Skip to content

How to use Sora: the complete 2026 guide to OpenAI's video model

Last updated: November 2026 Maintained by ToolChase Methodology

TL;DR

OpenAI discontinued the Sora consumer app and sora.com on April 26, 2026, so you can no longer sign up or generate videos there, and ChatGPT Plus and Pro no longer include Sora access. OpenAI shut down the Sora 2 API on September 24, 2026 (it had cost around $0.10 per second of 720p video). For everyone else, the practical move is a current alternative: Kling, Runway, or our full list of free AI video generators. The guide below is kept as a reference record.

Update: Sora was discontinued on April 26, 2026

OpenAI shut down the Sora web app and mobile app. OpenAI shut down the Sora 2 API on September 24, 2026, the last step in retiring Sora entirely. Sections below that describe sora.com sign-up and ChatGPT plan access reflect the product as it worked before the shutdown.

Independently researched Editorial standards

If you want to know how to use Sora, the OpenAI text-to-video model whose last release was Sora 2, this guide is now a record rather than a how-to: OpenAI shut the consumer product down on April 26, 2026 and the API closed on September 24, 2026. It covers how access worked, what the API offered until it closed, and which alternatives to move to.

Quick navigation

What Sora is, and what is new in Sora 2

Sora was OpenAI's flagship text-to-video model. You wrote a description of a scene, characters, environment, camera move, lighting, and the model returned a short, photorealistic video clip that respected basic physics and continuity. The first version launched publicly at the end of 2024. The last production model was Sora 2, which is what ran on sora.com until the shutdown.

The Sora 2 release closed most of the gaps that frustrated early users of the original model. Faces looked less plastic, hands and small objects broke less often, and physics interactions like splashing water, falling cloth, or moving crowds looked more natural. Sora 2 also handled synchronized audio in supported regions, so a clip of a beach could return with realistic wave sound and ambient noise rather than silence.

If you have heard people search for "sora 2 how to use" or "sora how to use sora", they are almost always describing this newer build. The workflow was identical to Sora 1, same dashboard, same prompt box, but the output quality, audio, and editing tools (remix, extend, storyboard) all lived in this generation. When this guide says "Sora", it means Sora 2 unless we explicitly call out the older release.

For a broader landscape view of where Sora sat relative to other models, our best AI video generators of 2026 roundup covers the alternatives in depth, and our full Sora review tracks the changing limits and feature set.

OpenAI Sora access: required plan tiers

This is how access worked until the April 26, 2026 shutdown: you paid for a qualifying ChatGPT subscription, then signed in at sora.com. That path is now closed. The last remaining access was the Sora 2 API, which OpenAI shut down on September 24, 2026; everything below is kept as a historical reference.

  • → Free ChatGPT: No Sora access. You could read about the model and watch sample clips, but you could not generate.
  • → ChatGPT Plus, $20/mo: Included Sora 2. Generations capped per month, output up to 720p, clips up to ~10 seconds, standard queue.
  • → ChatGPT Pro, from $100/mo: Higher monthly cap (roughly an order of magnitude more), 1080p, clips up to ~20 seconds, priority queue, faster generations, full storyboard access.
  • → ChatGPT Team and Enterprise: Sora access depended on the workspace administrator. Most Team plans inherited the Plus feature set; Enterprise customers checked with their account owner.
  • → Sora API: Available to developers through the OpenAI API with usage-based pricing until OpenAI shut it down on September 24, 2026. Access controls and rate limits applied per organization.

There was no separate "Sora-only" subscription. Whatever plan you used for ChatGPT determined how much you could generate. Between Plus and Pro, the rule of thumb was simple: if you generated video more than a few times a week, or you needed clips longer than 10 seconds, the Pro tier paid for itself in saved time. For a single occasional generation, Plus was fine.

Geographic availability mattered too. Sora was restricted in the United Kingdom, parts of the European Union, and a handful of other regions for regulatory reasons, even with a valid paid plan. If sora.com refused your sign-in despite an active subscription, your country was the most likely cause. A static IP from a supported region (where you legitimately resided) was required, and VPN-based workarounds were against OpenAI's terms.

How to use Sora AI for the first time

Knowing how to use Sora AI on day one came down to five clean steps. This was the same path whether someone searched "sora ai how to use", "sora how to use", or "sora how to use sora app".

