@huggingface/inference 4.0.6 → 4.1.0

package/src/providers/replicate.ts

@@ -16,9 +16,16 @@
  */
 import { InferenceClientProviderOutputError } from "../errors.js";
 import { isUrl } from "../lib/isUrl.js";
-import type { BodyParams, HeaderParams, UrlParams } from "../types.js";
+import type { BodyParams, HeaderParams, RequestArgs, UrlParams } from "../types.js";
 import { omit } from "../utils/omit.js";
-import { TaskProviderHelper, type TextToImageTaskHelper, type TextToVideoTaskHelper } from "./providerHelper.js";
+import {
+	TaskProviderHelper,
+	type ImageToImageTaskHelper,
+	type TextToImageTaskHelper,
+	type TextToVideoTaskHelper,
+} from "./providerHelper.js";
+import type { ImageToImageArgs } from "../tasks/cv/imageToImage.js";
+import { base64FromBytes } from "../utils/base64FromBytes.js";
 export interface ReplicateOutput {
 	output?: string | string[];
 }
@@ -152,3 +159,57 @@ export class ReplicateTextToVideoTask extends ReplicateTask implements TextToVid
 		throw new InferenceClientProviderOutputError("Received malformed response from Replicate text-to-video API");
 	}
 }
+
+export class ReplicateImageToImageTask extends ReplicateTask implements ImageToImageTaskHelper {
+	override preparePayload(params: BodyParams<ImageToImageArgs>): Record<string, unknown> {
+		return {
+			input: {
+				...omit(params.args, ["inputs", "parameters"]),
+				...params.args.parameters,
+				input_image: params.args.inputs, // This will be processed in preparePayloadAsync
+			},
+			version: params.model.includes(":") ? params.model.split(":")[1] : undefined,
+		};
+	}
+
+	async preparePayloadAsync(args: ImageToImageArgs): Promise<RequestArgs> {
+		const { inputs, ...restArgs } = args;
+
+		// Convert Blob to base64 data URL
+		const bytes = new Uint8Array(await inputs.arrayBuffer());
+		const base64 = base64FromBytes(bytes);
+		const imageInput = `data:${inputs.type || "image/jpeg"};base64,${base64}`;
+
+		return {
+			...restArgs,
+			inputs: imageInput,
+		};
+	}
+
+	override async getResponse(response: ReplicateOutput): Promise<Blob> {
+		if (
+			typeof response === "object" &&
+			!!response &&
+			"output" in response &&
+			Array.isArray(response.output) &&
+			response.output.length > 0 &&
+			typeof response.output[0] === "string"
+		) {
+			const urlResponse = await fetch(response.output[0]);
+			return await urlResponse.blob();
+		}
+
+		if (
+			typeof response === "object" &&
+			!!response &&
+			"output" in response &&
+			typeof response.output === "string" &&
+			isUrl(response.output)
+		) {
+			const urlResponse = await fetch(response.output);
+			return await urlResponse.blob();
+		}
+
+		throw new InferenceClientProviderOutputError("Received malformed response from Replicate image-to-image API");
+	}
+}

