Problem Description

I am currently using Replicate.com as my host, however, I would like to switch to either Lambda Labs or Google Cloud to reduce server costs. You should create an API which can interact with anything-v4.0 model and be as easy to use as Replicate. I can input text and receive an image back. An image to image option would also be nice.

• Link to anything-v4.0 model: https://huggingface.co/andite/anything-v4.0

• Link to current Replicate model: https://replicate.com/cjwbw/anything-v3-better-vae/versions/09a5805203f4c12da649ec1923bb7729517ca25fcac790e640eaa9ed66573b65

Acceptance Criteria

• An easy to use API which takes in http requests which utilizes anything-v4.0 to create images from text or images from images

• I use NodeJS for my backend. The API should be nodeJS compatible. I don't want to make my requests in Python. The request should look like this:
  {prompt: "Anime person", width: 512, height: 512, negative_prompt: "Negative prompt"}
  Or
  {input: image}
  It should return an image back.

• The NSFW content filter which causes the model to return black images should be removed.

• Should employ anythingv3 vae or some other vae which negates the bleached images returned by the model.

• Link to different vae discussion: https://huggingface.co/andite/anything-v4.0/discussions/47

• Should be deployed on Lambda Labs or Google Cloud (cheapest option is preferred)

• Should use T4 GPU, A100 GPU, or some price equivalent GPU

• Should be autoscaling compatible. So with increased usage, I should be able to get more instances. I do believe this is a default behavior of the platforms though.

Note:

• If you do a good job, I have a lot more models for you to deploy which I will of course pay your for

Strong Applicant

• Notes their relevant experience in publishing models
• Links to their Github