Veo 4 API

Veo 4 on fal-ai/veo4/text-to-video runs at roughly $0.75 per second at 720p 24fps, which puts a full 8 second cinematic beat at about $6.00. A 1080p render on the same endpoint costs closer to $1.25 per second, so the same 8 second job lands near $10. Image to video pricing on fal-ai/veo4/image-to-video tracks the same per second scale. Rates move quarterly, so check fal.ai/pricing before you quote a client. For cost math on a specific shot count, multiply your planned seconds by the resolution rate and add a 10 percent buffer for retries.

Sign up at fal.ai, generate a key from the dashboard, and drop it into your environment as FAL_KEY. From there, @fal-ai/client picks it up via fal.config({ credentials: process.env.FAL_KEY }) and every call to fal.subscribe('fal-ai/veo4/text-to-video', ...) authenticates automatically. In production, never hardcode the key; inject it through Vercel, Railway, or whichever runtime you use. Free signup credits cover your first few renders, and you can watch spend in real time on the cost dashboard before wiring Veo 4 into a client pipeline.

On fal-ai/veo4/text-to-video, the ceiling is 1080p and the floor is 720p. 720p jobs cost near $0.75 per second and are the right default for social cuts and internal review. 1080p jobs are the pick for broadcast deliverables, and the per second rate rises accordingly. 4K upscaling is a post step today, not a native option. Aspect ratios cover 16:9, 9:16, and 1:1, which covers landscape hero cuts, vertical social, and square for feeds. Frame rate is fixed at 24fps, which matches cinematic delivery without a frame rate conversion pass.

A single call to fal-ai/veo4/text-to-video renders up to 8 seconds. To produce longer sequences, chain shots by feeding the last frame of your first render into fal-ai/veo4/image-to-video and prompting the next beat. Veo 4 holds shot coherence well across a chain because its camera and grade controls stay consistent when you pin the same lens vocabulary across prompts. Plan your edits around the 8 second ceiling; for a 30 second spot, budget four Veo 4 renders plus a stitch pass in your NLE or in ffmpeg.

Install fal-client, set FAL_KEY in your environment, and call fal_client.subscribe('fal-ai/veo4/text-to-video', arguments={'prompt': '...', 'aspect_ratio': '16:9', 'duration_seconds': 8}). The Python SDK mirrors the TypeScript @fal-ai/client one to one, so the same input schema applies. For async jobs, use fal_client.submit and poll the queue with the returned request id. Both SDKs support webhook callbacks so you can push queue completion events to your own service instead of polling. Python is the right pick when Veo 4 sits inside a Django or FastAPI pipeline.

Yes. fal-ai/veo4/image-to-video accepts a still frame as input and animates it with a prompt, preserving the subject and grade from the source image while adding motion, camera path, and optional dialogue cues. Pricing matches fal-ai/veo4/text-to-video per second. This is the endpoint to use when you want to animate a client-approved still, extend a plate from a physical shoot, or chain consecutive 8 second beats by feeding the final frame of one render into the next. Aspect ratio inherits from the source unless you explicitly override it.

fal-ai/veo4/text-to-video runs through fal.ai's async queue, so failures return a structured error with a request id you can inspect via fal.queue.status. Default timeouts land near 10 minutes for 1080p renders; if a job exceeds that, the queue marks it failed and your credits for that submission are refunded. The right retry pattern is to catch the error, log the request id, and resubmit with a slightly softened prompt if the failure was a content filter hit. For transient queue errors, a single retry with exponential backoff resolves most cases.

Yes. fal-ai/veo4/text-to-video accepts a webhook_url on fal.queue.submit, and fal.ai posts the completion payload to that URL the moment the render lands. This is the right pattern for any production pipeline where you do not want a long lived connection. Your webhook handler receives the video URL, the request id, and the billing seconds used. Pair this with a signed secret on your endpoint to verify the callback origin. Webhooks cut queue polling overhead to zero and scale cleanly when you run dozens of renders in parallel.

Veo 4 on fal-ai/veo4/text-to-video leads on cinematic grading, shot coherence, and broadcast grade color. Kling 3.0 Pro is close on fidelity, cheaper per second, and supports 10 second shots, which is the pick when you need a longer beat. Seedance 2.0 is the volume play at roughly $0.18 per second, ideal for content calendars and quick variants. Grok Imagine is the fastest and loosest, best for idea passes and prototyping. HappyHorse 1.0 specializes in stylized character performance. Pick Veo 4 when the deliverable is a hero cut that has to grade cleanly.

One fal API key covers 600+ models, so fal-ai/veo4/text-to-video sits next to every other image, video, and audio endpoint you use. Infrastructure is serverless, with an async queue, webhook callbacks, and regional PoPs that reduce first frame latency. You ship in TypeScript or Python through a single @fal-ai/client SDK, watch spend live on the cost dashboard, get free signup credits to validate Veo 4 before committing a budget, and hit Slack or Discord when a render stalls at 2am. That combination is why production teams standardize on fal.ai rather than stitching vendor accounts together.