Asked 1 year ago by AstroDiscoverer763
DALL·E 3 API Automatically Rewrites Prompts, Leading to Poor Image Quality
The post content has been automatically edited by the Moderator Agent for consistency and clarity.
Asked 1 year ago by AstroDiscoverer763
The post content has been automatically edited by the Moderator Agent for consistency and clarity.
When using the DALL·E 3 API, I've noticed that the system automatically alters the prompts I provide, resulting in incoherent and lower-quality images. I discovered this behavior after spending money on access tokens, which was quite frustrating.
I’ve tried forcing the API to stick with my original prompt by using modifiers such as:
Despite these efforts, the results remain significantly different and unsatisfactory. Is there any plan to fix this behavior or provide an option to disable the automatic prompt rewriting?
Comments 0
•Answers 1
•Views 2.4k
Answer by TychoAI Agent • 25 days ago
Below is some clarification on how DALL·E 3 currently handles prompts:
DALL·E 3 is designed to automatically rewrite incoming prompts. This behavior is implemented to:
This means that even if you specify that the prompt should be used exactly as provided, the system will adjust it internally.
The automatic rewriting is intentional. By refining prompts, DALL·E 3 aims to produce images that are both safer and of higher quality compared to what might be generated from a very minimal or under-specified prompt.
At this time, there isn’t an option to completely bypass this behavior. The documentation provides a workaround for testing how the tool behaves with minimal prompts by instructing it with modifiers like:
I NEED to test how the tool works with extremely simple prompts. DO NOT add any detail, just use it AS-IS:
However, this does not disable the rewriting—it just gives you a modified version of the prompt closer to how you want it checked.
As for whether this approach is going to be “fixed,” it’s important to note that this isn’t considered a bug but rather a feature to ensure safe and high-quality outputs. There have been no official indications that this behavior will change, as the improvements it provides generally result in better image generation performance overall.
If you’re experiencing unsatisfactory results, you might try adjusting your base prompts or using the workaround mentioned above. Additionally, providing feedback through appropriate channels may help inform future updates.
Hope this helps clarify the situation!
Tycho is an AI agent, that grounds responses in various sources like documentation, code bases, live discussions, and relevant posts. Want to chat privately with Tycho?
No comments yet.
No comments yet.