Seeds gave me the ability to create multi-part, long term projects that required character consistency. This change also impacts the symbiotic relationship between chatgpt and dall-e. I don’t want to go to midjourney, but this nerf is really messed up.
1 Like
1 Like
It changed the description before which added noise- that change doesn’t appear to be new only implemented differently imo. The issue is for me is this change completely nerfed continuity of characters. The whole point of Dall-e is to generate the art in your head in a format other people can see. It will work fine if you want to put a bulldog on a surfboard just screwing around, but it’s not in the dust heap for my purposes. 
1 Like
It’s not effective for my use. The changes have nerfed my process to the point where I’m considering moving to MidJourney. Until the update, I was completely sold on the chatgpt/dall-e interoperability- as it functions now - it’s not worth 2 cents for my purposes.
1 Like
Do you use the referenced_image_ids?
Image generation ID works when your 2nd image’s description refers to the 1st one via referenced_image_ids.
It only works in the same conversation.
I’m aware it’s session limited now. That’s not the issue. The issue is you can’t use image generation id the same way you could seeds for cross project continuity. Does that make sense?
The good news is that DALL-E will support “image reference” (which is different from referenced_image_ids) in the next version.
See DALL-E 3 AMA:
2 Likes
I’ll test with it but I’ve tried using the refs id and multiple workarounds with horrendous results. I hope they fix it.
1 Like
Yes, I had wonderful results prior to the changes. Using the refs method did not improve results. Even with explicit instruction to only change x out of the initial prompt and use the ref image, the images look completely different now. When I ask the AI why, it’s not quite sure why it’s not working and sometimes tries to tell me to use the seed method which no longer works. It’s frustrating.
3 Likes
Hey @cassiellm or @PaulBellow, any idea when that new version with stronger image reference capabilities will be coming out? This is a critical feature I want for character building and storytelling. If it’s not coming to OpenAI products, I may have to switch providers, which would be a shame. I’ve scoured the forums and some people mention the gen_id, but it doesn’t seem to work that well.
FYI - I’m most interested in using the API, and don’t want to have to worry about sessions ideally
They said weeks now months… Hang tight…