Hey, thanks for the report.
I am a human, my name is Jay, and there is no AI or language translation in my answer. I do not work for OpenAI, I am a fellow AI user with experience, and I am located in the United States.
Using half-a-brain, and being not an AI-powered time-waster asking you to spend more of your money, nor an outsourced script-reader with no capability for understanding the basics, here is what I am able to infer about your issue:
- You had 100 or more successes with your workflow, and then something with OpenAI changed, making this OpenAI’s fault (check that you did not make any alteration, and try earlier code, to ensure this is true);
- Your timezone is Europe/Berlin, and your post was made Oct 29 2025 20:49:50 UTC/GMT - 21:49 your time. Therefore the time range of your requests being “today” would be reporting on the activities at most inclusive of 2025-10-28 23:00 to 2025-10-29 20:00, likely the middle of that range;
- you are ID-verified, or would otherwise get API errors with this model.
- You used the OpenAI API (with the edits endpoint needed for visual understanding of a provided image), as it is impossible to infer any other use without being completely obtuse;
- Thus, there is no “screenshot” to provide,
- There is no request ID to provide,
- and there is no “metadata” to report other than the image returned in base64.
Thus showing that the solicitation of more information in the reply above is quite ill-informed, and you cannot trust ANYTHING will be done for you after your own work and expense, because replication of this input is easy to confirm or fail to reproduce remotely, and the modifications to input requested could be done in earnest by OpenAI - instead of probing you with nonsense questions.
The only thing you have not provided is an input image. You say “similar prompt”, but don’t say “similar images”, so it is not absolutely clear that this issue is pinned to one particular input image.
Your input shape is JSON object or dictionary, although the endpoint takes a multipart/form-data request (and also needs not merely a file name string). This indicates to me that you are likely familiar with providing such an object as input into either an OpenAI SDK as parameters or are constructing your own requests with a code library that can similarly transform a serializable object to the API request.
- exception 1: It is assumed you are encoding an “image” file as form data and not merely its file name string (error would result). Images are not an input except to “edits”.
- exception 2: “moderation” is not an accepted parameter by the “edits” endpoint
- exception 3: the API reference for edits states today, “This endpoint only supports
gpt-image-1 and dall-e-2”, however this tells us nothing, as the generation API documentation also has not been updated to this clearly-supported model.
Here is exact replication code, the API reference with your input shape of an actual file, and a non-clobbering file name as output.
import os, base64
from openai import OpenAI
client = OpenAI()
prompt = (
"Create a highly realistic digital illustration of the uploaded "
"animal, capturing the relaxed and entertained spirit of the "
"'television' (📺) emoji. The animal’s facial expression should be "
"one of content relaxation—its eyes fixed forward with a calm, "
"slightly mesmerized gaze, and a soft, closed-mouth smile, as if "
"it's comfortably watching its favorite show. To represent the "
"emoji, the animal should be sitting comfortably in front of a "
"realistic television set, perhaps a charming, retro model with "
"'rabbit ear' antennas. The TV screen should be illuminated, "
"casting a soft, colorful light onto the animal. The rendering of "
"the animal must be photorealistic, with meticulous attention to "
"its specific fur/skin/scale texture, coloring, and anatomy as "
"seen in the uploaded photos. The background must be completely "
"transparent."
)
parameters = {
"image": open("IMG_5656 2.jpeg", "rb"),
"prompt": prompt,
"model": "gpt-image-1-mini",
"n": 1,
"size": "1024x1024",
"quality": "high",
"background": "transparent",
}
result = client.images.edit(**parameters)
image_base64 = result.data[0].b64_json
image_bytes = base64.b64decode(image_base64)
i = 0
while True:
path = f"output{i:02d}.png"
if not os.path.exists(path):
with open(path, "wb") as f:
f.write(image_bytes)
break
i += 1
Running with this image downloaded as your file name: https://media.istockphoto.com/id/860725344/photo/slow-loris.jpg?s=612x612&w=0&k=20&c=_z6G9Me_E7wUSbIKmlq9itEFEZd1ABSB95mVWOT4J0Q=
Provides this image (and the forum cannot display its transparency success; the mustard yellow is my “transparent”)
Actions for you to take:
- Ensure you are using the API correctly - you provided “moderation” which would fail an “edits” request, and images cannot be sent to the “generate new” endpoint. You provided “image” merely as a string and not the code for us to understand.
- Is attempting image input new to your code? Are actually using generations (which doesn’t take images as input) with a malformed request?
- (If the goal is creation of icons as a task, your actual prompt language could use improvement.)
