Sora Is Gone. Here Is Where Your Workflow Goes Next.

Last verified: October 29, 2026 · This is a living page. When we finish testing an alternative ourselves, its “pending” note is replaced with our logged results.

OpenAI has shut Sora down. If you built a workflow on it — social clips, product teasers, storyboards — this page is the honest map of what happened, what to save before it disappears, and where each job actually goes now, based where possible on tools we have run ourselves, on the same prompts, with the costs logged.

1. What happened — the verified facts

From OpenAI’s own Help Center (read by us on 2026-10-19):

• Sora web and app were discontinued on April 26, 2026.
• The Sora API was discontinued on September 24, 2026.
• Exports run through OpenAI’s sunset page at sora.chatgpt.com/sunset; OpenAI states your data is deleted after the final export window.
• OpenAI names no successor product on that page. Anyone selling you “the new Sora” is selling you something else — judge it on its own tests, not on the name.

That is the whole verified record. We do not publish rumors about why, or guesses about what OpenAI does next.

Export checklist — do this before the window closes

Our advice (not OpenAI’s wording), in order:

1. Go to the sunset export page above, signed into the account that owns the Sora creations.
2. Download everything you might ever reuse — finished videos first, then any source or reference material the export offers. Once the final window closes, OpenAI says the data is deleted; there is no recovery step after that.
3. Save your prompts and workflow notes alongside the files. The videos are the asset you will miss least; the prompts that produced your best results are the workflow you actually need to rebuild elsewhere.
4. Keep one untouched copy as your archive, and a second copy in the folder you will feed into your replacement tool for re-creation tests.
5. Note which clips were ever published, and where — if you re-create a published clip in a new tool, you will want to know which version your audience already saw.

2. Where to go, by job — not by hype

Job A: Generative clips (text-to-video / image-to-video) — TESTED BY US

We ran the same verbatim prompts through three alternatives on 2026-10-23, across six briefs: a serialized-drama scare scene, an everyday life scene, an emotional photo-to-life hook, a sci-fi art portrait, a cute-creature clip, and a food-promo scene. Full logs, raw outputs, and receipts are archived with each test. Headline results:

Read those rows the way a working creator should:

• If your Sora workflow was vertical shorts: Kling and Veo work as direct replacements today. Runway works too, but through image-to-video — budget the extra step, because its headline text-to-video mode would not give us a vertical file at all.
• If you publish commercially: Kling’s watermark is a real cost hidden inside the cheapest price. For client or brand work, Veo’s clean export may be worth more than Kling’s lower cost per minute. We are still checking whether Kling has a no-watermark download path on this tier; until we verify it, treat the watermark as present.
• If you hate garbled text in frame: two of three tools invented gibberish signage despite a “no text” instruction. Veo was the only one that stayed clean, and on food shots all three stayed clean only because the signage stayed out of focus — a prompt technique worth copying (see §3).

This comparison frame replaces the “pending” placeholder from our launch outline. Scores follow our published protocol weights (see /how-we-test) and will be finalized in Piece 2, the dedicated shootout article.

Job B: Avatar / talking explainer videos — NOT YET TESTED BY US

The tools most Sora users are pointed to for presenter-style video are HeyGen and Synthesia. Vendor claim — not tested: both market AI-avatar presenters, multilingual voiceover, and template-based explainer production (vendors’ own pages, read 2026-10-19).

Our head-to-head is already designed and frozen: the same 90-second script, verbatim, on both tools, plus a script-change re-render test and EN/ES/ZH language passes (protocol/sprint-b-scripts-v1-frozen.md). Until those runs are logged, we publish no verdict, no score, and no ranking for either tool.

Job C: Editing and repurposing what you already have — NOT YET TESTED BY US

If your Sora problem is not generating new clips but cutting, captioning, and re-cutting existing footage, the candidates are timeline/repurpose editors such as Descript, VEED, and CapCut. Vendor claim — not tested: each markets transcript-based editing and/or automatic short-form cutting (vendors’ own pages, read 2026-10-19).

Our repurpose test is also pre-designed: one fixed ~10-minute source video cut into three captioned vertical shorts, scored on caption accuracy, cut quality, and time to publishable. The source script is frozen (protocol/sprint-b-s3-source-script-fulltext.md); the runs happen in Sprint C, after the avatar winner generates the source. No verdict until then.

3. What does NOT carry over from Sora workflows

• Your prompts transfer as drafts, not as recipes. The same verbatim prompt produced visibly different casting, framing, and even aspect-ratio behavior across the three tools we ran. Re-test your five most important prompts on any candidate before you commit a subscription to it.
• “No text” does not mean no text. Two of three tools invented signage and timestamps anyway. What worked in our food test: keep any signage in the background and out of focus, and say so in the prompt.
• Vertical is a workflow choice, not a default. On Runway, the path you enter through (text-to-video vs. image-to-video) decided whether 9:16 existed at all. Check the export spec of your first test clip before generating a batch.
• Credits are not minutes. Every tool prices in its own credit unit, and retries burn them. Our cost figures above are per usable finished minute, retries included — the only version of the number that survives contact with a real deadline.
• Your Sora exports do not become training data for the next tool by magic. Re-creating a clip means generating it again, with the new tool’s failure modes. Keep the original exports as your quality reference.

4. Price honesty box — what $20 actually buys

Dated snapshot, verified at purchase on 2026-10-23; prices change, and this box is re-verified on the same cycle as the rest of the page.

• ~$7–9 (Kling Standard): the cheapest route we tested to true-vertical generative clips, at $9.54–$12.00 per usable finished minute in our runs — with a watermark on the standard export.
• ~$15 (Runway Standard): $17.28 per usable finished minute in our runs, no watermark, fastest on simple scenes — with the text-to-video vertical lock described above.
• ~$20 (Google AI Pro / Veo via Flow): $15.00 per usable finished minute in our runs, the cleanest output we saw, no watermark — with queue times that varied from ~9 minutes to nearly 30 minutes per clip in our sessions.

The lesson is not “cheapest wins.” It is that the sticker price, the credit price, and the usable-minute price are three different numbers, and only the third one is real.

5. How this page stays honest

• Everything labeled “tested by us” links to a test run under our published protocol (/how-we-test), with credits, retries, wall-clock time, and raw outputs archived.
• Everything else is labeled “vendor claim — not tested,” including tools we plan to test next.
• Some links on PickaSi are affiliate links; they never change a score or a ranking. Full details: /disclosure. As of this page’s first publication, PickaSi had no live affiliate links; our first marked (†) affiliate link went live on October 29, 2026.
• Found an error? Our corrections policy (/corrections) commits us to fixing it on the page, noting what changed and when — a standard we have already applied to our own first-week watermark correction.

Changelog

• v1 — October 29, 2026: First publication. Generative comparison frame filled with logged Sprint A results (T1–T6, 2026-10-23/24); avatar and repurpose sections published as vendor-claim placeholders pending Sprint B/C runs.