AI Image & Video API Providers, September 2026: FAL vs Replicate vs Direct APIs

Short answer (checked September 23, 2026): use an aggregator — FAL or Replicate — when you want many models behind one API key. Go direct to the model maker when one model family does most of your work: BytePlus ModelArk for Seedream and Seedance, Google for Veo and Nano Banana, OpenAI for GPT Image, Alibaba for Wan and Qwen Image, and xAI for Grok Imagine. Direct was the cheapest or tied route for most first-party models we checked; the main exception was GPT Image 2 at medium and high quality, which cost less on Replicate.

What changed since July: OpenAI's Sora API shuts down on September 24, 2026, with no replacement. DALL-E 3 shut down on May 12. Google removed Imagen 4 from the Gemini API on August 17 (it remains on Vertex AI). New models since then include Seedance 2.5, Wan 3.0, GPT Image 2.5, and Seedream 5.0 Flash.

For per-image and per-second prices across every route, see the AI image & video API pricing comparison.

Choosing a provider solves model access; it does not create a repeatable creative operation. Teamday gives you an image and video studio plus visual AI employees with a workspace, missions, files, and human review. Connect your own provider keys and usage bills to your provider account. Starter costs $99 a month. Generate images in the Teamday studio →

---

Provider Overview

---

Provider Deep Dives

1. FAL — the broadest aggregator

What it is: a generative-media API platform with one client and key for image, video, audio, and 3D models from many labs. In FAL's own State of Generative Media report, 50% of surveyed organizations used FAL for image generation and 44% for video.

Pricing highlights (FAL model pages, September 23, 2026):

Strengths:

• Broadest catalog of the platforms compared here
• Cheapest route found for Kling v3 (half of Replicate's price)
• Same price as the model maker for Veo 3.1 and Wan 3.0

Weaknesses:

• Seedance costs about twice BytePlus's direct price; Seedream 5.0 Pro and Nano Banana Pro also cost more than direct
• Free credits work only in the Sandbox and Playground, not through the API
• The general pricing page lagged the model pages in September; check the specific model page

Best for: teams that want many models behind one integration, and Kling video.

import { fal } from "@fal-ai/client";

fal.config({ credentials: process.env.FAL_KEY });

const result = await fal.subscribe("fal-ai/flux-2-flex", {
  input: {
    prompt: "A professional product photo of wireless headphones",
    image_size: "landscape_16_9"
  }
});

console.log(result.data.images[0].url);

---

2. Replicate — the developer-friendly aggregator

What it is: an API platform for running hosted and community models, with support for deploying your own models. 15% of organizations in FAL's survey used Replicate for images and 22% for video.

Pricing highlights (Replicate model pages, September 23, 2026):

Strengths:

• Cheapest aggregator route found for Seedream 5.0 Pro, Seedance 2.0 and GPT Image 2 at medium or high quality
• Custom model deployment for your own weights
• Clear per-model pricing on each model page

Weaknesses:

• Kling v3 costs twice FAL's price
• Some models FAL carries (for example Wan 3.0 and Ideogram V4) were not found on Replicate

Best for: prototyping, custom model hosting, and teams that already run open models there.

import Replicate from "replicate";

const replicate = new Replicate();

const output = await replicate.run("black-forest-labs/flux-2-pro", {
  input: {
    prompt: "A professional product photo of wireless headphones",
    aspect_ratio: "16:9"
  }
});

console.log(output);

---

3. BytePlus ModelArk — the direct route for Seedream and Seedance

What it is: ByteDance's international API for its Seedream image and Seedance video models.

Pricing highlights (BytePlus ModelArk pricing, September 23, 2026):

Per-second figures are BytePlus's own examples for a 5-second 16:9 clip. BytePlus listed 25% off Seedance 2.0 Fast and 60% off 2.0 Mini for enterprise users from August 7 to October 7, 2026.

Strengths:

• Cheapest route found for Seedream and Seedance — roughly half FAL's Seedance price
• Seedance 2.5 generates 4–30 second clips with native voice and music

Weaknesses:

• ByteDance models only
• Token billing makes cost estimates harder; log your own usage

Best for: cinematic marketing video and high-volume Seedream image work.

---

4. Google (Gemini API and Vertex AI) — Veo and Nano Banana

What it is: Google's direct APIs for Veo video, Nano Banana image models (Gemini image), and Imagen 4 on Vertex AI. In FAL's survey, 56% of organizations used Google AI Studio for video — the highest of any provider.

Pricing highlights (Gemini API pricing and Vertex AI pricing, September 23, 2026):

Batch runs Nano Banana 2 and Nano Banana Pro at half price. The original Nano Banana (gemini-2.5-flash-image, $0.039) shuts down on October 2, 2026.

