Interior Video Generation

Multi-model AI video generation for interior design — cinematic walkthroughs, product spotlights, before/after reveals, and social reels.

Overview

The generate-interior-video-v2 edge function routes to the optimal AI video model based on video type or an explicit model override. It handles the full lifecycle: model selection, async polling, timeout handling, and storage.

Edge Function: generate-interior-video-v2

Models

Model	Credits	Speed	Best For
`veo-2`	30	~2-3 min	Cinematic walkthroughs, floor-plan flythroughs
`kling-v3.0`	20	~60-90s	Product spotlights, before/after, social reels
`wan2.1-i2v-720p`	12	~90s	Budget option, general purpose
`runway-gen4-turbo`	40	~2-3 min	Premium quality output

Kling uses the native Kling SDK. Wan and Runway use Replicate. Veo-2 uses Google's API.

Video Types & Auto-Routing

If you pass video_type without model, the function auto-selects:

Video Type	Auto-Selected Model
`walkthrough`	`veo-2`
`floorplan_flythrough`	`veo-2`
`product_spotlight`	`kling-v3.0`
`before_after`	`kling-v3.0`
`social_reel`	`kling-v3.0`

API

POST /functions/v1/generate-interior-video-v2
Authorization: Bearer <jwt>
Content-Type: application/json

Request Body

Field	Type	Required	Description
`image_url`	string	Yes	Source room image URL
`prompt`	string	Yes	Text description of motion/camera movement
`video_type`	string	No*	`walkthrough` \| `product_spotlight` \| `before_after` \| `floorplan_flythrough` \| `social_reel`
`model`	string	No*	Explicit model override (see table above)
`aspect_ratio`	string	No	`16:9` (default) \| `9:16` \| `1:1`
`duration`	number	No	Duration in seconds (model-dependent max)
`workspace_id`	string	Yes	For credit deduction
`user_id`	string	No	For server-to-server calls

*One of video_type or model is required.

Aspect Ratios

Ratio	Use Case
`16:9`	Standard landscape (walkthroughs, web)
`9:16`	Vertical (Instagram Reels, TikTok)
`1:1`	Square (Instagram feed)

Response

Synchronous (generation completed within ~55s)

{
  "success": true,
  "video_url": "https://bgbavxtjlbvgplozizxu.supabase.co/storage/v1/object/public/generation-videos/...",
  "model": "kling-v3.0",
  "credits_used": 20,
  "duration_ms": 42000
}

Async (generation exceeded 55s timeout)

{
  "success": true,
  "status": "processing",
  "job_id": "uuid",
  "prediction_id": "replicate-prediction-id",
  "model": "veo-2",
  "credits_used": 30,
  "message": "Video generation in progress. Poll /functions/v1/generate-interior-video-v2?job_id=..."
}

Poll with GET /functions/v1/generate-interior-video-v2?job_id=<uuid> until status = 'completed'.

Async Polling Pattern

Replicate-backed models (Wan, Runway) and Veo-2 can take 2-5 minutes. If the edge function's 55s polling window elapses:

prediction_id is stored in the generation_videos table
Response returns job_id and status: 'processing'
Frontend polls the endpoint with ?job_id= on a 5-10s interval
Once Replicate/Veo reports succeeded, video is downloaded → re-uploaded to Supabase Storage → generation_videos updated

Storage

Generated videos are stored at:

generation-videos/{job-uuid}.mp4

Credit Handling

Credits are debited upfront before the model call. They are non-refundable (generation infrastructure costs are incurred regardless of output quality).

Database

Generated videos are tracked in the generation_videos table:

Column	Description
`id`	Job UUID
`prediction_id`	Replicate/external prediction ID
`status`	`processing` \| `completed` \| `failed`
`video_url`	Final Supabase Storage URL
`model`	Model used
`credits_debited`	Credits consumed
`workspace_id`	Owner workspace

Social Video Generation

For social-media-specific short videos, the separate generate-social-video edge function provides a simplified interface:

Model	Credits
`kling-1.6-pro`	15
`veo-2` (delegated to `generate-interior-video-v2`)	30

POST /functions/v1/generate-social-video
{
  "prompt": "...",
  "image_url": "...",
  "model": "kling-1.6-pro",
  "aspect_ratio": "9:16",
  "workspace_id": "uuid"
}

Last Updated: March 2026