Negative prompts for the inference api

jetpackjules · February 18, 2023, 12:40am

Hello, for the free inference api, Is there a way to add negative prompts? I heard that huggingface recently added this feature, and was wondering how to include it in my request…

Thanks in advance!

jetpackjules · February 27, 2023, 7:10pm

EDIT: this would be for the Stable Diffusion inference api…

Any updates?

radames · February 27, 2023, 9:10pm

hi @jetpackjules,

Yes we’ve recently add the negative prompt to the API

Here is a request example in Javascript

fetch(
  "https://api-inference.huggingface.co/models/stabilityai/stable-diffusion-2",
  {
    method: "POST",
    headers: {
      "content-type": "application/json",
    },
    body: JSON.stringify({
      inputs:
        "award winning high resolution photo of a giant tortoise/((ladybird)) hybrid, [trending on artstation]",
      negative_prompt: "blurry",
    }),
  }
)
  .then((res) => res.blob())
  .then((blob) => {
    const tab = window.open((target = "_blank"));
    tab.location.href = window.URL.createObjectURL(blob);
  });

github.com

huggingface/api-inference-community/blob/6f39af264188099a8929cd2a0148b85c7690b0c4/docker_images/diffusers/app/pipelines/text_to_image.py#L57


      
                  self.ldm.unet.load_attn_procs(
                      model_id, use_auth_token=os.getenv("HF_API_TOKEN")
                  )
          
          
    if isinstance(self.ldm, (StableDiffusionPipeline, AltDiffusionPipeline)):
                  self.ldm.scheduler = DPMSolverMultistepScheduler.from_config(
                      self.ldm.scheduler.config
                  )
          
          
def __call__(
              self, inputs: str, inference_steps=25, negative_prompt=None, **kwargs
          ) -> "Image.Image":
              """
              Args:
                  inputs (:obj:`str`):
                      a string containing some text
              Return:
                  A :obj:`PIL.Image` with the raw image representation as PIL.
              """
          
          
    if kwargs:

Blockquote

jetpackjules · February 27, 2023, 9:53pm

Awesome, Thankyou very much!

jetpackjules · February 27, 2023, 10:19pm

Does this also work for Stable Diffusion 1-5, or just 2?

radames · February 27, 2023, 10:37pm

Hi @jetpackjules atm the pipeline is very basic and only supports 512px inference, however we could change that, you can send your opinion needs here Document usage of text-to-image · Issue #152 · huggingface/api-inference-community · GitHub

radames · February 27, 2023, 10:45pm

@jetpackjules sorry, it’s automatic, you have to set the sample size on the vae config file

Compare

jetpackjules · February 27, 2023, 11:06pm

For the negative prompt?

sjsjd · August 22, 2023, 7:17am

Hey, how to download an image from this data and display it in nodejs?

sjsjd · August 22, 2023, 7:18am

I use your code