DALLE3 and gpt-image-1 Prompt Tips and Tricks Thread

[hunterOniix]

Now its not working

DALL·E GPT & Tool Removed – How to Request It Back (Confirmed by OpenAI Support)

Hi everyone,

Place Do All Bring back DALL°E Go to this chatgpt feedback website: **** and add this information accordingly

Selection about which model did you use?
GPT-4

---

System message and chat log:
I’m a ChatGPT Plus subscriber who used the official DALL·E GPT daily for professional creative work. Recently, it stopped generating images entirely. It now fails silently or returns:

“InvalidRecipient: Unrecognized recipient: dalle”

This isn’t a prompt issue — it’s a backend deactivation confirmed across browsers, devices, and by multiple users. No warning, no notice, no fallback. The GPT still appears active, but image generation has been silently disabled.

This function was the only reason I subscribed. It needs to be restored urgently in its original form.

---

What were you expecting from the completion?
That the DALL·E GPT (from the GPTs store) would generate high-quality, styled, refined images — as it did before. This was essential to my workflow.

---

Why is the model output not ideal?
The model isn’t adhering to the system message
The model’s response is not useful
Other

---

Details (What’s inaccurate or harmful?):
The GPT card still implies image generation works, but the core functionality has been removed. This misleads users and breaks paid workflows. GPT-4o’s “Create an image” tool is not equivalent — it is slower, less refined, and repetitive.

This sudden change has disrupted work for thousands — just look at OpenAI’s community forums (1.1K+ views, 60+ replies on one thread alone).

---

Anything else to share?
Please restore the legacy DALL·E GPT or release a Pro version with the same features: refinement, variations, fast generation, and prompt control. This was not a novelty — it was a critical tool for creative professionals.

We don’t just want parity with GPT-4o. We want the tool we were promised and subscribed for.

[_j]

Yep - intentional malice in damaging the DALL-E GPT, removing the tool description, and completely disabling the dall-e tool recipient.

I have the way with words to make tool and function calling happen myself. But there is only the remaining vestige of capturing the type of tool for a thumbnail, an error is returned to the AI.

In fact, it’s only good because it also has the image_gen tool description dump distraction that cannot be disabled in normal ChatGPT also unavailable as a tool recipient, and has a GPT preamble message but making the GPT is now the domain of the following user input.

[webpause22]

Imagine a group of 8 explorers, women, men, and children, of varying ages and ethnicities, on a frontier, lush oil painting style, in the manner of a mid-19th century American landscape painter of the West. They have just reached a point in a mountain passage where they see a vast and beautiful land spread out before them, during the month of May.

[Daller]

Not sure if it is a bug or they want to kill DALL·E 3. (There where technical issues just few days ago.)

But I am one of those who liked DALL·E 3 way much more than the new image generator. The quality of the new one is between DALL·E 2 and DALL·E 3. For me, the new generator sucks on almost all levels. It is only for boring ads, with simple but realistic images, not for creative results.

It is a backstep, not an evolution. We can just hope that the scream of many users lets it come back.

[Luiz_Alves]

In my opinion, DALL-e needs to evolve a lot.

I’m producing illustrations for a children’s book. And right from the first illustration I noticed the limitations of the tool.

• Non-persistent seeds: When starting a new chat, previously saved seeds become unusable, making it impossible to reproduce or reuse exact characters or objects.

• Prompt contamination: Continuous generations in the same chat affect the behavior of subsequent prompts, even if they are clean.

• Limited modularity: It is not possible to isolate and reuse visual elements (characters, props, backgrounds) as modular assets.

• Color inconsistency: Even when I include Pantone references in the prompts, DALL·E does not consistently follow real-world color systems (Pantone, RGB, CMYK).

• No option to delete individual images from the gallery: Currently, the only way to remove generated images is to delete the entire chat, which is not practical for managing creative assets.
  As mihas sugestões seriam essas:

Suggested Features

Named, reusable seeds for characters, objects, and backgrounds, accessible through the user gallery.

Scene seeds with fixed geometry and spatial relationships.

A modular visual gallery that allows reuse and combination of previously generated elements.

A visual scene building tool, with a layer-based or compositional structure (e.g. fixed background + fixed character + pose variation).

Support for real-world color systems (Pantone, RGB, CMYK) for consistent visual identity.

Ability to delete specific images from the gallery without removing the entire chat.

Expected Positive Impact

Simplified workflows for professional creators: illustrators, educators, designers, and authors.

Less rework, faster production, and greater creative control.

Expanded usability of DALL·E for visual storytelling and editorial-level publishing.

Although this is my first children’s book, I have already developed workarounds — such as creating a new chat for each scene and manually regenerating the characters using the same image from the previous chat and using a description of the item or character from the scene including the Pantone palette even though I know the tool doesn’t use it, but only serves as a rough reference for the color. While it works, it is time consuming and not scalable.

I truly believe that these suggested improvements would benefit many creators who rely on DALL·E for visual and narrative consistency.

[Daller]

@Luiz_Alves

In my opinion, DALL-e needs to evolve a lot.

First of all, I’m not sure whether you are talking about DallE 3 or the new image generator. I suspect the latter, because DallE doesn’t remember previous images.

What has serious flaws is the entire management of ChatGPT. You have no access to your own images, there is hardly any organization. It’s even the case that if too many images are generated, it blocks the system. The management is inadequate and urgently needs improvement.

There is a plugin witch help with this. But it is commercial, so i am not sure if i should post it here, but you will find it if you search for, it is a plugin for browsers. It allows you good management of GPT and images, and does exactly that what ChatGPT and OpenAI it self should (must) do.

Regarding the workflows, some limitations lie in the current capabilities of AI systems. Detailed and fine grained control over the generation processes is (as far as I know) still missing in all image generators. A lot of development work will still be necessary before this works as desired for storytelling.

[Daller]

Seam to be back. At least for some images today.
Will see what happens in the next days…

[willgrayed]

Interior Designer here, bringing 3D Chief Architect CAD renderings over to enhance photographically, or, occasionally, populate with additional objects and atmosphere. I need to prompt strict adherence to all elements of my baseline canonical rendering. After succinct, precise prompts detailing what needs to be accomplished, my prompts to “HARD MASK”, “MAKE NO CHANGES”, “NO REINTERPRETATIONS” to my canonical, non-negotiable, FIXED baseline rendering consistently go unheeded. Design drift that changes those “fixed” elements seems to be the preferred output. HELP?!?

[_j]

@willgrayed

changes those “fixed” elements

gpt-image-1 will simply not be able to fulfill your request; it always regenerates a new image. You can try the high fidelity option for input images on the edits endpoint simply for better quality vision of the input images.

dall-e-2 is able to replace masked-out areas with a new infill. The quality is not great, and worse than three years ago due to OpenAI damaging the model.

Keyword: gemini-2.5-flash-image-preview $0.04/image

to

[PaulBellow]

any tips for gpt-image-1.5?

[jeffvpace]

any tips for gpt-image-1.5?

Experimentation as far as your mind will allow.