Best Sora 2 Alternatives for App Clips and the Video API

Sora is gone, and nothing official takes its place. OpenAI closed the Sora app on April 26, 2026, shut down the Sora 2 API on September 24, 2026, and its Sora 2 model page now says "there is no one-to-one replacement API."

If you made clips in the Sora app, start with Google Vids, which has let any Google account generate videos for free since September 23, or with the 50 free daily credits in Google Flow. If you called the Sora API, Google's own documentation names Gemini Omni Flash as its default video model, at about $0.10 per second of 720p video, the same rate sora-2 charged.

Everything else depends on what those two can't give you: single takes longer than 10 seconds, a cheaper 1080p rate, commercial rights on a small plan, or a character that stays the same across shots. No tool below imports Sora projects, cameos, or video IDs, so plan each one as a fresh setup. For shutdown dates, app exports, and leftover credits, see what still works after the Sora 2 API shutdown. All prices are in US dollars.

Pick a replacement by what you used Sora for

App and plan prices are as of September 30, 2026; Google plan prices are US prices.

If you made clips in the Sora app

Free: Google Vids and Google Flow

Since September 23, 2026, anyone with a Google or Google Workspace account can generate videos in Google Vids at no cost with Gemini Omni 1.1 Flash, according to Google's announcement. Open vids.new in a desktop browser and choose "Create AI videos." You can set a duration, extend scenes, and generate new scenes in 1080p. Google doesn't publish a free limit; for more generations on a personal account, it points to its paid Google AI plans. Vids is a web app with no API, and every clip carries an invisible SynthID watermark.

Google Flow gives every account 50 credits a day, with or without a subscription. Unused daily credits don't roll over. What 50 credits buy depends on the model:

• An 8-second Omni Flash clip at 720p costs 12 credits, so a free day covers four of them (48 credits).
• Veo 3.1 Lite costs 10 credits per generation for non-Ultra accounts, or five clips a day.
• Veo 3.1 Fast costs 20 credits per generation, or two clips a day.

Credits are charged per generation, and one request can produce two videos. Without a subscription, video generation may be unavailable during peak hours, roughly 2 to 5 PM UTC (10 a.m. to 1 p.m. EDT). Paid Google AI plans add monthly Flow credits on top of the daily 50: AI Plus at $4.99 a month adds 200, AI Pro at $19.99 adds 1,000, and AI Ultra at $99.99 adds 10,000. At 12 credits per clip, AI Pro's extra credits come to about 83 eight-second Omni clips at 720p a month.

Paid apps for more volume or commercial use

Among the plans that state commercial rights on their pricing pages, Kling Standard is the cheapest. Its FAQ lets paid subscribers use generated content commercially, as long as it isn't used to build a competing product. Pika Starter costs about the same but excludes a commercial license. Hailuo's $8.40 rate is billed as $100.80 for the year up front. Luma costs more at the entry level but includes commercial use on every plan.

SuperGrok makes sense if you already pay for Grok; 1080p video needs SuperGrok Plus at $100 a month. The Grok Imagine guide covers its video limits and API. If your Sora videos were mostly a talking presenter, HeyGen is closer to that job: its free plan makes 3 videos a month of up to 1 minute, and Creator costs $29 a month. For per-clip costs across more plans, see how much an AI video generator costs per month and per clip.

If you called the Sora API

Google's video generation overview is direct about which of its two models to start with: "Use Gemini Omni Flash as your default model for video generation." The overview cites video coherence, reasoning across text, image, audio, and video inputs, character consistency, and multi-turn conversational editing, and it reserves Veo 3.1 for "scene extension, last-frame control, or integration with legacy pipelines." The Omni API page is narrower on audio: uploading audio references isn't supported yet, and audio inside a video reference is ignored. Omni Flash is generally available on the paid tier of the Gemini API, while the Veo 3.1 model IDs still carry a -preview suffix.

Neither Google model has a free API tier. Google charges only for videos that finish generating, and a Veo video blocked by safety filters isn't charged. For the few free options on the API side, see whether a free AI video API exists.

If your Sora jobs run on Microsoft Foundry, its sora-2 preview deployment retires on October 15, 2026, so the same move applies there within two weeks.

