Everything you need to get started with GPT Image 2 — from your first generation in ChatGPT to Thinking Mode, multi-turn editing, and reference images.
Six essential topics to master GPT Image 2, from account access to advanced prompting.
You do not need a paid plan to use GPT Image 2. The Free tier includes access with Instant Mode and standard rate limits. Plus and Pro plans unlock Thinking Mode and higher rate limits. Go to chatgpt.com and sign in or create a free account. GPT Image 2 is accessible on all plans.
Thinking Mode is GPT Image 2's most significant feature for quality. Before generating, the model runs an agentic planning step: it searches the web for relevant visual references if needed, reasons through the layout and composition, and self-corrects potential issues. It produces noticeably better results for prompts with multiple subjects, precise spatial arrangements, infographics, and any image that requires accurate real-world information. Thinking Mode requires a Plus ($20/mo), Pro ($100/mo), or Pro ($200/mo) subscription.
One of GPT Image 2's most practical capabilities is context-aware multi-turn editing. After generating an image, you can describe specific changes in follow-up messages. The model preserves all unchanged elements and applies only the changes you specify. The key to successful editing prompts is the
GPT Image 2 accepts up to 16 reference images per generation request. Reference images are useful for maintaining character consistency across multiple generations, matching an established visual style, and transferring color palettes or textures from existing assets. To add reference images in Chat
GPT Image 2's reasoning architecture means it responds better to intent-driven descriptions than keyword lists. Think of writing a brief for a photographer or art director, not writing search tags. Instead of "moody lighting," write "single overhead lamp, warm tungsten light, deep shadows on left."
Most GPT Image 2 generation problems fall into a small number of categories, each with a clear fix.
Common questions about using GPT Image 2 in ChatGPT and via the API.
Go to chatgpt.com and sign in. GPT Image 2 is available on all plans including Free. Simply describe what you want to generate — ChatGPT automatically uses GPT Image 2 for image requests. You can also select it explicitly from the model selector in the interface.
Thinking Mode requires a ChatGPT Plus ($20/mo), Pro ($100/mo), or Pro ($200/mo) subscription. When available, look for the Thinking Mode toggle in the ChatGPT interface before submitting your prompt. When enabled, GPT Image 2 plans layout and self-checks output before generating, producing higher-quality results for complex prompts.
Type a follow-up message in the same chat describing what you want to change. Specify what changes and what stays the same. GPT Image 2 uses the full conversation context to apply targeted edits. Do not start a new chat — the model needs the context of the original generation.
Click the attachment icon in the ChatGPT chat input and upload your reference images before typing your prompt. In ChatGPT, you can attach up to 16 reference images. When using Thinking Mode for character consistency, up to 8 reference images are used. Then describe in your prompt which aspects should be maintained — character appearance, color palette, art style, etc.
The most common causes: (1) vague style words instead of physical descriptions, (2) text in the image not wrapped in quotes, (3) too many elements in a single prompt, (4) Thinking Mode off for a complex prompt. Fix by adding physical specifics, quoting exact text, and enabling Thinking Mode for complex generations.
Try GPT Image 2 free — no credit card required. Generate your first image in seconds.