Sora Is Gone: 5 Alternatives That Still Take Your Face (September 2026)
OpenAI retired the Sora app in April 2026 and the API goes dark on 24 September. Here is where the 'me in an AI video' workflow moves, with per-second prices and the real-face policy of each option.
OpenAI wound Sora down in two stages: the app and web experience were discontinued on 26 April 2026, and the API endpoints stopped serving on 24 September 2026. If you built anything on Sora 2 or Sora 2 Pro, the deadline has passed and the migration is overdue.
The feature people actually miss is not text-to-video — plenty of models do that. It is putting yourself in the shot. Here is where that goes.
The five that still work
| Option | Model | Price | Takes a real face |
|---|---|---|---|
| HappyHorse 1.1 ref2v | HappyHorse 1.1 reference-to-video | $0.14/s | yes, up to 9 refs |
| Kling 3.0 Motion Control | Kling 3.0 Pro Motion Control | $0.168/s | yes, with likeness limits |
| MiniMax H3 Max Recast | MiniMax H3 Max Recast | ~$0.30–0.45/s (reported; fal lists a $0.05 base unit) | not yet tested by us |
| MiniMax H3 ref2v | MiniMax H3 reference-to-video | $0.26/s (the H3 Max line lists $0.05 base) | yes |
| OmniHuman 2.0 | OmniHuman 2.0 | $0.105/s | yes, audio-driven |
Prices for the HappyHorse, Kling, MiniMax and OmniHuman rows were re-read on 4 September and 18 October 2026 (OmniHuman moved to 2.0, the row above). Sora's API is now closed, so there is no waiting period left.
HappyHorse 1.1 reference-to-video is the closest thing to the Sora cameo experience: hand it up to nine photos of yourself plus a prompt, get 1080p in a genuine 9:16 or 4:5 frame with native audio and multilingual lip-sync. It is the only one on this list with the full set of vertical aspect ratios out of the box.
Kling 3.0 Motion Control solves a different problem — it copies the movement of a 3–30 second reference clip onto your character. If what you liked about Sora was describing a scene, this is not it. If what you want is to perform someone else's choreography, nothing else on the list competes.
MiniMax H3 Max Recast (October 2026) replaces people in an existing 5–30 second clip from 1–4 reference photos while keeping the cuts and the audio. It is the most direct answer to "put me in this video", but it is new and we have not run it on our own clips; see our guide.
MiniMax H3, released 31 July 2026 (the Max line followed in August), takes up to nine image references plus reference clips for motion and audio, at 2K. It is new enough that there is no meaningful body of user reports yet — treat it as promising rather than proven.
OmniHuman 2.0 (it replaced 1.5 on fal in October 2026) is the audio-driven option: a photo plus an audio track produces a performance matched to the audio. Not a scene generator.
What is not a replacement
Seedance 2.0 and 2.5 both refuse real human likenesses, and the filter sits upstream of every reseller — so "use Seedance through a different provider" is not a workaround. Veo is not an option here either: Google's Veo endpoints are no longer served through fal. If a tool markets itself as a Sora replacement and runs on either of those, it cannot do the cameo use case.
Seedance 2.5 (announced 31 July 2026, now on fal) does not change that. It takes up to 50 references and produces a genuine 30-second single take, but ByteDance's BytePlus platform confirmed on 31 August 2026 that real-face references are rejected, and the filter sits above the model, so hosting it elsewhere does not change the verdict (details). At roughly $0.462 per output second it is also the most expensive per-second rate in the autumn wave. Of the rest of the autumn releases, MiniMax H3 Max and Wan 3.0 now list base rates on fal (without published resolution multipliers), Gemini Omni Flash 1.1 blocks the likeness of real people and is not offered in the EEA, UK or Switzerland, and LTX 2.5's price unit is still undisclosed; the autumn round-up has the full list. None of them displaces the five above for the cameo job.
Consumer face-swap apps in the $10–15/month bracket do a narrower job than they imply: most are face-only swaps at 720p with a 15-second cap, which looks composited rather than performed.
Migration checklist
- Inventory anything still calling Sora endpoints; the API closed on 24 September.
- Decide which of the two jobs you need — new scene from references, or existing clip with a new performer. They are different models.
- Re-check prices before committing. Every number above is dated, and the endpoints reprice without announcement.
- Export anything stored in a Sora account. OpenAI has stated the data goes with the shutdown.
How to do this without code: The cameo use case, putting yourself in a described scene, is the Me in scene tool. In AutoHustle this is the Me in scene tool: a few photos of you and one sentence describing the scene (about 2 credits per second, quoted before you press Generate). See How to put yourself in any scene and Put yourself in a movie scene.
Related reading: AI Video Generation Pricing Compared (October 2026): What a Second Really Costs · How to Put Your Face in Any Instagram Reel (2026 Walkthrough)
For developers: the fal endpoints behind this article
Prices in the article are per-second or per-unit rates from the fal.ai pricing API on the dates stated; re-check them before you build, because endpoints reprice without notice. The IDs:
| Model | fal endpoint |
|---|---|
| HappyHorse 1.1 reference-to-video | alibaba/happy-horse/v1.1/reference-to-video |
| Kling 3.0 Pro Motion Control | fal-ai/kling-video/v3/pro/motion-control |
| MiniMax H3 Max Recast | minimax/h3-max/recast |
| MiniMax H3 reference-to-video | minimax/h3/reference-to-video |
| OmniHuman 2.0 | bytedance/omnihuman/v2 |