  1. Confirm your plan. You visited chatgpt.com, clicked your profile, and verified Plus or Pro was active. Free accounts had to upgrade first, because Sora did not unlock without a paid plan.
  2. Open sora.com. You went to sora.com and clicked "Sign in," using the same email tied to your ChatGPT subscription, and landed on the Sora dashboard with a feed of community clips.
  3. Click "New video." The button sat at the top of the dashboard. It opened the prompt composer with controls for aspect ratio, length, and resolution.
  4. Write your first prompt. You kept it under 200 words and described one shot: subject, action, environment, lighting, camera. Example: "A red fox trotting across a snowy clearing at dawn, low cinematic camera, soft golden light, 24 fps."
  5. Hit "Generate." Generation took 1-4 minutes. Sora returned the clip to your library, and from there you could download it, share it, remix it, or extend it.

That was it. The first generation was the slowest because you were learning what the model rewarded. After ten or fifteen runs, your prompts got tighter and your output quality climbed sharply. Most professional users kept a notebook of "prompts that worked" so they could riff on proven formulas instead of starting cold each time.

How to use the Sora app, sign-up and interface

Many people search for "how to use sora app" or "sora app how to use" and assume there was a separate, hidden product. There was not. The "Sora app" was the same web app at sora.com, plus a native iOS app in supported countries. Both shared one account, one library, and one billing relationship with your ChatGPT subscription.

To sign up, you installed the Sora app from the App Store (where available) or opened sora.com in any modern browser, tapped "Continue with OpenAI", completed the sign-in, and accepted the content policy. The first time you logged in, the app pulled in your existing ChatGPT identity, so you did not need a separate account.

The interface had four primary surfaces. The feed showed community clips and what your followed accounts had shared. The library listed every generation you owned, with filters for status, aspect ratio, and prompt text. The composer was where you wrote prompts, attached reference images, set duration and ratio, and hit generate. The storyboard editor was where you sequenced multiple prompts into a single multi-shot clip. Most beginners spent 90% of their time in the composer.

The iOS app worked best as a capture and share device. It was excellent for queuing a generation while you were away from your desk, then reviewing the result on a larger screen later. Heavy prompt iteration was faster on desktop because you could paste, edit, and reference notes in parallel windows.

How to use Sora in ChatGPT

The "how to use Sora in ChatGPT" workflow has changed over time. At launch, video generation lived inside chatgpt.com itself. Later, video work moved to sora.com, with ChatGPT used for prompt drafting. Both paths closed on April 26, 2026. Here is how each one worked.

Direct generation in ChatGPT. In supported regions, paid ChatGPT users could ask the assistant to generate Sora video directly inside a chat thread. The chat handed off to Sora, ran the generation, and returned the clip inline. This was convenient but lacked the storyboard editor, fine-grained controls, and library you got on sora.com.

ChatGPT as a prompt writer. A common workflow was to use ChatGPT (or Claude) to brainstorm and rewrite Sora prompts, then paste the finished prompt into sora.com. ChatGPT is excellent at converting a rough idea, "show me a moody noir detective scene", into a tight, structured prompt with subject, action, lighting, camera, and atmosphere already baked in.

ChatGPT for editing notes. After a generation landed, you dropped the prompt and a description of what looked wrong back into ChatGPT and asked for a revised prompt. This was faster than rewriting from scratch and tended to produce better second drafts than re-prompting Sora cold.

The two surfaces shared an account but not a context. ChatGPT did not see your Sora library, and Sora did not see your chat history. Copy-paste was the connecting tissue.

OpenAI Sora text-to-video: a prompt structure that works

People searching "openai sora text to video how to use", "openai sora video generation how to use", "sora ai video generator how to use", and "openai sora video generator how to use" are all really asking the same thing: what should I actually type? After hundreds of generations, the structure that delivered the most consistent results was the SACLAM formula, Subject, Action, Camera, Lighting, Atmosphere, Motion.

