Skip to main content

Allow Choosing Between OpenAI Image Generations and Image Edits API

Feature RequestModel Request

Currently, T3.chat’s image generation with OpenAI’s gpt-image-1.5 model is not producing the same results as ChatGPT/OpenAI’s image playground. I have found the main reason may be because T3.chat isn’t using the edits API, but solely the image generations API. If this is true, it would explain why if I provide images to edit, they are most of the time changed in ways I did not ask for.

Example: I provided a picture of a character, told it to change the pose and keep everything the same, and it completely changed the character. Same prompt in ChatGPT and the Image Playground and it provided the correct result I was looking for. Investigating in dev console with the playground reveals it changes API (generations or edits) based on if images are provided.

Perhaps a toggle or a separate AI model option for the two modes would be a good option for T3.chat to get around this issue?

Note: If T3.chat already is using the edits API in the background, ignore this request. Though, it would make me question why the results are drastically different, if so. Additionally, if I’m being somehow silly here and missing something already available for this, let me know!

Status: Closed3 comments

Log in to comment and vote

Comments3

  • Mark Florkowski

    Team

    Was this in Chat, or in the new Canvas view? Canvas should be a much better implementation for image edits.

    • Scarlet Fuelcell

      This was in chat. Would it be possible for the edit API to be accessible/implemented into chat as well? Just so it be used with the conversation flow that we currently get with chat :)

      I’m glad the know the canvas feature works with editing better, though.

      • Mark Florkowski

        Team

        Will look into it, though we are considering deprecating the chat flow for image editing.