DALL-E Is Gone From ChatGPT - What Creators Should Use for AI Images Now

On August 30, 2026, OpenAI retired the original "DALL-E" GPT inside ChatGPT, one of the very first custom GPTs and the tool that introduced a huge share of casual creators to AI image generation. OpenAI told users to download any images they wanted to keep before the cutoff. Image generation itself has not left ChatGPT: it lives on through ChatGPT's newer built-in image tool, and third-party custom GPTs with image generation enabled are untouched. It is specifically OpenAI's original, dedicated DALL-E GPT and its saved conversation history that are gone.

What this actually changes for creators

If you had a habit of opening the DALL-E GPT directly, that entry point and its thread history no longer exist. If you generate images through plain ChatGPT chat, nothing about your workflow breaks. Either way, it is a good moment to notice something creators using ChatGPT for images have run into for a while: it is a general-purpose chat assistant with an image tool bolted on, not a purpose-built image generator.

Why a dedicated image model beats a chat assistant with an image tool

• Consistent aspect ratio and resolution controls, instead of the assistant guessing what you probably want.
• Reference images you can pin as the identity or style anchor across a whole batch, not just describe in words.
• Model choice - different image models render faces, product shots, text-in-image and illustration styles very differently, and a single chat tool only gives you one of them at a time.
• A visible cost per generation, so you know what a batch of variations is actually going to run before you click.

Dedicated image models worth trying instead

• GPT Image 2 - OpenAI's own current-generation image model, available directly in Karya's Image Studio outside of ChatGPT.
• Nano Banana Pro - Google's image model, strong at photorealistic edits and consistent characters across a set of images.
• Seedream 5 - ByteDance's image model, a solid pick for product and marketing shots.
• Qwen Image 3 - Alibaba's image model, good at rendering readable text inside an image.

Trying it in Karya

All four of the models above are live in Karya's Image Studio today, alongside the rest of the catalog. Karya shows the exact credit cost before you generate, there is no subscription, and credits never expire - packs start at $3, so switching models to compare results costs nothing beyond what you actually generate.