  • → Subject: who or what is on screen, in concrete terms ("a young woman in a yellow raincoat", not "a person").
  • → Action: what they are doing, with one clear verb ("walking", "looking up", "stirring a pot").
  • → Camera: shot type and movement ("medium close-up, slow dolly forward", "wide aerial, locked off").
  • → Lighting: source, direction, and mood ("warm golden-hour sun from camera left", "harsh fluorescent overheads").
  • → Atmosphere: environment and weather ("light rain, glistening pavement, neon signs reflecting in puddles").
  • → Motion: tempo and continuity ("subject moves slowly, camera tracks at the same pace, no cuts").

An example that follows the formula: "A young woman in a yellow raincoat (subject) walking down a narrow alley at night (action), medium tracking shot from behind, slow dolly forward (camera), neon shop signs casting cyan and magenta highlights on her shoulders (lighting), light rain, wet asphalt, faint steam from a manhole (atmosphere), measured pace, 24 fps, no cuts, 8 seconds (motion)."

That single prompt was much more likely to return a usable clip than "girl walking in the rain". Sora rewarded specificity and punished ambiguity. The model had to make a thousand micro-decisions; every concrete noun and verb you provided was one decision it did not have to guess.

Two more rules of thumb. First, you named the style if you had one, "shot on 35mm film, anamorphic lens, slight grain" was a different result than "rendered in pixar 3D animation style". Second, you kept the description to one shot. If you wanted multiple shots, you used the storyboard editor (covered below) instead of asking for cuts inside one prompt.

For prompt-engineering basics that translate to other AI video tools as well, our AI image-to-video tools guide walks through the same principles for image-anchored generations.

How to use Sora 2 on PC

"How to use Sora 2 on PC" is one of the most common search queries because there was no installable Windows or macOS desktop app. Sora was a web product, which meant it worked identically on Windows 10 and 11, macOS, ChromeOS, and Linux as long as you had a modern browser.

  1. Open Chrome, Edge, Firefox, Brave, or Safari. Sora supported all evergreen browsers. Outdated versions of Internet Explorer or legacy Edge did not work.
  2. Visit sora.com. You signed in with your ChatGPT Plus or Pro account.
  3. Optional: install as a PWA. In Chrome and Edge, the install icon in the address bar added Sora to your taskbar or dock. It ran in its own window with no browser chrome, close to a native app feel.
  4. Use a wired connection if possible. Generations were server-side, so your only network requirement was a stable connection during upload of reference images and download of finished clips.
  5. Plan for storage. 1080p clips at 20 seconds were 30-60 MB each. If you generated dozens per day, your downloads folder needed a drive with room to spare.

The PC experience was the most productive way to work with Sora because you could keep ChatGPT, your prompt notes, and the Sora composer side by side. Many editors ran a three-window layout: Sora on the left for generation, ChatGPT in the middle for prompt iteration, and a notes app on the right for prompt history.

Aspect ratios, length limits, and resolution

Sora 2 supported three primary aspect ratios. Each one was purpose-built for a distribution channel, so you picked the ratio before writing the prompt, because Sora composed the frame differently for each.

  • → 16:9 landscape. The default. Best for YouTube, embedded video, websites, and any presentation context. Sora framed subjects centrally with environment around them.
  • → 9:16 portrait. Vertical for TikTok, Instagram Reels, YouTube Shorts. Sora favored close-ups and subject-driven framing because there was no horizontal space to work with.
  • → 1:1 square. Best for static social posts and ads inside Instagram and LinkedIn feeds. Sora tightened framing further than 16:9 but kept more environment than 9:16.

On length, Plus accounts capped at roughly 10 seconds and Pro accounts at roughly 20 seconds per generation. Resolution capped at 720p on Plus and 1080p on Pro. Both tiers limited the total number of generations per month, Plus ran out fastest because each clip counted the same regardless of length, so 10-second clips were more efficient on Plus than 4-second clips. The usage indicator in the dashboard showed what was left before a long batch.

For a 4K master, you rendered at 1080p in Sora and upscaled externally with a tool like Runway's upscaler or a dedicated AI upscaler. Sora itself did not generate at 4K.

Remix and extend a video

Two of the most useful features in Sora 2 were remix and extend. Remix took an existing clip, yours or another user's, where allowed, and regenerated it with modifications. Extend took a clip and added more seconds to the end, continuing the motion and scene logically.

