Feature request for DALLE-2: some kind of control panel showing how many generations/tokens we have left with counters/timers

I was just thinking it would be nice to have some sort of dashboard alongside our generations letting us know how many generations we have left, token counts, and (in the future) costs.

Also, guidance/explanation as to why certain keywords or phrases trigger the content filter. In most cases, many of us are triggering the filter without really understanding why and so we have to poke the black box to try and work around it. In most cases, we are certainly not trying to breach the guidelines, but it can be difficult to know why it’s so touchy.

15 Likes

Good idea, that should be implemented

1 Like

A few more thoughts:

  • A “test this prompt” feature would be nice. Rather than risk getting banned, it would be cool if there was a way we could quickly test our prompts for breaking the rules. Maybe it would just talk to the content filter (or whatever you’re using now) and just give us back an instant pass/fail with an explanation. I suppose anyone could do this with the content filter API.
  • A “DALLE2 prompt builder” - I have thought about doing this as a sort of chatbot. Basically it would ask you some boilerplate questions to describe your image, give you some options to pick from, and so on. At the end, the entire chat log would be fed through a finetuned model so that the “perfect” DALLE2 prompt could be generated. I might work on this anyways.
2 Likes

I’ve made a mockup of the control panel I’d like to see. I am going to be working on a DALLE assistant this morning. Basically it will be a chatbot that can ask you clarifying questions to help build a prompt. I’d also like to see some dropdowns available that can help control the style/setting/scene/colors/etc. By having a fill-in-the-blank approach to prompt-building, you will get users getting closer to exactly what they want. I forgot a “camera” drop down. That could include things like “wide angle” and “zoom lens” and all the other clever things folks on Reddit and Twitter are coming up with. There can also be an “effects” dropdown for things like “motion blur” and “sepia”.

A control panel like this will be absolutely critical (with the ability to save presets) for commercial applications. For instance, I am using this for storyboarding right now, which would be entirely different from a marketing department trying to make logos.

I guess what I’m saying is that you’ll need to eventually be making “DALLE Studio”.

@pw - any of this stuff on your feature roadmap? Also, please feel free to reach out to discuss if you want.

Here’s the chatbot first draft

Yes, the content filter raises the spectre of losing access to DALL-E - in all cases that I have been hit with the filter pop-up, I have been entering something innocent. It would be good to have some indication of where the lines are – and a potential for challenging the premise with some manual feedback.

1 Like

I second this.
Displaying how many Generations are left
Trigger word feedback

On addition for DALLE it could be interesting when you get a trigger word feedback if you could tell if you think that warning was: “fair”, “maybe fair” or “false”.
Even “maybe fair” could help to understand on which stuff uses struggle to understand what is allowed and what not. It seems a lot of people run into warnings innocent.

1 Like

Or better yet it could just suggest safer alternatives

1 Like

I entered an Artist’s name and got a flag… super sensitive

2 Likes

“Also, guidance/explanation as to why certain keywords or phrases trigger the content filter. In most cases, many of us are triggering the filter without really understanding why and so we have to poke the black box to try and work around it.”

Especially second this. But I’m nervous about getting my account banned, so I’m too fearful to “poke the black box”.

2 Likes