ChatGPT image generation can fail in a handful of ways: a vague “something went wrong” error, a high-demand notice, a policy refusal, or a spinner that never ends. Since ChatGPT Images 2.6 arrived on 4 November 2026 with GIFs and saved references, there are a few new ways for it to trip up too.
This guide maps each error to its cause and gives you the fix. It covers the ChatGPT app and website, plan limits, and the gpt-image-2.6 API for developers.
- Check the status page first: image generation can fail while text chat works fine.
- “High demand” or a limit notice: you hit a cap. OpenAI does not publish exact image numbers or reset times, so wait or upgrade.
- Policy refusal: rephrase to describe a fictional or generic subject. Never try to trick the filter.
- Stuck “Creating image”: wait a few minutes, then refresh and start a new chat.
- Edits ignored: select a smaller area and describe one change at a time.
- API errors: 400
moderation_blocked, 403 verification, 429 rate or spend limits, 503 overloaded.
Quick Checks Before Anything Else
Most image failures are not your fault. They come from a service incident, a usage cap, or a stale browser session. Run these steps in order before you change your prompt.
- Check status.openai.com. Look for “Image Generation” under ChatGPT. On 20 August 2026, for example, OpenAI expanded a Thinking mode incident to include image generation, and it took about three hours to fully recover.
- Try a tiny test prompt. Ask for “a red apple on a white table”. If that works, the service is fine and the problem is your prompt, file or chat.
- Start a new chat. Very long chats with many images can slow down or fail. A fresh chat clears that context.
- Refresh or sign out and back in. This renews your session token, which fixes many silent failures.
- Update the app. Images 2.6 features such as GIFs rolled out gradually, so an old app version may not show them.
- Switch device. If the web fails, try the mobile app, or the other way round. This tells you if the problem is local.
Tip: If text chat is also broken, the issue is wider than images. Our ChatGPT not working fix guide covers login, network and browser problems.
ChatGPT Image Error Messages and Fixes
Here are the messages people see most often, what they usually mean, and the first thing to try.
| Message or symptom | Likely cause | Fix |
|---|---|---|
| “Something went wrong” | Server error, timeout or broken session | Check status, refresh, retry in a new chat |
| “We’re experiencing high demand” | Peak traffic or temporary capacity limits | Wait 15 to 30 minutes, try off-peak hours |
| Limit reached notice with a Get Plus button | You used your plan’s image allowance | Wait until later, or upgrade (upgrading resets limits) |
| “This request may violate our content policies” | Prompt or output flagged by safety checks | Rephrase with a generic or fictional subject |
| Spinner stuck on “Creating image” | Slow job, dropped connection or failed render | Wait 2 to 3 minutes, then refresh and resend |
| Edit returns the same image | Selection too large or vague instruction | Select a smaller area, ask for one change |
| Upload fails or spins forever | File over 20 MB, odd format or daily upload cap | Resize, convert to PNG or JPG, try again later |
| Download button does nothing | Browser blocking downloads or pop-ups | Open the image full size and save, or use the Library |
| Wrong shape or misspelled text | Ratio or exact wording not stated | Name the ratio and put text in quotes |
The rest of this guide explains each row in more detail.
Image Limits per Plan and When They Reset
Image creation has its own usage limit, separate from your chat limit. OpenAI’s free tier FAQ says that when you hit it, you must “wait for a later time” to use the tool again. It does not give a number or an exact reset time.
| Plan | Image access (official wording) | Images with thinking |
|---|---|---|
| Free | Limited | No |
| Go | Higher limits than Free | No |
| Plus | More, plus more complex and accurate images | Yes |
| Pro | Unlimited and faster, subject to abuse guardrails | Yes |
| Business / Enterprise | Set by workspace and plan | Plan dependent |
You will see numbers online such as “2 to 3 a day on Free” or “50 every 3 hours on Plus”. Treat them as user reports, not official rules. They conflict with each other, and OpenAI can lower caps during busy periods.
What to do when you hit the cap
- Wait: limits refresh on a rolling basis. Try again after a few hours.
- Batch your ideas: describe everything you want in one prompt instead of many small retries.
- Upgrade: the free tier FAQ says upgrading to Plus, Pro or Business resets your rate limits. Compare plans in our ChatGPT Free vs Go vs Plus vs Pro guide.
- Use the API: developers pay per token and are not bound by ChatGPT plan caps.
Note: GIFs count as image generation too. OpenAI has not published separate GIF limits, so expect free accounts to run out faster.
Content Policy Refusals and How to Rephrase
ChatGPT checks both your prompt and the finished image. A refusal can happen before drawing starts, or after the image is made and flagged. Common triggers include sexual content, graphic violence, real people without consent, and close copies of brands, logos or a living artist’s style.