Remix workflow. You opened any clip in your library, clicked "Remix", and edited the prompt. You could rewrite portions, change the subject, swap the lighting, or pivot the camera. Sora kept the overall composition and re-generated with the modified prompt. This was the fastest way to iterate when one clip was 80% right and you only needed to fix one element.

Extend workflow. You clicked "Extend" on a finished clip and supplied a continuation prompt. Sora analyzed the last frame and generated a new segment that continued from there. The result felt seamless when you described the continuation in motion terms ("she keeps walking, then turns left at the next corner") rather than scene-cut terms ("then she enters a different room"). Storyboard handled hard cuts.

Extends consumed a fresh generation credit, so they cost the same as a new clip. Remixes also consumed one credit, which mattered when your monthly cap was tight.

Storyboard mode, multi-shot generations

Storyboard mode was the answer when one prompt was not enough. It let you sequence multiple prompts into a single timeline, with each prompt rendering its own shot, then stitched the shots together into one continuous video. This was how a 60-second cinematic sequence was built on Sora.

  1. Open the storyboard editor from the dashboard. It showed a horizontal timeline divided into shot slots.
  2. Add shot 1. You wrote a prompt for the first shot using the SACLAM formula and set its length (1-10 seconds on Plus, up to 20 seconds on Pro).
  3. Add shot 2 and beyond. You repeated this for as many shots as you needed. Most narrative sequences landed between three and eight shots.
  4. Set transitions. Each shot transition could be a hard cut or a cross-fade. Cuts felt cinematic; fades were softer and matched a mood-piece tone.
  5. Render the full sequence. Each shot consumed a generation credit, plus a small overhead for the assembly. Pro accounts could render storyboards up to several minutes long; Plus accounts capped shorter.

The storyboard was also where you fixed continuity errors after the fact. If shot 3 looked great but shot 4 changed the protagonist's outfit, you could re-render only shot 4 instead of redoing the whole sequence. Each shot was an independent generation that happened to share the timeline.

Costs and credits

Sora's billing model was a fixed monthly subscription, not pay-as-you-go. Both Plus and Pro included a monthly allowance of generations; once you exhausted it, you waited for the cycle to reset. There were no overage charges.

  • → ChatGPT Plus, $20/mo. Included a capped monthly Sora budget at 720p, clips up to 10 seconds, and standard queue priority.
  • → ChatGPT Pro, from $100/mo. Roughly an order of magnitude more generations per month, 1080p output, clips up to 20 seconds, faster queue, full storyboard.
  • → Sora API. Pay-per-generation pricing for developers until OpenAI shut the API down on September 24, 2026. Useful for automation but priced higher per clip than the bundled subscriptions.

One credit equaled one finished clip, regardless of length within the tier limits. A 10-second Plus clip and a 4-second Plus clip both cost one credit. This was why most Plus users defaulted to maximum length, the per-second cost dropped the longer the clip was. Pro users had more headroom but the same logic applied.

The exact monthly cap and credit accounting moved over time, as OpenAI revised both periodically, and the dashboard showed your current usage and remaining credits. Our full Sora review tracks the current numbers, and our Runway vs Sora comparison stacks Sora's value against the main alternative.

Common errors and how to fix them

Most Sora errors fell into one of five buckets. Here is how each was fixed.

  • → "This prompt cannot be processed." Content policy block. Common triggers: named real people, copyrighted characters, weapons in active use, explicit content, certain political imagery. The fix was to rewrite the prompt to remove the named entity and describe a generic style or archetype instead.
  • → "Generation failed, please try again." Server-side issue, usually transient. The fix was to wait 60 seconds and retry the same prompt, and if it failed three times in a row, to simplify the prompt or try a different aspect ratio.
  • → "You have reached your monthly limit." You had exhausted your credit allowance. The options were to wait for the cycle reset (visible in the dashboard) or upgrade Plus to Pro for the rest of the month.
  • → "This region is not supported." Geographic restriction. Sora was unavailable in the UK and parts of the EU even with paid plans, and there was no legitimate workaround.
  • → Output looks wrong (warped faces, bad hands, unnatural physics). Not an error per se. Sora 2 was much better than Sora 1 here, but small objects and rapid limb motion were still failure modes. The fix was a wider shot, slower action, or different camera angle, and avoiding prompts that asked for fingers gripping small props or fast martial arts.