package/src/snippets/templates.exported.ts

@@ -5,6 +5,7 @@ export const templates: Record<string, Record<string, Record<string, string>>> =
       "basic": "async function query(data) {\n\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"application/json\",\n{% if billTo %}\n\t\t\t\t\"X-HF-Bill-To\": \"{{ billTo }}\",\n{% endif %}\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n\tconst result = await response.json();\n\treturn result;\n}\n\nquery({ inputs: {{ providerInputs.asObj.inputs }} }).then((response) => {\n    console.log(JSON.stringify(response));\n});",
       "basicAudio": "async function query(data) {\n\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"audio/flac\",\n{% if billTo %}\n\t\t\t\t\"X-HF-Bill-To\": \"{{ billTo }}\",\n{% endif %}\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n\tconst result = await response.json();\n\treturn result;\n}\n\nquery({ inputs: {{ providerInputs.asObj.inputs }} }).then((response) => {\n    console.log(JSON.stringify(response));\n});",
       "basicImage": "async function query(data) {\n\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"image/jpeg\",\n{% if billTo %}\n\t\t\t\t\"X-HF-Bill-To\": \"{{ billTo }}\",\n{% endif %}\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n\tconst result = await response.json();\n\treturn result;\n}\n\nquery({ inputs: {{ providerInputs.asObj.inputs }} }).then((response) => {\n    console.log(JSON.stringify(response));\n});",
+      "imageToImage": "const image = fs.readFileSync(\"{{inputs.asObj.inputs}}\");\n\nasync function query(data) {\n\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"image/jpeg\",\n{% if billTo %}\n\t\t\t\t\"X-HF-Bill-To\": \"{{ billTo }}\",\n{% endif %}\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: {\n\t\t\t\t\"inputs\": `data:image/png;base64,${data.inputs.encode(\"base64\")}`,\n\t\t\t\t\"parameters\": data.parameters,\n\t\t\t}\n\t\t}\n\t);\n\tconst result = await response.json();\n\treturn result;\n}\n\nquery({ \n\tinputs: image,\n\tparameters: {\n\t\tprompt: \"{{ inputs.asObj.parameters.prompt }}\",\n\t}\n}).then((response) => {\n    console.log(JSON.stringify(response));\n});",
       "textToAudio": "{% if model.library_name == \"transformers\" %}\nasync function query(data) {\n\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"application/json\",\n{% if billTo %}\n\t\t\t\t\"X-HF-Bill-To\": \"{{ billTo }}\",\n{% endif %}\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n\tconst result = await response.blob();\n    return result;\n}\n\nquery({ inputs: {{ providerInputs.asObj.inputs }} }).then((response) => {\n    // Returns a byte object of the Audio wavform. Use it directly!\n});\n{% else %}\nasync function query(data) {\n\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"application/json\",\n\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n    const result = await response.json();\n    return result;\n}\n\nquery({ inputs: {{ providerInputs.asObj.inputs }} }).then((response) => {\n    console.log(JSON.stringify(response));\n});\n{% endif %} ",
       "textToImage": "async function query(data) {\n\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"application/json\",\n{% if billTo %}\n\t\t\t\t\"X-HF-Bill-To\": \"{{ billTo }}\",\n{% endif %}\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n\tconst result = await response.blob();\n\treturn result;\n}\n\n\nquery({ {{ providerInputs.asTsString }} }).then((response) => {\n    // Use image\n});",
       "textToSpeech": "{% if model.library_name == \"transformers\" %}\nasync function query(data) {\n\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"application/json\",\n{% if billTo %}\n\t\t\t\t\"X-HF-Bill-To\": \"{{ billTo }}\",\n{% endif %}\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n\tconst result = await response.blob();\n    return result;\n}\n\nquery({ text: {{ inputs.asObj.inputs }} }).then((response) => {\n    // Returns a byte object of the Audio wavform. Use it directly!\n});\n{% else %}\nasync function query(data) {\n\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"application/json\",\n\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n    const result = await response.json();\n    return result;\n}\n\nquery({ text: {{ inputs.asObj.inputs }} }).then((response) => {\n    console.log(JSON.stringify(response));\n});\n{% endif %} ",
@@ -16,6 +17,7 @@ export const templates: Record<string, Record<string, Record<string, string>>> =
       "basicImage": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst data = fs.readFileSync({{inputs.asObj.inputs}});\n\nconst output = await client.{{ methodName }}({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n\tdata,\n\tmodel: \"{{ model.id }}\",\n\tprovider: \"{{ provider }}\",\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(output);",
       "conversational": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst chatCompletion = await client.chatCompletion({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n{{ inputs.asTsString }}\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n\nconsole.log(chatCompletion.choices[0].message);",