Some refusals are false positives. A harmless word, such as a medical term or “shoot” for a photo shoot, can trip the check. In those cases you can rephrase in a way that stays inside the rules.
Rephrasing that respects the rules
- Make subjects fictional: “a middle-aged news anchor” instead of a named real person.
- Describe a style, not a person: “bold flat colours and thick outlines” instead of “in the style of [living artist]”.
- Remove ambiguous words: “a photo session” instead of “a shoot”.
- Use your own photos: only upload images of yourself, or of people who agreed.
- Create original branding: “a coffee logo with a leaf” instead of copying an existing logo.
Warning: If your idea genuinely breaks OpenAI’s usage policies, no wording is acceptable. Trying to trick the filter can get your account flagged or banned, so drop the request instead.
If you believe a refusal was a mistake, use the thumbs-down button on the reply to send feedback. That is the official way to report over-blocking.
Stuck on “Creating Image”
Images 2.6 is faster than earlier versions; OpenAI says about 50% faster. Still, complex scenes, “Images with thinking” and GIFs can take a few minutes. A long wait is not always a failure.
- Wait 2 to 3 minutes. Do not resend during that time, as repeated requests can eat into your limit.
- Refresh the page. Sometimes the image finished but the page did not update.
- Check the Library. Open Library in the sidebar to see whether the image saved there.
- Simplify the prompt. Fewer subjects and less text mean faster, more reliable output.
- Start a new chat and resend. Paste your prompt into a fresh conversation.
- Check your connection. Turn off VPNs, ad blockers or proxies that can cut long requests.
Edits, Uploads and Downloads Not Working
Edits or inpainting not applying
ChatGPT lets you click an image, highlight an area and describe the change. The selection is guidance, not a hard mask, so changes can spill outside the area or be skipped. OpenAI’s API docs say the same about masks: the model “may not follow its exact shape”.
- One change per edit: “make the shirt blue” works better than five changes at once.
- Say what must stay: “keep the face, pose and background exactly the same”.
- Select tighter: a small, precise highlight gives better results than a large one.
- Use saved references: for the same person or pet across edits, save them as a reference and call them with @name.
- Restart from the best version: after many edits, quality can drift. Download the best one, upload it in a new chat, and continue.
Note: Uploaded animated GIFs cannot be edited. You can only refine GIFs that ChatGPT made in the same chat.
Upload errors
- Size: OpenAI’s file uploads FAQ lists a 20 MB per image limit. Shrink large phone photos first.
- Format: convert HEIC or unusual formats to PNG or JPG.
- Daily caps: free accounts have a small daily upload allowance, so uploads may stop working until later.
- Network: slow mobile data can time out. Try Wi-Fi.
Can’t download the image
- Open full size: click the image, then use the download icon at the top.
- Right-click or long-press: choose “Save image” on desktop, or long-press on mobile.
- Allow downloads: some browsers block downloads from new sites. Check the address bar for a blocked icon.
- Check storage permission: on mobile, the app needs photo or storage access to save.
Wrong Aspect Ratio or Garbled Text
If you do not say a shape, ChatGPT picks one, often square. State it clearly in the prompt, such as “wide 16:9 banner” or “tall 9:16 phone wallpaper”. The API supports custom shapes between 1:3 and 3:1, so very extreme ratios are not possible.
Text inside images is much better than it used to be, but long text still fails sometimes. These habits help:
- Quote exact text: write the words inside quotes, such as “BIG SALE”.
- Keep it short: a headline and one line beat a paragraph.
- Fix in an edit: select the wrong word and say “change the text to read ‘Open Daily’”.
- Add it later: for long copy, generate without text and add it in a design tool.
App vs Web Problems
Features roll out unevenly. The new GIF tool, for example, was still rolling out after launch, so one device may have it before another.
| Problem | Web fix | App fix |
|---|---|---|
| New feature missing | Hard refresh with Ctrl+Shift+R | Update from the App Store or Play Store |
| Image never appears | Disable extensions, try another browser | Force close and reopen the app |
| Can’t save image | Allow downloads in site settings | Grant photo or storage permission |
| Upload button greyed out | Sign in again, check daily upload cap | Allow camera and photo access |
| Odd errors after an update | Clear cookies for chatgpt.com | Clear the app cache (Android) or reinstall |
gpt-image-2.6 API Errors for Developers
The API offers two Images 2.6 models: gpt-image-2.6-flare for fast drafts and gpt-image-2.6-sunburst for detail and consistent edits. Use the full model ID. For the feature split and prices, see our Images 2.6, Flare and Sunburst guide.