Per-second prices next to Sora

Google and BytePlus prices are as of September 26, 2026; the others are as of September 30, 2026. Pika dates its own figures September 18, 2026, and they include a 5% platform fee.

Omni Flash's rate comes from Google's token math: 5,792 output tokens per second of 720p video at $17.50 per million tokens is about $0.101 a second.

To see what a switch does to a budget, multiply the rate by 800 seconds, which is 100 clips of 8 seconds each:

• At 720p, sora-2 cost $80. Omni Flash and Veo 3.1 Fast come to about the same, $80. Veo 3.1 Lite is $40, MiniMax-H3 at 768P is $64, and Kling 3.0 is $67.20 without audio or $100.80 with it. Grok Imagine 1.5 is $112, Seedance 2.5 is $184.80, and Veo 3.1 Standard is $320.
• At 1080p, the same 100 clips on Sora 2 Pro cost $560. Veo 3.1 Lite comes to $64, Veo 3.1 Fast to $96, and Kling 3.0 to $134.40 with audio. Grok Imagine 1.5 is $200, Veo 3.1 Standard is $320, and Seedance 2.5 is $455.20.

These are list prices for clips that finish generating. A clip you reject still counts, so track spending per clip you actually use during a trial.

Beyond Google: what each alternative is for

Seedance 2.5 is the one to test when single takes or reference counts matter. ByteDance's launch post says it generates up to 30 seconds at a time, extends over several rounds, accepts up to 30 image, 10 video, and 10 audio references, and supports timestamp-level editing. BytePlus lists the model as dreamina-seedance-2-5-260628 with 4- to 30-second output. These are the vendor's own claims, and ByteDance notes weaker physical plausibility and multi-subject interaction. At 720p it costs more than twice Omni Flash's rate. BytePlus is the international contract; mainland China uses a separate Volcano Engine contract. For resolution-by-resolution costs, see Seedance 2.5 API pricing.

Kling 3.0 sells API capacity in prepaid units at a $0.14 list price per unit, with a $9.8 trial pack of 100 units. Its per-second rate at 720p without audio is lower than Omni Flash's. Kling's API plans don't include Kling 4.0.

Grok Imagine Video 1.5 accepts text, image, and audio input. Its 480p rate is $0.08 a second, but its 720p rate is higher than Omni Flash's.

MiniMax-H3 charges $0.08 a second at 768P, a step above 720p, and $0.13 at 2K, with no plan or membership required.

Runway, Luma, and Pika each price their own models through their own APIs. Runway bills credits at $0.01 each, Luma prices whole 5- or 10-second clips, and Pika adds a $10 monthly membership on top of per-second rates.

None of these vendors describes its API as compatible with Sora requests. Each needs its own client code, and the next section covers what changes for the Google pair.

Moving Sora code to Veo 3.1 or Omni Flash

The two Google models run on different APIs, so choose one before you rewrite the worker. The Veo 3.1 documentation says Veo is only available through the generateContent API. The Omni Flash documentation runs on the newer Interactions API.

A minimal Veo job with the Python SDK. It follows the pattern in Google's Veo docs and reads your key from GEMINI_API_KEY:

import time
from google import genai
from google.genai import types

client = genai.Client()

operation = client.models.generate_videos(
    model="veo-3.1-fast-generate-preview",
    prompt="A ceramic mug on a desk as morning light moves across it.",
    config=types.GenerateVideosConfig(aspect_ratio="9:16"),
)
print("Operation:", operation.name)  # store this before you start polling

while not operation.done:
    time.sleep(10)
    operation = client.operations.get(operation)

video = operation.response.generated_videos[0]
client.files.download(file=video.video, destination="mug.mp4")

The same shot on Omni Flash, followed by an extension in a second turn:

import base64
from google import genai

client = genai.Client()

first = client.interactions.create(
    model="gemini-omni-1.1-flash",
    input="A ceramic mug on a desk as morning light moves across it.",
    response_format={"type": "video", "aspect_ratio": "9:16"},
)
with open("mug.mp4", "wb") as f:
    f.write(base64.b64decode(first.output_video.data))

second = client.interactions.create(
    model="gemini-omni-1.1-flash",
    previous_interaction_id=first.id,
    input="Continue the scene: a hand reaches in and lifts the mug.",
)
with open("mug-extended.mp4", "wb") as f:
    f.write(base64.b64decode(second.output_video.data))

output_video is a convenience field in the SDK only; raw REST responses put the video in the interaction's steps. For videos larger than 4 MB, Google recommends delivery: "uri" in response_format, then polling the file until its state is ACTIVE before you download it.

