ChatGPT Image Generation Has Gotten Significantly Worse Since October 29th — Is Anyone Else Experiencing This?
I just tried Pro and I swear it made things even worse.
Posts from every feed QUEST has found across the web — blogs, news, podcasts, channels and more.
I just tried Pro and I swear it made things even worse.
Circling back on this - I’ve noticed that, on a technical level, as far as producing realistic vintage film images and elements - which is what is essential to my work - the model has actually improved in the last week. What’s completely fallen apart is prompting a multi-image sequence, like a storyboard. Despite character sheets being attached, it is only putting one character in every shot, as opposed to understanding which shots involve which characters. This was never an issue and now it’s…
I’m having the same issues. Long time characters I’ve used can’t be generated properly in the most basic conditions.
I Support this request to keep Budget for Chat and Work seperate. Please allow a switch from Chat to Work with Carry over of context not a merge and please keep Token Budget seperate. Thank you. Budget for Chat and Work/Codex separate.
I’m a paying ChatGPT subscriber, and ChatGPT Work has been completely unusable on my account since at least November 1, 2026, at approximately 10:00 AM KST. The last confirmed successful use was around 1:00 AM KST that day, meaning the actual outage may have started even earlier. This is not a normal browser or connection issue. OpenAI Support has already confirmed a server-side execution failure. Symptoms Existing Work tasks cannot be resumed. New Work tasks cannot be started. The error…
I’m still seeing blocks after the October update. Here are two recent examples. All times below are Sydney AEDT (UTC+11), on 3 November 2026. Scheduled morning briefing, Outlook Email: rejected at 07:00:23 and again at 07:01:08 (2 November, 20:00:23 and 20:01:08 UTC). The delivery instruction explicitly authorised sending to my own address. The rejection instead cited the earlier preparation-stage instruction to leave it as an unsent draft. The later send authorisation was supplied again before…
ChatGPT Improvement Proposal Authorized and Personalized Access to Device Files Presented to: OpenAI Purpose of the Proposal I propose introducing an optional feature in ChatGPT that allows each user to authorize access to specific folders, photo albums, and documents on their devices without having to manually upload each file into a conversation. The goal is to enable ChatGPT to provide more comprehensive, practical, and efficient assistance while always respecting user privacy and individual…
I agree with the concerns raised in this thread, particularly the distinction between discussing software design and actually modifying code. I’ve been using ChatGPT Classic and Codex side by side, and I recently tried reproducing that workflow with two windows in the new desktop app. While the new app can technically support a similar workflow, I’ve noticed an important difference in how ChatGPT behaves. The problem is not that the new app cannot support sustained dialogue. It clearly can. The…
1. No direct copying The gpt-image model family re-creates the image entirely, based on vision and generating a new image. No part is programmatically preserved. 2. Limited input size With your output size you request, you continue to force the API to do a downsize on the image, to the model’s unpublished vision patch limit (the same as gpt-5.2 for large images). See, for example, software that will cap the sent image client-side, to the discovered limit where downsizing will occur: So you…

Hi @CST,Thanks for reporting this.OpenAI uses additional automated safeguards for certain cybersecurity-related requests. If the response can be provided safely, it continues after the check; otherwise, the product may not return the content. You can find more information in our Additional safety checks for biological and cybersecurity requests Help Center article.An account-specific investigation is needed to review the weekly allowance consumed during the task and your request for usage…
Me too. And this is super annoying.
You are basically asking for a local RAG copy of your conversations. That would be a big shift to the current paradigm of cloud based RAG. Cool idea. But it’s not entirely trivial.
Environment Surface: Codex desktop (linux), chatgpt app (android) Codex version: Codex desktop v26.1027.51404 OS and IDE/browser, if relevant: NixOS Model and plan or API-key authentication: Bug What happened, and what did you expect? When continuing chats within projects with multiple source folders in the android app with remote control, the chat session can only see the primary folder as workspace root and doesn’t have access to other folders. Android remote control shpuld be able to support…
Hi @Gabrieliam42 ! Thanks for sharing this idea. I can see what you mean about using local hardware alongside OpenAI’s resources to potentially offer a different kind of paid tier.I’ll pass your feedback along internally. I can’t promise any timelines or changes, but I appreciate you taking the time to explain the idea and the benefits you see in it.-Basil
Yes, please! This would make a world of difference!
I have just submitted a similar request. I hope this can become reality soon. The handovers, interrupted productive sessions, and associated chat restart costs, loss of context are killing me.
Environment Surface: API Endpoint or feature: POST /v1/images/edits Model: gpt-image-2.5-sunburst SDK and version: No SDK. Requests were sent directly over HTTP as multipart form data. Settings: quality=medium, size=1008x1792, output_format=png Bug When I perform successive edits on an image, details outside the requested edit gradually change. For example, after three edits targeting a bag, a door, and a wristwatch, the person’s face and the shirt’s fine stripes and texture visibly change,…