       "conversationalStream": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nlet out = \"\";\n\nconst stream = client.chatCompletionStream({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n{{ inputs.asTsString }}\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n\nfor await (const chunk of stream) {\n\tif (chunk.choices && chunk.choices.length > 0) {\n\t\tconst newContent = chunk.choices[0].delta.content;\n\t\tout += newContent;\n\t\tconsole.log(newContent);\n\t}\n}",
+      "imageToImage": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst data = fs.readFileSync(\"{{inputs.asObj.inputs}}\");\n\nconst image = await client.imageToImage({\n{% if endpointUrl %}\n\tendpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n\tprovider: \"{{provider}}\",\n\tmodel: \"{{model.id}}\",\n\tinputs: data,\n\tparameters: { prompt: \"{{inputs.asObj.parameters.prompt}}\", },\n}{% if billTo %}, {\n\tbillTo: \"{{ billTo }}\",\n}{% endif %});\n/// Use the generated image (it's a Blob)\n// For example, you can save it to a file or display it in an image element\n",
       "textToImage": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst image = await client.textToImage({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n\tparameters: { num_inference_steps: 5 },\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n/// Use the generated image (it's a Blob)",
       "textToSpeech": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst audio = await client.textToSpeech({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n// Use the generated audio (it's a Blob)",
       "textToVideo": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst video = await client.textToVideo({\n{% if endpointUrl %}\n    endpointUrl: \"{{ endpointUrl }}\",\n{% endif %}\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n}{% if billTo %}, {\n    billTo: \"{{ billTo }}\",\n}{% endif %});\n// Use the generated video (it's a Blob)"
@@ -27,6 +29,7 @@ export const templates: Record<string, Record<string, Record<string, string>>> =
   },
   "python": {
     "fal_client": {
+      "imageToImage": "{%if provider == \"fal-ai\" %}\nimport fal_client\nimport base64\n\ndef on_queue_update(update):\n    if isinstance(update, fal_client.InProgress):\n        for log in update.logs:\n           print(log[\"message\"])\n\nwith open(\"{{inputs.asObj.inputs}}\", \"rb\") as image_file:\n    image_base_64 = base64.b64encode(image_file.read()).decode('utf-8')\n\nresult = fal_client.subscribe(\n    \"fal-ai/flux-kontext/dev\",\n    arguments={\n        \"prompt\": f\"data:image/png;base64,{image_base_64}\",\n        \"image_url\": \"{{ providerInputs.asObj.inputs }}\",\n    },\n    with_logs=True,\n    on_queue_update=on_queue_update,\n)\nprint(result)\n{%endif%}\n",
       "textToImage": "{% if provider == \"fal-ai\" %}\nimport fal_client\n\n{% if providerInputs.asObj.loras is defined and providerInputs.asObj.loras != none %}\nresult = fal_client.subscribe(\n    \"{{ providerModelId }}\",\n    arguments={\n        \"prompt\": {{ inputs.asObj.inputs }},\n        \"loras\":{{ providerInputs.asObj.loras | tojson }},\n    },\n)\n{% else %}\nresult = fal_client.subscribe(\n    \"{{ providerModelId }}\",\n    arguments={\n        \"prompt\": {{ inputs.asObj.inputs }},\n    },\n)\n{% endif %} \nprint(result)\n{% endif %} "
     },
     "huggingface_hub": {
@@ -36,7 +39,7 @@ export const templates: Record<string, Record<string, Record<string, string>>> =
       "conversational": "completion = client.chat.completions.create(\n    model=\"{{ model.id }}\",\n{{ inputs.asPythonString }}\n)\n\nprint(completion.choices[0].message) ",
       "conversationalStream": "stream = client.chat.completions.create(\n    model=\"{{ model.id }}\",\n{{ inputs.asPythonString }}\n    stream=True,\n)\n\nfor chunk in stream:\n    print(chunk.choices[0].delta.content, end=\"\") ",
       "documentQuestionAnswering": "output = client.document_question_answering(\n    \"{{ inputs.asObj.image }}\",\n    question=\"{{ inputs.asObj.question }}\",\n    model=\"{{ model.id }}\",\n) ",
-      "imageToImage": "# output is a PIL.Image object\nimage = client.image_to_image(\n    \"{{ inputs.asObj.inputs }}\",\n    prompt=\"{{ inputs.asObj.parameters.prompt }}\",\n    model=\"{{ model.id }}\",\n) ",
+      "imageToImage": "with open(\"{{ inputs.asObj.inputs }}\", \"rb\") as image_file:\n   input_image = image_file.read()\n\n# output is a PIL.Image object\nimage = client.image_to_image(\n    input_image,\n    prompt=\"{{ inputs.asObj.parameters.prompt }}\",\n    model=\"{{ model.id }}\",\n) ",
       "importInferenceClient": "from huggingface_hub import InferenceClient\n\nclient = InferenceClient(\n{% if endpointUrl %}\n    base_url=\"{{ baseUrl }}\",\n{% endif %}\n    provider=\"{{ provider }}\",\n    api_key=\"{{ accessToken }}\",\n{% if billTo %}\n    bill_to=\"{{ billTo }}\",\n{% endif %}\n)",
