From d240eeb8dcccb6de2ec6d39a0162bdcff563503a Mon Sep 17 00:00:00 2001
From: Celina Hanouti <hanouticelina@gmail.com>
Date: Fri, 12 Dec 2025 15:38:01 +0100
Subject: [PATCH 1/2] add fal-ai support for image-text-to-image

---
 .../inference/src/lib/getProviderHelper.ts    |  9 +++--
 packages/inference/src/providers/fal-ai.ts    | 39 ++++++++++++++++++-
 2 files changed, 43 insertions(+), 5 deletions(-)

diff --git a/packages/inference/src/lib/getProviderHelper.ts b/packages/inference/src/lib/getProviderHelper.ts
index a733a37b65..1371d9530d 100644
--- a/packages/inference/src/lib/getProviderHelper.ts
+++ b/packages/inference/src/lib/getProviderHelper.ts
@@ -76,13 +76,14 @@ export const PROVIDERS: Record<InferenceProvider, Partial<Record<InferenceTask,
 		conversational: new Cohere.CohereConversationalTask(),
 	},
 	"fal-ai": {
-		"text-to-image": new FalAI.FalAITextToImageTask(),
-		"text-to-speech": new FalAI.FalAITextToSpeechTask(),
-		"text-to-video": new FalAI.FalAITextToVideoTask(),
-		"image-to-image": new FalAI.FalAIImageToImageTask(),
 		"automatic-speech-recognition": new FalAI.FalAIAutomaticSpeechRecognitionTask(),
+		"image-text-to-image": new FalAI.FalAIImageTextToImageTask(),
+		"image-to-image": new FalAI.FalAIImageToImageTask(),
 		"image-segmentation": new FalAI.FalAIImageSegmentationTask(),
 		"image-to-video": new FalAI.FalAIImageToVideoTask(),
+		"text-to-image": new FalAI.FalAITextToImageTask(),
+		"text-to-speech": new FalAI.FalAITextToSpeechTask(),
+		"text-to-video": new FalAI.FalAITextToVideoTask(),
 	},
 	"featherless-ai": {
 		conversational: new FeatherlessAI.FeatherlessAIConversationalTask(),
diff --git a/packages/inference/src/providers/fal-ai.ts b/packages/inference/src/providers/fal-ai.ts
index fe8fa40eb6..772d8d536f 100644
--- a/packages/inference/src/providers/fal-ai.ts
+++ b/packages/inference/src/providers/fal-ai.ts
@@ -21,7 +21,11 @@ import { isUrl } from "../lib/isUrl.js";
 import type { BodyParams, HeaderParams, InferenceTask, ModelId, RequestArgs, UrlParams } from "../types.js";
 import { delay } from "../utils/delay.js";
 import { omit } from "../utils/omit.js";
-import type { ImageSegmentationTaskHelper, ImageToImageTaskHelper } from "./providerHelper.js";
+import type {
+	ImageSegmentationTaskHelper,
+	ImageToImageTaskHelper,
+	ImageTextToImageTaskHelper,
+} from "./providerHelper.js";
 import {
 	type AutomaticSpeechRecognitionTaskHelper,
 	TaskProviderHelper,
@@ -37,6 +41,7 @@ import {
 	InferenceClientProviderOutputError,
 } from "../errors.js";
 import type { ImageToImageArgs, ImageToVideoArgs } from "../tasks/index.js";
+import type { ImageTextToImageArgs } from "../tasks/cv/imageTextToImage.js";
 import type { ImageSegmentationArgs } from "../tasks/cv/imageSegmentation.js";
 
 export interface FalAiQueueOutput {
@@ -284,6 +289,38 @@ export class FalAIImageToImageTask extends FalAiQueueTask implements ImageToImag
 	}
 }
 
+export class FalAIImageTextToImageTask extends FalAIImageToImageTask implements ImageTextToImageTaskHelper {
+	private useTextOnlyEndpoint = false;
+
+	constructor() {
+		super();
+		this.task = "image-text-to-image";
+	}
+
+	override makeRoute(params: UrlParams): string {
+		let model = params.model;
+		if (this.useTextOnlyEndpoint) {
+			// Strip last path segment: fal-ai/flux-2/edit => fal-ai/flux-2
+			model = model.split("/").slice(0, -1).join("/");
+			this.useTextOnlyEndpoint = false;
+		}
+		return super.makeRoute({ ...params, model });
+	}
+
+	override async preparePayloadAsync(args: ImageTextToImageArgs): Promise<RequestArgs> {
+		if (args.inputs) {
+			return super.preparePayloadAsync(args as ImageToImageArgs);
+		}
+		// No image provided: strip endpoint suffix and use `text-to-image` endpoint
+		this.useTextOnlyEndpoint = true;
+		return {
+			...omit(args, ["inputs", "parameters"]),
+			...(args.parameters as Record<string, unknown>),
+			prompt: args.parameters?.prompt,
+		} as RequestArgs;
+	}
+}
+
 export class FalAITextToVideoTask extends FalAiQueueTask implements TextToVideoTaskHelper {
 	task: InferenceTask;
 	constructor() {