The model call is the small part of the change. These are the parts that break quietly:

1. Job records. A Sora video ID means nothing now. Store the Veo operation name or the Omni interaction ID next to your own job ID before you wait on anything, so a failed poll can resume instead of paying for a second generation.
2. Downloads. Veo deletes generated files after 2 days. Copy every finished video into your own storage right away and treat Google's link as temporary.
3. Length promises. If your product offered single clips longer than 8 seconds, pick the replacement on purpose: Omni extensions up to 40 seconds, Veo extensions at 720p, Seedance 2.5's 30-second takes, or edited cuts. Silently returning 8 seconds breaks the product even when the call succeeds.
4. Resolution. Veo generates 1080p and 4K directly, but only at 8 seconds. Omni Flash's 1080p and 4K outputs are upscaled.
5. Prompts. Omni doesn't support negative prompts, system instructions, or temperature. Move "no text on screen"-style instructions into the prompt itself. Veo's docs list English as fully supported and other languages as not evaluated.
6. People and regions. If you serve users in the EU, UK, Switzerland, or the Middle East and North Africa, test requests that include people from those regions before launch.
7. Watermarks. Both models add an invisible SynthID watermark to every output, and Google's SynthID verification can detect it. Plan any AI-content labeling with that in mind.

For turning a still image into a clip with either model, see how to create a video with Nano Banana through Gemini Omni or Veo 3.1.

Sora characters and cameos have no import path

Sora's cameos and saved characters, along with remix and storyboard projects, have no export format that another tool accepts, and no vendor documents an import for them. Getting a consistent character back means rebuilding it from reference material in the new model:

• Veo 3.1 takes up to 3 reference images of a person, character, or product to keep its appearance. Reference-image generations must be 8 seconds long and allow only adult people (allow_adult).
• Omni Flash accepts subject reference images, plus up to 3 video references of up to 3 seconds each, which Google says "work best with likenesses." Uploaded images of certain recognizable people aren't supported.
• Seedance 2.5 accepts up to 30 image references per request, according to ByteDance's launch post.

A practical starting set is a few clean stills of the character from the videos you exported from Sora: front, profile, and full body, in the lighting you want. If you have short Sora clips, Omni Flash can also edit or extend an uploaded video of 10 seconds or less, except in the EEA, Switzerland, and the UK. Veo can only extend videos that Veo generated.

FAQ

Is there an official replacement for Sora?

No. OpenAI's Sora 2 model page says "There is no one-to-one replacement API," and its deprecation notice lists no successor. Gemini Omni Flash and Veo 3.1 Fast cost about what sora-2 did at 720p, $0.10 a second. Neither accepts Sora requests unchanged.

What is the best free alternative to Sora?

Start with Google Vids, which generates Omni 1.1 videos at 1080p for any Google account and publishes no limit. Google Flow is the other free option: its 50 daily credits cover four 8-second Omni Flash clips at 720p. Both are web apps. Google's video models have no free API tier.

Is Veo 4 available?

No. As of September 30, 2026, Google's model list, Veo documentation, and API pricing page show Veo 3.1 as the newest Veo model. Google has not listed a Veo 4 model, whatever some third-party pages call it.

Which API is cheapest to switch to from Sora 2?

At 720p, Veo 3.1 Lite charges $0.05 a second, MiniMax-H3 $0.08 at 768P, Kling 3.0 $0.084 without audio, and Omni Flash and Veo 3.1 Fast about $0.10, compared with sora-2 at $0.10. For 1080p work that ran on Sora 2 Pro at $0.70 a second, Veo 3.1 Lite charges $0.08, Veo 3.1 Fast $0.12, and Kling 3.0 $0.112 to $0.168. Pika 2.5 starts at $0.04 a second but needs a $10 monthly membership.

Can I still use Sora through Azure?

Only briefly. Microsoft Foundry's sora-2 preview deployment retires on October 15, 2026, so it only buys Azure customers two more weeks. The Sora 2 API shutdown guide lists every channel and date.