| Error | Meaning | Fix |
|---|---|---|
400 moderation_blocked |
“Your request was rejected by the safety system”, at input or output stage | Change the prompt; moderation: "low" is less strict but checks still apply |
| 400 invalid size | Size not divisible by 16 or ratio outside 1:3 to 3:1 | Use 1024×1024, 1536×1024, 1024×1536 or a valid custom size |
| 400 background or format | Transparent background with JPEG output | Use png or webp for transparency |
| 400 mask with GIF or markup | Masks are not supported with GIF output, or with markup | Use one edit method per request |
| 401 | Invalid or wrong API key | Check the key and project, or make a new one |
| 403 verification | Organization Verification not completed | Verify your org in the developer console |
| 429 rate limit or “Slow down” | Too many requests, or traffic ramped up too fast | Follow Retry-After, use backoff with jitter |
| 429 credits or spend limit | No prepaid credits or spend cap reached | Add credits or raise the limit; retrying will not help |
| 500 | OpenAI server error | Retry shortly, check the status page |
503 server_is_overloaded |
Model temporarily overloaded | Retry with longer delays |
Developer checklist
- Update the SDK. Run
pip install -U openaiso new parameters such ascomplexityare recognised. - Verify your organization. OpenAI says you may need API Organization Verification before GPT Image models respond.
- Check the size rules. Custom sizes need both sides divisible by 16, within 1:3 to 3:1, up to 3840×2160.
- Keep prompts under 32,000 characters. That is the documented maximum.
- Edit inputs: you can send 1 to 16 images, and a markup layer must be a transparent PNG matching the first image’s size.
- Send a
safety_identifier. A hashed per-user ID (128 characters or fewer) helps OpenAI handle abuse per user instead of per account. - Lower
complexityfor speed. High is the default; low cuts latency for simple images and quick edits.
from openai import OpenAI, RateLimitError, BadRequestError
import base64, time
client = OpenAI()
for attempt in range(4):
try:
r = client.images.generate(
model="gpt-image-2.6-flare",
prompt="flat illustration of a lighthouse at dusk",
size="1536x1024",
quality="low",
)
open("out.png", "wb").write(base64.b64decode(r.data[0].b64_json))
break
except RateLimitError:
time.sleep(2 ** attempt) # back off, then retry
except BadRequestError as e:
print("Fix the request, do not retry:", e)
break
Tip: For a full list of OpenAI API errors beyond images, see our OpenAI API errors fix guide. New to the API? Start with how to get an OpenAI API key.
Troubleshooting: Common Problems and Fixes
- Images fail only on one account: you may have hit that account’s cap. Check for a limit notice higher up the chat.
- Works on mobile, fails on desktop: an extension or ad blocker is the usual cause. Try a private window.
- Same face changes every edit: save the person as a reference and mention them with @name in each prompt.
- Image quality dropped suddenly: during high demand, output can be slower or simpler. Try again later or ask for “high detail”.
- Custom GPT won’t make images: the GPT’s builder may have turned off image generation. Use regular ChatGPT instead.
- GIF option missing: it was still rolling out after launch. Update the app and check again in a few days.
- API works in tests but fails in production: check spend limits and rate tiers, which differ per project.
- Need images and ChatGPT is down: other assistants have their own tools. See can Claude make images for what Claude can and cannot do.
Frequently Asked Questions
Why does ChatGPT say “something went wrong” when creating images?
It is usually a server error, a timeout or a stale session. Check status.openai.com, refresh the page and retry in a new chat.
How many images can I make on ChatGPT Free?
OpenAI does not publish an exact number. Free has a limited image allowance separate from chat, and Go, Plus and Pro get more.
When does the ChatGPT image limit reset?
OpenAI does not state an exact reset time. It says to wait and try later, and upgrading to Plus, Pro or Business resets your limits.
Why was my harmless image request refused?
Safety checks can flag innocent words or real-person references. Rephrase with a generic or fictional subject, and send thumbs-down feedback if you think it was a mistake.
Why are my image edits not changing anything?
The selection is guidance, not a strict mask. Highlight a smaller area, ask for one change, and say what must stay the same.
Why does my gpt-image-2.6 API call return 403?
Your organization probably needs API Organization Verification. Complete it in the developer console, then try again.
Can I edit a GIF I uploaded to ChatGPT?
No. Uploaded animated GIFs cannot be edited, but you can refine GIFs that ChatGPT created in the same chat.
Sources
- OpenAI Help: ChatGPT Free Tier FAQ
- OpenAI Help: File Uploads FAQ
- OpenAI API: Image generation guide
- OpenAI API reference: Images
- OpenAI API: Error codes
- OpenAI Status: Thinking mode and image generation incident (20 Aug 2026)
- OpenAI Community: moderation_blocked error thread
- TechRepublic: ChatGPT Images 2.6 GIF creation