Strengths:

• Veo 3.1 Lite is the cheapest route found to video with generated audio
• Veo 3.1 Fast with audio costs less direct ($0.10/s) than on FAL or Replicate ($0.15/s)

Weaknesses:

• Model and endpoint churn: Veo 2 and 3.0 shut down on the Gemini API on June 30, and Imagen 4 on August 17
• Some options, such as Veo without audio and Imagen 4, are Vertex-only

Best for: Veo video, Nano Banana images, and teams already on Google Cloud.

---

5. OpenAI — text-in-image specialist, leaving video

What it is: OpenAI's direct API for GPT Image models. Sora video is ending.

Pricing highlights (OpenAI pricing and image-generation calculator, September 23, 2026):

Deprecations (OpenAI deprecations): DALL-E 2 and 3 shut down May 12, 2026; the Videos API and Sora 2 shut down September 24, 2026; gpt-image-1 shuts down October 23, 2026; gpt-image-1.5, gpt-image-1-mini and chatgpt-image-latest shut down December 1, 2026, with gpt-image-2 as the replacement.

Strengths:

• Strong text rendering for infographics, diagrams, and labels
• Image editing with reference images

Weaknesses:

• No video API after September 24
• High quality settings cost far more than low ($0.211 vs $0.006 for GPT Image 2)

Best for: text-heavy images and precise edits.

import OpenAI from "openai";

const openai = new OpenAI();

const response = await openai.images.generate({
  model: "gpt-image-2",
  prompt: "A professional infographic showing '5 Steps to Success' with icons",
  size: "1536x1024",
  quality: "medium"
});

// GPT Image models return base64 image data
console.log(response.data[0].b64_json?.slice(0, 32));

---

6. Alibaba Model Studio — Wan and Qwen Image

What it is: Alibaba Cloud's API for Wan video and image models and Qwen Image.

Pricing highlights (Model Studio pricing, international region, September 23, 2026):

Wan 3.0 bills input plus output duration. Free quotas apply in the Singapore region only.

Best for: low-cost video at 480p–720p and text-capable images at $0.03–$0.04.

---

7. xAI — Grok Imagine

What it is: xAI's image and video models, billed per image and per second by resolution (xAI models).

Best for: inexpensive images and short video without an aggregator.

---

8. Runway API — editing-oriented video

What it is: Runway's developer API, billed in credits at $0.01 each (Runway API pricing).

Best for: video editing and restyling workflows; Gen-4 Turbo for cheap drafts.

---

Head-to-Head: Where Each Model Is Cheapest

Who uses which provider

FAL's State of Generative Media report (Volume 1) surveyed which infrastructure providers organizations use. Respondents could name several, so the figures do not add up to 100%.

The same report found a median of 14 different models in enterprise production deployments. That is the practical argument for an aggregator — or for a tool that handles several providers for you. Note that FAL published the report.

---

Decision Matrix

---

The Teamday Shortcut: Skip the API Shopping

Comparing APIs assumes you are building an app. If you are a marketer, founder, or operations lead who wants the finished image or video, API keys, rate limits, and model-switching code are overhead.

Teamday's image and video studios put these models behind one prompt box:

• Images: Seedream 5.0 and 4.5, GPT Image 2.5 (Flare and Sunburst) and GPT Image 2, Gemini 3.1 Flash Image, Flux 2 Flex, Qwen Image 2.0, Wan 2.7 Image, and Grok Imagine Image
• Video: Seedance 2.5, 2.0 and 2.0 Fast, Kling 3 Pro and 2.6 Pro, Wan 2.6, HappyHorse 1.0, and Grok Imagine Video
• Providers: ByteDance ModelArk, OpenAI, Google, FAL, Alibaba Model Studio, and xAI

Connect your own provider keys and each generation bills to your provider account at that provider's price. Any Teamday agent can generate images and video from chat — including Iris, the AI image generator, Reel, the AI video producer, and Nova, the AI marketing agent. They are part of Teamday's AI creative team and work alongside Maya, the AI content creator, on the AI marketing team.

"Reel, cut me a 30-second brand film for our landing page." "Iris, generate 10 Instagram carousel variations for this launch."

---

Conclusion

Our recommendation: use one aggregator for breadth (FAL for most catalogs, Replicate if you host custom models), and add a direct key for the one or two model families that make up most of your volume. Price the route, not the platform. Move anything on Sora or DALL-E now.

---

Related Resources

• AI Image & Video API Pricing 2026
• FAL.AI vs Replicate: Detailed Comparison
• Best AI Image Models 2026
• Best AI Video Models 2026
• Best AI Models in 2026: Frontier Models Compared
• Changelog: Seedance 2.0 for every team
• Iris — AI image generator and Reel — AI video producer