AI endpoint for Image generation which allows image+prompt -> image

Gemini API
,
1

I was playing around a lot with AI, recently I saw an amazing progress with CHAT GPT and Dall-E, I was trying to make a small project to create an AI home remodeling POC and was managed to do that but what I was looking for is an API where I can upload image and prompt which tells the AI what changes modifications I would like to do to the image, I was not able to find something similar, I end up with sending image get the info out of the image as a prompt and recompose it as new prompt for Dall-E the result sometime surprising me but most of the time not :frowning:
you can give it a try if you like, my home remodeling ai project
I found a model called imagen by google but I am not sure how can I use it and if I can use it at all, did any one here tried it?
Is there any model which can get image+prompt and generate image based on that image and given prompt?

1 Like
2

Regarding the “surprising” responses, have you fine-tuned your current model?

3

No, can you please share more info about fine tuning?

5

Check out Gemini’s doc on fine tuning

6

The Imagen models can certainly do what you’re asking, at least some of the models of it can.

Imagen models are available through the Vertex AI API (tho there are rumors it is coming to the AI Studio API as well). The biggest issue is that they are limited access, and you need to contact Google Cloud Support to get more than basic access.

See this page for an overview and this one for more about the API and how to request access.

1 Like
7

thank you for your clarification, it’s Interesting, I have no idea how is the vertex api works, might be worth waiting for the rumours to come true :star::sweat_smile:

9

I would like to share my results with you, imagen was approved for my case, the only way I found useful for my case is to map the classes which I want to edit and the only model which is editable is imagen2, one of the thing which botters me is the fact you can add up to 5 class ids, in some case of my use case I will need more then 5 :frowning:
But results are quite nice for now, my POC start to look nice you welcome to test it out if you want:Modifieed by moderator

1 Like