[Inference] get the raw response for textToImage

[krampstudio]

Allow to get the full response for text-to-image inference, instead of directly the blob or the url:

const res = await textToImage({ inputs:  "A dog with a baseball cap" },{ outputType: "json" });

we have the following ouput (depends on the provider)

{
  "images": [
    {
      "url": "data:image/png;base64,iVBORw0KGgoAAAAN......rkJggg==",
      "width": 1024,
      "height": 768,
      "content_type": "image/png"
    }
  ],
  "timings": {
    "inference": 1.2445615287870169
  },
  "seed": 982874191,
  "has_nsfw_concepts": [
    false
  ],
  "prompt": "A dog with a baseball cap"
} 

needed for (internal https://github.com/huggingface-internal/moon-landing/pull/14421)

[coyotte508]

Why do we need that?

In the internal link, we only do const imageBlob = isImage(res) ? res : new Blob([res], { type: "image/png" }); so it seems we could just call textToImage with output: "blob" ?

[krampstudio]

Why do we need that?

In the internal link, we only do const imageBlob = isImage(res) ? res : new Blob([res], { type: "image/png" }); so it seems we could just call textToImage with output: "blob" ?

I've updated the internal link, we need the full response for educational purpose: we show it to the user.

And more generaly, it can be usefull for developers to access the full response.

[julien-c]

no objection on my side