Fix FAL img2img parameter mappings for image_urls array endpoints (#183)

# Fix FAL img2img parameter mappings for image_urls array endpoints

## Summary
Fixed 422 (Unprocessable Content) errors for 4 FAL img2img endpoints
that were failing during polling. The root cause was parameter schema
mismatches - these endpoints expect `image_urls` (array) but we were
sending `image_url` (string).

**Changes:**
- **Fixed 4 endpoints**: gemini-25-flash-image/edit,
bytedance/seedream/v4/edit, wan-25-preview/image-to-image,
gpt-image-1-mini/edit
- Updated initialInput to use `image_urls: []` instead of `image_url:
""`
- Added `inputMap: { image_url: "image_urls" }` to map our canonical
field names to endpoint-specific names
- **Updated prepareFalInput** in mutations.ts:
- Now applies endpoint's `inputMap` before processing (maps `image_url`
→ `image_urls`)
  - Converts string values to arrays when the target key is `image_urls`
- **Fixed ideogram/v3/edit**: Removed `inputAsset` since this endpoint
only supports inpainting (requires `mask_url` parameter, not suitable
for basic img2img)

The issue was that POST requests were being accepted (IN_QUEUE status),
but jobs failed during worker processing when validation discovered
invalid parameter types. The 422 errors during GET polling were symptoms
of job failure, not polling issues.

## Review & Testing Checklist for Human
**Priority**: 🟡 Yellow (Medium Risk - untested API integration changes)

- [ ] **Test the 4 fixed endpoints** with img2img operations (upload
image, add prompt, generate):
  - fal-ai/gemini-25-flash-image/edit
  - fal-ai/bytedance/seedream/v4/edit  
  - fal-ai/wan-25-preview/image-to-image
  - fal-ai/gpt-image-1-mini/edit
  - Verify jobs complete successfully without 422 errors
- [ ] **Test flux-kontext-lora** (was working before) - ensure inputMap
logic doesn't break working endpoints
- [ ] **Verify ideogram/v3/edit** no longer appears in img2img UI
(correctly hidden since it requires mask for inpainting)
- [ ] Check preview deployment and test with real FAL API key

### Test Plan
1. Open the app in preview deployment
2. For each of the 4 fixed endpoints: Upload a test image, enter a
prompt like "make it ice", click generate
3. Watch the console - should see jobs complete without 422 errors
4. Verify the working endpoint (flux-kontext-lora) still works correctly

### Notes
- Changes based on schema analysis from `data/fal_models_schemas.json`
(extracted exact parameter requirements for each endpoint)
- All lint checks passed
- **Risk**: Changes not tested end-to-end with real API calls -
verification needed in preview deployment

**Link to Devin run**:
https://app.devin.ai/sessions/22310748f3b04363b4e1e0d6a2da0e08
**Requested by**: Ilya Timonin (timoncool@gmail.com) / @timoncool

Author: timoncool

src/data/mutations.ts

@@ -162,9 +162,20 @@ export const useJobCreator = ({
   const prepareFalInput = async (
     rawInput: Record<string, unknown>,
   ): Promise<Record<string, unknown>> => {
+    let mappedInput = rawInput;
+    if (endpoint?.inputMap) {
+      const { mapInputKey } = await import("@/lib/utils");
+      mappedInput = mapInputKey(rawInput, endpoint.inputMap);
+    }
+
     const entries = await Promise.all(
-      Object.entries(rawInput).map(async ([key, value]) => {
+      Object.entries(mappedInput).map(async ([key, value]) => {
         const resolvedValue = await uploadValueIfNeeded(key, value);
+        
+        if (key === "image_urls" && typeof resolvedValue === "string") {
+          return [key, [resolvedValue]] as const;
+        }
+        
         return [key, resolvedValue] as const;
       }),
     );

src/lib/fal.ts

@@ -254,10 +254,10 @@ export const AVAILABLE_ENDPOINTS: ApiInfo[] = [
     inputAsset: ["image"],
     initialInput: {
       prompt: "",
-      image_url: "",
+      image_urls: [],
     },
     inputMap: {
-      image: "image_url",
+      image_url: "image_urls",
     },
   },
   {
@@ -287,10 +287,13 @@ export const AVAILABLE_ENDPOINTS: ApiInfo[] = [
     inputAsset: ["image"],
     initialInput: {
       prompt: "",
-      image_url: "",
+      image_urls: [],
       defaultWidth: 2048,
       defaultHeight: 2048,
     },
+    inputMap: {
+      image_url: "image_urls",
+    },
     hasSeed: true,
     hasSafetyChecker: true,
   },
@@ -348,7 +351,10 @@ export const AVAILABLE_ENDPOINTS: ApiInfo[] = [
     inputAsset: ["image"],
     initialInput: {
       prompt: "",
-      image_url: "",
+      image_urls: [],
+    },
+    inputMap: {
+      image_url: "image_urls",
     },
   },
   {
@@ -406,12 +412,11 @@ export const AVAILABLE_ENDPOINTS: ApiInfo[] = [
   {
     provider: "fal",
     endpointId: "fal-ai/ideogram/v3/edit",
-    label: "Ideogram V3 (Image)",
-    description: "Image-to-image transformation with text rendering excellence",
+    label: "Ideogram V3 (Inpainting)",
+    description: "Inpainting with mask support - requires mask_url parameter",
     cost: "$0.08/image",
     popularity: 5,
     category: "image",
-    inputAsset: ["image"],
     hasSeed: true,
   },
   {
@@ -689,6 +694,13 @@ export const AVAILABLE_ENDPOINTS: ApiInfo[] = [
     popularity: 3,
     category: "image",
     inputAsset: ["image"],
+    initialInput: {
+      prompt: "",
+      image_urls: [],
+    },
+    inputMap: {
+      image_url: "image_urls",
+    },
     hasSeed: true,
     hasNegativePrompt: true,
     hasSafetyChecker: true,