GPT Image 2 is OpenAI’s newest image model, released on April 21, 2026 as part of ChatGPT Images 2.0. It replaces GPT Image 1 across ChatGPT, Codex, and the OpenAI API, and it lands in Picsart the same day. Try it inside the Picsart AI Image Generator, the AI Playground, or as a node in Picsart Flow. No extra subscription, no tab-hopping, no waiting.
This release is less of a refresh and more of a rethink. GPT Image 2 follows instructions more precisely, holds on to fine details, composes complex scenes without flinching, and can stop to reason about a prompt before a single pixel is rendered.
Meet GPT Image 2
GPT Image 2 is the direct successor to GPT Image 1 and ships with two modes built for two different speeds of work.
Instant Mode is the quick pass. A prompt goes in, an image comes out, and the model keeps things moving. It’s the right call for fast drafts, mood exploration, and anything that doesn’t need the model to slow down and think.
Thinking Mode is the new part, and it’s the big one. Before generating, the model runs a reasoning step. It breaks a prompt into components, figures out how those components should fit together, pulls in uploaded references when they help, and checks its own work before the result lands. OpenAI describes the shift as a move from approximating a prompt to actually following it.
See what’s new
Reasoning now lives inside image generation. Thinking Mode treats a prompt less like a suggestion and more like a brief. Layouts with specific elements in specific spots, dense compositions, and multi-step visual instructions get handled with real intention instead of an educated guess. The result is fewer weird misreads and more finished-looking first outputs.
Up to eight coherent images come back in a single pass. With Thinking Mode on, one request returns a set, not just a picture. Characters, objects, and style stay consistent across the batch, which makes the model a solid fit for storyboards, ad variants, catalog pages, comic panels, and anything that needs to feel like it belongs to the same world.
Non-Latin scripts finally render cleanly. Japanese, Korean, Chinese, Hindi, and Bengali all see real improvements, alongside sharper Latin-script typography. Posters, labels, diagrams, and comics can keep their language in the design instead of fighting it.
Small text, icons, UI elements, and dense compositions hold up under pressure. The details that used to break image models, like tiny captions and fiddly interfaces, come through noticeably cleaner than anything in the GPT Image 1 era.
Style range widens. Cinematic stills, pixel art, manga, photorealism, editorial illustration, and plenty in between all come out stronger, with the model holding style better across a single generation and across a set.
Aspect ratios stretch from 3:1 to 1:3. Wide banner, square social, tall vertical story, magazine spread, and everything between each gets a native frame, so post-generation cropping stops being the default step.
Output reaches up to 2K (around 2000px) through the API, with sharper edges and more reliable fine-grain detail at that size.
Find GPT Image 2 across Picsart
The AI Image Generator is the quickest route in. Open the full-screen workspace, pick GPT Image 2 from the model list, describe the scene, and generate. The results drop straight into Picsart, ready to edit, restyle, resize, or export.
Generate inside the AI Image Generator
Step 1. Open the AI Image Generator at picsart.com/ai-image-generator, then pick GPT Image 2 from the Ai Model dropdown in the prompt bar.
Step 2. Choose Instant Mode for quick drafts or Thinking Mode for complex prompts and consistent image sets, then write the prompt.
Step 3. Set the aspect ratio through the Auto dropdown (3:1 to 1:3) and pick a style, or leave both on default.
Step 4. Hit Generate. In Thinking Mode, request up to eight coherent images in a single pass.
The AI Playground is the spot for creators who want more room to iterate. GPT Image 2 sits next to Picsart’s other image models, so one prompt can run through a lineup of AIs for a side-by-side look. Aspect ratio, style, and resolution are all tunable, and every result saves to the active board.
Iterate inside the AI Playground
Step 1. Open the AI Playground, pick a board (or hit New board), and set the media toggle to Image.
Step 2. Open the model picker, head to the Image tab, and select GPT Image 2.
Step 3. Choose Instant Mode or Thinking Mode, write the prompt or tap Inspire me for a starter, then set the aspect ratio and image count (up to eight in Thinking Mode).
Step 4. Generate, and every result lands on the active board, ready to remix.
Picsart Flow is where GPT Image 2 becomes part of a bigger pipeline. Drop it into a flow as a node, connect it to prompt generation, style references, localization steps, and export settings, then rerun the whole workflow anytime a new campaign, product drop, or batch lands. One flow, built once, reused endlessly.
Scale up inside Picsart Flow
Step 1. Open Picsart Flow at picsart.com/flow and start a new flow or open an existing one.
Step 2. Add an AI Image Generation node to the canvas, open the node’s model picker, and choose GPT Image
Step 3. Pick Instant Mode or Thinking Mode, then wire in the inputs and any downstream nodes for resizing, translation, or export.
Step 4. Run the flow, and rerun it anytime a new campaign or batch needs visuals.
Start generating with GPT Image 2
GPT Image 2 is live in Picsart right now. Quick drafts, reasoning-driven generations, multilingual posters, eight-image sets, ultra-wide banners, ultra-tall verticals, and 2K outputs all live under one roof. Open the AI Image Generator for the fastest path, jump into the AI Playground to compare and iterate, or wire it into Picsart Flow for a workflow that runs itself.
