ChatGPT image generation represents one of the most powerful features of modern generative AI, yet it is currently undergoing a significant architectural transition. As of August 2026, many users reporting that image generation has stopped working are not necessarily experiencing a global outage. Instead, they are encountering a mix of service retirements, new safety protocols, and technical timeout thresholds inherent to the latest image models.

Understanding why the service is failing requires a look at the current status of OpenAI’s infrastructure and the specific settings that might be obstructing the process.

Current Service Status and the August 2026 Transition

The primary reason many users are seeing errors is the scheduled retirement of the standalone "DALL-E GPT." OpenAI has officially announced that this specific preconfigured tool will be sunsetted on August 30, 2026.

The Retirement of DALL-E GPT

For a long time, users navigated to a specific "DALL-E" specialized GPT in the sidebar to create visuals. However, this standalone application is being replaced by the integrated "ChatGPT Images" feature. If you are trying to access image generation through an old bookmark or a specific DALL-E GPT link, you may encounter stability issues or "service unavailable" messages.

The future of image creation within the ecosystem now resides directly in the main chat interface. Powered by the gpt-image-1 and gpt-image-2 models, users no longer need to switch tools; they simply need to prompt the standard ChatGPT model to "create an image." If you have important images stored within the legacy DALL-E GPT, it is critical to download them before the August 30 deadline, as that specific chat history may become inaccessible or lose its rendering capabilities.

Is There a Global Outage?

As of today, the core image generation API is operational. While intermittent glitches occurred around August 5, these were localized incidents. If your request is stuck on a spinning loader, the issue most likely lies with local configuration, account-specific limits, or the new "Moderation" filters rather than a total server shutdown.

Top Technical Reasons for Image Generation Failure

When the interface fails to return a visual, the root cause usually falls into one of three categories: connection timeouts, session corruption, or software conflicts.

1. The Rendering Timeout Barrier

One of the most overlooked aspects of the new gpt-image-2 model is its intensive processing requirements. In technical benchmarks, generating a high-quality 1024px image can take anywhere from 145 to 280 seconds.

Many standard web browsers and network gateways are configured to time out after 60 or 120 seconds of silence. If the OpenAI servers are busy, the image might still be generating in the background, but your browser's CDN (Content Delivery Network) or local network may have already severed the connection. This often results in a "Network Error" or a blank response.

The Fix: If a prompt is taking unusually long, do not refresh immediately. Wait at least three minutes. If it still fails, try switching to "Medium" quality instead of "High," which can reduce rendering time by up to 100 seconds.

2. Conflicting Custom Instructions

A common culprit for the "I am unable to create images" text response is a conflict within "Custom Instructions." If your global instructions include directives like "always provide text-based responses" or "do not use external tools," the model may inadvertently block its own access to the DALL-E API.

Even if you haven't explicitly forbidden images, certain persona settings—like forcing the model to act as a "minimalist text-only terminal"—will override the image generation capability.

The Fix:

  1. Click on your profile name in the bottom left corner.
  2. Select "Customize ChatGPT."
  3. Temporarily toggle off "Enable for new chats" under Custom Instructions.
  4. Start a brand-new conversation and try the image prompt again.

3. Session and Authentication Expiry

ChatGPT sessions can occasionally enter a "degraded" state where text generation continues to work, but the multi-modal bridge (which handles images and file uploads) fails to authenticate. This happens frequently if you leave a tab open for several days or switch between multiple devices.

The Fix: Log out of your account entirely, clear your browser's cookies specifically for openai.com, and log back in. This forces a refresh of your authentication token and reconnects you to the most recent server instance.

Solving Image Issues on Mobile Devices

Mobile users face a unique set of challenges. The mobile browser versions of Chrome and Safari often load a "light" version of the ChatGPT interface to save data and memory. Unfortunately, this light interface is frequently incompatible with the complex UI elements required to display generated images.

The Chrome "Desktop Mode" Solution

If you are using an Android device and the image never appears after you hit send, the mobile UI might be the bottleneck.

  1. Open Chrome on your Android phone and navigate to the ChatGPT site.
  2. Tap the three vertical dots in the top right corner.
  3. Check the box for "Desktop site."
  4. The page will reload, and the layout will look exactly like the computer version.
  5. Attempt the image prompt again.

The Safari "AA" Menu for iPhone

iOS users can perform a similar fix:

  1. Tap the "AA" icon in the Safari address bar.
  2. Select "Request Desktop Website."
  3. This often resolves the "Image not showing" bug where the text description appears but the box containing the visual remains empty.

The Case for the Official App

For the most stable experience, OpenAI recommends moving away from mobile browsers entirely. The official ChatGPT app for iOS and Android has built-in native support for image rendering, which bypasses the rendering limitations of mobile browsers. If your browser isn't working, the App Store version is almost always the quickest path to restoration.

Understanding Content Policy and Safety Filters

Often, the service isn't "broken"—it is intentionally stopping the generation due to safety guardrails. OpenAI has implemented a dual-layer safety system for 2026: the Input Filter and the Output Filter.

