monthly.

What does the GPT Image API cost to use?

Per 1M tokens, gpt-image-2 costs $8.00 input and $30.00 output, gpt-image-1.5 costs $8.00 and $32.00, and gpt-image-1-mini costs $2.50 and $8.00, each with a 50% Batch discount.

When are DALL-E 2 and DALL-E 3 retired?

Both DALL-E 2 and DALL-E 3 are removed from the API on May 12, 2026, after a deprecation announced November 14, 2025, with the gpt-image family as the successor.

What image sizes and formats does it support?

gpt-image-2 accepts a maximum edge of 3840px and aspect ratios up to 3:1, with presets from 1024x1024 to 3840x2160, and outputs PNG, JPEG, or WebP.

How are generated images marked as AI content?

Images from ChatGPT, the API, and Codex carry a C2PA Content Credentials manifest and an invisible SynthID watermark from Google DeepMind, verifiable through OpenAI's public verification tool.

Does OpenAI train on API image data?

Data sent to the OpenAI API is not used to train its models unless a customer opts in, a policy in effect since March 1, 2023.

Can generated images be edited after creation?

The API supports mask-based inpainting, where a marked region under 50MB with an alpha channel is regenerated from the prompt, plus up to 4 reference images for consistency.

Comments are closed.

OpenAI DALL-E / GPT Image

Text-to-image generation and editing across ChatGPT and API