fix text-image parameters (#132)

Since the community api now [supports
kwargs,](huggingface/api-inference-community@414c5a0),
we can pass more diffusers parameters via the api, I choose a couple of
them that [make
sense](https://github.com/huggingface/diffusers/blob/fbc9a736dd5d8c20144ea44ef5266c87303ecacd/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py#L525-L542)

For the test, we could check the output image dimensions, but it would
require an import such as `canvas`, to convert the blob image to an
image and get its dimensions. wdyt @coyotte508

The test with negative_prompt will fail until this [PR is
merged](huggingface/api-inference-community#228),
as it is now it requires `negative_prompt: ["...negative prompt"]` to be
a list not a single value `negative_prompt: "...negative prompt"`

---------

Co-authored-by: coyotte508 <[email protected]>

Author: radames

README.md

@@ -92,8 +92,10 @@ await inference.translation({
 
 await inference.textToImage({
   inputs: 'award winning high resolution photo of a giant tortoise/((ladybird)) hybrid, [trending on artstation]',
-  negative_prompt: 'blurry',
   model: 'stabilityai/stable-diffusion-2',
+  parameters: {
+    negative_prompt: 'blurry',
+  }
 })
 ```
 

packages/inference/README.md

@@ -4,8 +4,6 @@ A Typescript powered wrapper for the Hugging Face Inference API. Learn more abou
 
 Check out the [full documentation](https://huggingface.co/docs/huggingface.js/inference/README) or try out a live [interactive notebook](https://observablehq.com/@huggingface/hello-huggingface-js-inference).
 
-
-
 ## Install
 
 ```console
@@ -18,7 +16,7 @@ pnpm add @huggingface/inference
 
 ## Usage
 
-❗**Important note:** Using an API key is optional to get started, however you will be rate limited eventually. Join [Hugging Face](https://huggingface.co/join) and then visit [access tokens](https://huggingface.co/settings/tokens) to generate your API key for **free**. 
+❗**Important note:** Using an API key is optional to get started, however you will be rate limited eventually. Join [Hugging Face](https://huggingface.co/join) and then visit [access tokens](https://huggingface.co/settings/tokens) to generate your API key for **free**.
 
 Your API key should be kept private. If you need to protect it in front-end applications, we suggest setting up a proxy server that stores the API key.
 
@@ -154,8 +152,10 @@ await hf.imageSegmentation({
 
 await hf.textToImage({
   inputs: 'award winning high resolution photo of a giant tortoise/((ladybird)) hybrid, [trending on artstation]',
-  negative_prompt: 'blurry',
   model: 'stabilityai/stable-diffusion-2',
+  parameters: {
+    negative_prompt: 'blurry',
+  }
 })
 ```
 

packages/inference/src/HfInference.ts

@@ -582,10 +582,28 @@ export type TextToImageArgs = Args & {
 	 */
 	inputs: string;
 
-	/**
-	 * An optional negative prompt for the image generation
-	 */
-	negative_prompt?: string;
+	parameters?: {
+		/**
+		 * An optional negative prompt for the image generation
+		 */
+		negative_prompt?: string;
+		/**
+		 * The height in pixels of the generated image
+		 */
+		height?: number;
+		/**
+		 * The width in pixels of the generated image
+		 */
+		width?: number;
+		/**
+		 * The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference.
+		 */
+		num_inference_steps?: number;
+		/**
+		 * Guidance scale: Higher guidance scale encourages to generate images that are closely linked to the text `prompt`, usually at the expense of lower image quality.
+		 */
+		guidance_scale?: number;
+	};
 };
 
 export type TextToImageReturn = Blob;

packages/inference/test/HfInference.spec.ts

@@ -397,10 +397,26 @@ describe.concurrent(
 		it("textToImage", async () => {
 			const res = await hf.textToImage({
 				inputs: "award winning high resolution photo of a giant tortoise/((ladybird)) hybrid, [trending on artstation]",
-				negative_prompt: "blurry",
 				model: "stabilityai/stable-diffusion-2",
 			});
+			expect(res).toBeInstanceOf(Blob);
+		});
 
+		it("textToImage with parameters", async () => {
+			const width = 512;
+			const height = 128;
+			const num_inference_steps = 10;
+
+			const res = await hf.textToImage({
+				inputs: "award winning high resolution photo of a giant tortoise/((ladybird)) hybrid, [trending on artstation]",
+				model: "stabilityai/stable-diffusion-2",
+				parameters: {
+					negative_prompt: "blurry",
+					width,
+					height,
+					num_inference_steps,
+				},
+			});
 			expect(res).toBeInstanceOf(Blob);
 		});
 	},

packages/inference/test/tapes.json

@@ -388,5 +388,45 @@
         "vary": "Origin, Access-Control-Request-Method, Access-Control-Request-Headers"
       }
     }
+  },
+  "5d075f1b72a1944d61597a7bc7c3251c3d6b7f3b4deadce4fb6de54964bfeb74": {
+    "url": "https://api-inference.huggingface.co/models/stabilityai/stable-diffusion-2",
+    "init": {
+      "headers": {
+        "Content-Type": "application/json"
+      },
+      "method": "POST"
+    },
+    "response": {
+      "body": "",
+      "status": 200,
+      "statusText": "OK",
+      "headers": {
+        "access-control-allow-credentials": "true",
+        "connection": "keep-alive",
+        "content-type": "image/jpeg",
+        "vary": "Origin, Access-Control-Request-Method, Access-Control-Request-Headers"
+      }
+    }
+  },
+  "9f9b1e0f1931cca2a77f4f39284fd1ca2e598b393d0c7ed36f05e66897cb762d": {
+    "url": "https://api-inference.huggingface.co/models/stabilityai/stable-diffusion-2",
+    "init": {
+      "headers": {
+        "Content-Type": "application/json"
+      },
+      "method": "POST"
+    },
+    "response": {
+      "body": "",
+      "status": 200,
+      "statusText": "OK",
+      "headers": {
+        "access-control-allow-credentials": "true",
+        "connection": "keep-alive",
+        "content-type": "image/jpeg",
+        "vary": "Origin, Access-Control-Request-Method, Access-Control-Request-Headers"
+      }
+    }
   }
 }