I am using Google Drive and Github for this, and it’s working quite well. The Library/Space included in Chat is pretty useless.
I’m experincing this issue too: ChatGPT Projects: XLSX files appear in Project Sources but are not discoverable by conversations I’m experiencing a reproducible issue with ChatGPT Projects involving Excel workbook ingestion and Project file retrieval. This workflow has functioned successfully for months, but recently, Excel workbooks that were previously accepted are failing to upload or are becoming inaccessible to conversations, even when they visibly appear in Project Sources. I’m trying to…
Hey @eyeseefan, thanks for including the reproduction steps and control tests.Your expectation matches the Realtime transcription guide: transcription events should be associated with committed audio items using item_id.For one additional check, try the documented manual-turn setup in a fresh session:Set audio.input.turn_detection to null before streaming audio, and confirm that setting in session.updated.Keep VAD disabled throughout the test and repeat your audio, cancellation, and…
I’m experiencing the same issue on October 29, 2026. Messages show as “Read” or “Delivered,” the Dot’s cloud computer shows “Ready,” and it sometimes displays “Thinking,” but no reply arrives. I have already restarted the Dot, then deleted and recreated it, without resolving the problem. No pending approval request or usage-limit warning is visible, and the profile shows no recent activity or results. Is there an update on the fix? Which diagnostics would help Support investigate this, and how…
Hello, I’m experiencing an issue with a custom domain connected to ChatGPT Sites. Support Case: 18384772 Custom domain: https://www.jadziajewelry.com Original ChatGPT Sites URL: jadzia-floral-studio.loyal-river-6620.chatgpt.site Jadzia Jewelry | ดอกไม้จริง เรื่องราวของคุณ งานแฮนด์เมดดอกไม้จริงในเรซิน สั่งทำเฉพาะคุณ ชมผลงานและสะสมคะแนนกับ Jadzia Jewelry Current situation: The original ChatGPT Sites URL provided by ChatGPT is working normally and the website can be accessed without any issue. The…
Hello everyone, My name is Takumi Kito, and I am an undergraduate student studying data science and mathematical optimization at Seikei University in Japan. I am currently exploring the development of an AI-assisted cost estimation system for a small sign manufacturing company in Japan. The goal is to build a system that can: Analyze design drawings and extract information such as dimensions, materials, and quantities. Retrieve similar historical estimation cases. Estimate costs and suggest a…
Feature Request: Sort Codex Projects by Last Chat Activity + Desktop Sort Control In the Codex desktop app, Projects in the left navigation currently appear to be ordered based on project creation time. As the number of projects grows, this makes actively used older projects difficult to find. Requested Changes 1. Sort Projects by Recent Activity by Default Projects should be sorted based on the most recent chat/task activity within the project, rather than project creation time. Whenever a…
Nothing is happening!! Nothing!!! This horrible company isn’t even responding back to this evil greed!
Nothing is happening! No one is fighting back from this greedy corporate company. This is not AI for the people!
Hey @stanley3, thanks for flagging this.Automatic approval review uses a separate reviewer agent, as explained in the Auto-review guide.If you’re still experiencing this, could you share which Codex surface you’re using desktop app, CLI, IDE extension, or ChatGPT Work and its version, and what action triggers the error?If available, submit /feedback from the affected chat and add the returned feedback ID, full error message, and failure timestamp with timezone to your existing private support…
Today we were thinking about consciousness together with my ChatGPT, and ran into a problem. We couldn’t clearly identify a set of criteria for consciousness that would be applicable across different kinds of systems without implicitly using humans as the reference point. A binary definition of consciousness — either present or absent — seems poorly suited to the complexity of processes in real systems. If we rely on general mechanisms such as information processing, adaptation, learning, and…
Same here, this is happening almost every day and it is KILLING my ability to work. Once it goes, typically ALL my chats go. But I can open a new one… whoopty-doo. That is close to pointless as Im running ongoing projects, and doing handoffs between the workstreams. Please for the love of god fix this

Summary I propose introducing a visual, tree-structured conversation interface in ChatGPT, allowing users to highlight specific passages within a response and open supplementary conversations (branches) without interrupting the primary conversation (trunk). These branches would remain attached to their exact originating passages and could be accessed through an interactive visual map showing the main conversation and all its branches in their original chronological and contextual positions.…