From fcff53eb1f666c12194f62765b62b2345e1da72b Mon Sep 17 00:00:00 2001
From: Celina Hanouti <hanouticelina@gmail.com>
Date: Mon, 15 Dec 2025 10:48:12 +0100
Subject: [PATCH 2/2] add image-text-to-video for fal

---
 .../inference/src/lib/getProviderHelper.ts    |  1 +
 packages/inference/src/providers/fal-ai.ts    | 34 +++++++++++++++++++
 2 files changed, 35 insertions(+)

diff --git a/packages/inference/src/lib/getProviderHelper.ts b/packages/inference/src/lib/getProviderHelper.ts
index 1371d9530d..3c991913be 100644
--- a/packages/inference/src/lib/getProviderHelper.ts
+++ b/packages/inference/src/lib/getProviderHelper.ts
@@ -78,6 +78,7 @@ export const PROVIDERS: Record<InferenceProvider, Partial<Record<InferenceTask,
 	"fal-ai": {
 		"automatic-speech-recognition": new FalAI.FalAIAutomaticSpeechRecognitionTask(),
 		"image-text-to-image": new FalAI.FalAIImageTextToImageTask(),
+		"image-text-to-video": new FalAI.FalAIImageTextToVideoTask(),
 		"image-to-image": new FalAI.FalAIImageToImageTask(),
 		"image-segmentation": new FalAI.FalAIImageSegmentationTask(),
 		"image-to-video": new FalAI.FalAIImageToVideoTask(),
diff --git a/packages/inference/src/providers/fal-ai.ts b/packages/inference/src/providers/fal-ai.ts
index 772d8d536f..e1ae60105b 100644
--- a/packages/inference/src/providers/fal-ai.ts
+++ b/packages/inference/src/providers/fal-ai.ts
@@ -25,6 +25,7 @@ import type {
 	ImageSegmentationTaskHelper,
 	ImageToImageTaskHelper,
 	ImageTextToImageTaskHelper,
+	ImageTextToVideoTaskHelper,
 } from "./providerHelper.js";
 import {
 	type AutomaticSpeechRecognitionTaskHelper,
@@ -42,6 +43,7 @@ import {
 } from "../errors.js";
 import type { ImageToImageArgs, ImageToVideoArgs } from "../tasks/index.js";
 import type { ImageTextToImageArgs } from "../tasks/cv/imageTextToImage.js";
+import type { ImageTextToVideoArgs } from "../tasks/cv/imageTextToVideo.js";
 import type { ImageSegmentationArgs } from "../tasks/cv/imageSegmentation.js";
 
 export interface FalAiQueueOutput {
@@ -437,6 +439,38 @@ export class FalAIImageToVideoTask extends FalAiQueueTask implements ImageToVide
 	}
 }
 
+export class FalAIImageTextToVideoTask extends FalAIImageToVideoTask implements ImageTextToVideoTaskHelper {
+	private useTextOnlyEndpoint = false;
+
+	constructor() {
+		super();
+		this.task = "image-text-to-video";
+	}
+
+	override makeRoute(params: UrlParams): string {
+		let model = params.model;
+		if (this.useTextOnlyEndpoint) {
+			// Strip last path segment: fal-ai/ltxv-13b-098-distilled/image-to-video => fal-ai/ltxv-13b-098-distilled
+			model = model.split("/").slice(0, -1).join("/");
+			this.useTextOnlyEndpoint = false;
+		}
+		return super.makeRoute({ ...params, model });
+	}
+
+	override async preparePayloadAsync(args: ImageTextToVideoArgs): Promise<RequestArgs> {
+		if (args.inputs) {
+			return super.preparePayloadAsync(args as ImageToVideoArgs);
+		}
+		// No image provided: strip endpoint suffix and use `text-to-video` endpoint
+		this.useTextOnlyEndpoint = true;
+		return {
+			...omit(args, ["inputs", "parameters"]),
+			...(args.parameters as Record<string, unknown>),
+			prompt: args.parameters?.prompt,
+		} as RequestArgs;
+	}
+}
+
 export class FalAIAutomaticSpeechRecognitionTask extends FalAITask implements AutomaticSpeechRecognitionTaskHelper {
 	override prepareHeaders(params: HeaderParams, binary: boolean): Record<string, string> {
 		const headers = super.prepareHeaders(params, binary);