If a generation looked 80% right, remixing it beat regenerating. Remix gave Sora the framing it had already produced as a starting point, which usually fixed the small thing without breaking the overall composition.

FAQ

Is Sora free?

No, and as of April 26, 2026 Sora is not available at all as a consumer product: OpenAI discontinued the app and sora.com. Before the shutdown it required ChatGPT Plus or Pro. The last way to generate with Sora was the paid Sora 2 API, which OpenAI shut down on September 24, 2026; it had cost roughly $0.10 per second of 720p video. There is no free tier and no consumer plan.

Do I need ChatGPT Plus to use Sora?

Not anymore. ChatGPT Plus and Pro no longer include Sora access: OpenAI discontinued the Sora consumer product on April 26, 2026, and no ChatGPT plan grants video generation with Sora today. OpenAI shut down the Sora 2 API on September 24, 2026. If you want text-to-video inside a subscription you already pay for, look at current alternatives like Kling or Runway instead.

What is Sora 2?

Sora 2 was the second generation of OpenAI's text-to-video model. Compared with the first Sora release, Sora 2 produced sharper images, more accurate physics, more natural human motion, better prompt adherence, and synchronized audio in supported regions. It also added the dedicated sora.com app, the storyboard editor for multi-shot scenes, and the remix feature for iterating on existing clips.

How long can a Sora video be?

Before the shutdown, Sora videos on ChatGPT Plus could run up to about 10 seconds at 720p. On ChatGPT Pro, you could generate clips up to about 20 seconds at 1080p. Longer clips also consumed more of your monthly generation budget. For a longer scene, you generated a base clip then used the extend feature to add more seconds, or built a multi-shot sequence with the storyboard editor.

Can I generate Sora videos on my phone?

No. Sora is no longer available on any device: OpenAI shut down sora.com and the Sora mobile apps on April 26, 2026, and shut down the Sora 2 API on September 24, 2026. Before the shutdown, the web app ran in mobile browsers and the app shared one OpenAI account and library with ChatGPT. For text-to-video today, use an actively supported tool such as Kling or Runway.

Why does Sora reject my prompt?

Sora rejected prompts that violated OpenAI's content policy. Common triggers included real public figures or celebrities by name, copyrighted characters and franchises, explicit or sexual content, graphic violence, hate speech, weapons in active use, and certain political imagery. The fix was to rephrase the prompt to remove the named entity, describe the visual style instead of a brand, and avoid charged language.

How do I get sora.com access?

You cannot: sora.com was shut down on April 26, 2026 along with the Sora mobile app, and signing in is no longer possible on any plan. OpenAI shut down the Sora 2 API, offered to developers through the OpenAI platform, on September 24, 2026. For a consumer-style experience, use a current tool such as Kling, Runway, or Hailuo.

Sora vs Runway, which should I use?

For anyone choosing today, use Runway: Sora was discontinued as a consumer product on April 26, 2026, so it is no longer an option at all (OpenAI shut down the API on September 24, 2026). Historically, Sora 2 produced the most photorealistic clip from a single text prompt with minimal control overhead. Use Runway when you need fine-grained controls, motion brushes, camera keyframes, image references, and longer-form editing inside a real timeline, with Gen-4.5 for text to video and Gen-4 or Gen-4 Turbo for image to video. Sora is no longer bundled with ChatGPT, so there is no subscription path to it anymore. Runway is a better fit for production teams and editors who need predictable, controllable output.

Where to go next

Sora 2 set the bar for photorealistic consumer text-to-video, but it is discontinued, so there is nothing left to sign up for and no Plus or Pro tier that includes it. If you need precise creative controls, motion brushes, camera keyframes, image-driven shots, multi-track timelines, start with Runway, and evaluate Pika, Luma AI, and Kling for the rest of your workflow.

For a deeper feature breakdown, read the full Sora review. For a head-to-head against the main alternative, the Runway vs Sora comparison covers pricing, output quality, and use-case fit. For the broader market, the best AI video generators of 2026 ranks every credible model with verified pricing and feature lists.

Full Sora review → Runway vs Sora Best AI video generators 2026 AI image-to-video tools More guides

External references

OpenAI Sora product page sora.com, sign in OpenAI Help Center ChatGPT pricing

See something outdated? Report an issue