Input Filter (The Moderation Block)

If you receive a message saying "Your request was rejected by the safety system," the model blocked the generation before it even started. This is usually triggered by specific keywords.

  • Real People: Prompts involving celebrities, politicians, or even public figures in the tech industry are strictly blocked to prevent deepfakes.
  • Copyrighted IP: Asking for "Mickey Mouse" or "Iron Man" will trigger a rejection.
  • Sensitive Contexts: Recent updates have made the filters particularly sensitive to prompts involving "teenagers" or "children" in specific settings to prevent potential misuse.

Output Filter (The "So Sorry" Error)

If the model says, "We're so sorry but the image we created didn't meet our content policy," it means the AI did generate the image, but the final visual contained something that triggered a post-generation scan. This is more frustrating because it consumes your usage limit but provides no result.

How to Bypass Policy Rejections:

  • Be Descriptive, Not Specific: Instead of asking for a specific character, describe the aesthetic. Use "a futuristic hero in red and gold armor" instead of "Iron Man."
  • Avoid Proper Nouns: Do not use names of living people.
  • Focus on Art Styles: Instead of naming a specific artist, describe the style, such as "Impressionist oil painting with thick brushstrokes."

Usage Limits and the 24-Hour Reset

For users on Free or Plus plans, image generation is subject to dynamic rate limiting. Unlike text messages, which are relatively "cheap" in terms of compute, every image generation requires significant GPU time.

The "Clock to Reset" Error

If you receive an error stating you must "wait for your clock to reset," you have hit your quota for the current window.

  • Plus Users: Usually have a high limit, but this can be throttled during peak hours (10 AM to 2 PM EST).
  • Free Users: May be limited to just a few images per day or even per week depending on global demand.

Note: If you have been waiting for 24 to 48 hours and the limit still hasn't reset, this is likely a bug. Starting a new chat session usually clears this, as usage limits are sometimes "stuck" to specific conversation threads.

Advanced Troubleshooting for Power Users and API Clients

For those accessing image models via third-party clients or the API, the issues are often more technical.

API 429 Errors

The image-generations endpoint has a much lower concurrency limit than the chat-completions endpoint. While you might be able to send 50 text prompts per minute, trying to generate more than 3 to 5 images simultaneously will almost certainly result in a 429 "Too Many Requests" error.

Parameter Incompatibility

If you are using a desktop client like Cherry Studio or a custom script, ensure you are not passing the response_format parameter in a way that the gpt-image-2 model doesn't recognize. The latest API requires a clean request body; sending unnecessary headers can cause the entire generation to fail with an "Unknown Parameter" error.

Step-by-Step Restoration Checklist

If you cannot generate images right now, follow these steps in order:

  1. Check the Status Page: Visit status.openai.com to ensure the "Image Generation" service isn't listed as "Major Outage."
  2. Verify the Model: Ensure you are in a standard GPT-4o or GPT-4 chat, not a legacy DALL-E GPT that may have been retired.
  3. Start a New Chat: This is the #1 fix for 80% of issues. It clears the context window and resets the tool-calling logic.
  4. Simplify the Prompt: If you are using a long, complex prompt, try "Generate a simple blue circle." If this works, your original prompt was likely triggering a safety filter.
  5. Disable VPN/Adblockers: Some VPNs and privacy extensions block the WebSocket connections OpenAI uses to stream image data.
  6. Switch to the App: If the web version is failing, the mobile app provides a more robust environment for image rendering.

Summary of the Current Situation

The current inability to generate images is largely a byproduct of the transition toward more integrated AI models and stricter safety enforcement. The retirement of the standalone DALL-E GPT on August 30, 2026, marks a shift toward a more unified ChatGPT experience, but this transition comes with temporary friction. By ensuring your custom instructions are clear, your browser is in desktop mode, and your prompts avoid copyrighted or sensitive material, you can resolve most image generation issues instantly.

Frequently Asked Questions

Why does ChatGPT say it can't create images anymore?

This is usually due to a conflict in Custom Instructions or because you are using an older, unsupported browser version. It can also happen if the model's "tool-calling" ability fails; starting a new chat usually fixes this.

Did OpenAI remove DALL-E?

No, DALL-E is still the engine behind the images. However, the specific "DALL-E GPT" tool is being retired in late August 2026 in favor of integrating the capability directly into the standard ChatGPT conversation.

Why do my images keep failing at 99%?

This is typically a network timeout. High-quality images take a long time to render, and your internet connection or the OpenAI gateway may be closing the connection before the file is fully transferred. Try using a more stable connection or lowering the image quality settings.

Can I generate real people?

OpenAI has strict policies against generating realistic images of public figures to prevent misinformation. If your prompt includes a celebrity's name, it will be blocked.

Why is there a 24-hour wait limit?

Image generation is computationally expensive. All users have a quota. If you hit this quota, you must wait for the rolling window to reset, which is typically 24 hours from your first generation in that cycle.