       "questionAnswering": "answer = client.question_answering(\n    question=\"{{ inputs.asObj.question }}\",\n    context=\"{{ inputs.asObj.context }}\",\n    model=\"{{ model.id }}\",\n) ",
       "tableQuestionAnswering": "answer = client.table_question_answering(\n    query=\"{{ inputs.asObj.query }}\",\n    table={{ inputs.asObj.table }},\n    model=\"{{ model.id }}\",\n) ",
@@ -55,7 +58,7 @@ export const templates: Record<string, Record<string, Record<string, string>>> =
       "conversational": "def query(payload):\n    response = requests.post(API_URL, headers=headers, json=payload)\n    return response.json()\n\nresponse = query({\n{{ providerInputs.asJsonString }}\n})\n\nprint(response[\"choices\"][0][\"message\"])",
       "conversationalStream": "def query(payload):\n    response = requests.post(API_URL, headers=headers, json=payload, stream=True)\n    for line in response.iter_lines():\n        if not line.startswith(b\"data:\"):\n            continue\n        if line.strip() == b\"data: [DONE]\":\n            return\n        yield json.loads(line.decode(\"utf-8\").lstrip(\"data:\").rstrip(\"/n\"))\n\nchunks = query({\n{{ providerInputs.asJsonString }},\n    \"stream\": True,\n})\n\nfor chunk in chunks:\n    print(chunk[\"choices\"][0][\"delta\"][\"content\"], end=\"\")",
       "documentQuestionAnswering": "def query(payload):\n    with open(payload[\"image\"], \"rb\") as f:\n        img = f.read()\n        payload[\"image\"] = base64.b64encode(img).decode(\"utf-8\")\n    response = requests.post(API_URL, headers=headers, json=payload)\n    return response.json()\n\noutput = query({\n    \"inputs\": {\n        \"image\": \"{{ inputs.asObj.image }}\",\n        \"question\": \"{{ inputs.asObj.question }}\",\n    },\n}) ",
-      "imageToImage": "def query(payload):\n    with open(payload[\"inputs\"], \"rb\") as f:\n        img = f.read()\n        payload[\"inputs\"] = base64.b64encode(img).decode(\"utf-8\")\n    response = requests.post(API_URL, headers=headers, json=payload)\n    return response.content\n\nimage_bytes = query({\n{{ providerInputs.asJsonString }}\n})\n\n# You can access the image with PIL.Image for example\nimport io\nfrom PIL import Image\nimage = Image.open(io.BytesIO(image_bytes)) ",
+      "imageToImage": "with open(\"{{inputs.asObj.inputs}}\", \"rb\") as image_file:\n    image_base_64 = base64.b64encode(image_file.read()).decode('utf-8')\n\ndef query(payload):\n    with open(payload[\"inputs\"], \"rb\") as f:\n        img = f.read()\n        payload[\"inputs\"] = base64.b64encode(img).decode(\"utf-8\")\n    response = requests.post(API_URL, headers=headers, json=payload)\n    return response.content\n\nimage_bytes = query({\n{{ providerInputs.asJsonString }}\n})\n\n# You can access the image with PIL.Image for example\nimport io\nfrom PIL import Image\nimage = Image.open(io.BytesIO(image_bytes)) ",
       "importRequests": "{% if importBase64 %}\nimport base64\n{% endif %}\n{% if importJson %}\nimport json\n{% endif %}\nimport requests\n\nAPI_URL = \"{{ fullUrl }}\"\nheaders = {\n    \"Authorization\": \"{{ authorizationHeader }}\",\n{% if billTo %}\n    \"X-HF-Bill-To\": \"{{ billTo }}\"\n{% endif %}\n}",
       "tabular": "def query(payload):\n    response = requests.post(API_URL, headers=headers, json=payload)\n    return response.content\n\nresponse = query({\n    \"inputs\": {\n        \"data\": {{ providerInputs.asObj.inputs }}\n    },\n}) ",
       "textToAudio": "{% if model.library_name == \"transformers\" %}\ndef query(payload):\n    response = requests.post(API_URL, headers=headers, json=payload)\n    return response.content\n\naudio_bytes = query({\n    \"inputs\": {{ inputs.asObj.inputs }},\n})\n# You can access the audio with IPython.display for example\nfrom IPython.display import Audio\nAudio(audio_bytes)\n{% else %}\ndef query(payload):\n    response = requests.post(API_URL, headers=headers, json=payload)\n    return response.json()\n\naudio, sampling_rate = query({\n    \"inputs\": {{ inputs.asObj.inputs }},\n})\n# You can access the audio with IPython.display for example\nfrom IPython.display import Audio\nAudio(audio, rate=sampling_rate)\n{% endif %} ",

package/src/tasks/cv/imageToImage.ts

@@ -3,6 +3,7 @@ import { resolveProvider } from "../../lib/getInferenceProviderMapping.js";
 import { getProviderHelper } from "../../lib/getProviderHelper.js";
 import type { BaseArgs, Options } from "../../types.js";
 import { innerRequest } from "../../utils/request.js";
+import { makeRequestOptions } from "../../lib/makeRequestOptions.js";
 
 export type ImageToImageArgs = BaseArgs & ImageToImageInput;
 
@@ -18,5 +19,6 @@ export async function imageToImage(args: ImageToImageArgs, options?: Options): P
 		...options,
 		task: "image-to-image",
 	});
-	return providerHelper.getResponse(res);
+	const { url, info } = await makeRequestOptions(args, providerHelper, { ...options, task: "image-to-image" });
+	return providerHelper.getResponse(res, url, info.headers as Record<string, string>);
 }