AI Video Models on fal in October 2026: What Is New, What It Costs, and What Is Gone
The month's changes in one page: Recast and Kling O3 for replacing a person, OmniHuman 2.0 and H3 Max Lip Sync for talking photos, Nano Banana 2.1 and GPT Image 2.5 for stills, plus Veo and Sora leaving fal. Prices read on 18 October 2026, with a status for each model in our own pipeline.
Two months after the late-August wave described in every AI video model that shipped this autumn, fal's catalogue has moved again, and the shape of the change is different. August was about generators. October is about tools for specific jobs: replacing a person in an existing clip, giving a recording a speaker, editing a still. This page lists what is new, what it costs on 18 October 2026, and where each model stands with us. "Candidate" means we have read the schema and prices but have not run it on our own material, and nothing is on sale in the Studio until it has.
Replace a person in an existing video
| Model | Price per second | Limits | Our status |
|---|---|---|---|
| MiniMax H3 Max Recast | $0.30 / $0.45 (768P / 1080P, reported) | 5–30 s in, one to four photos, keeps cuts and audio | candidate, test planned |
| Kling O3 edit | $0.14 (4K $0.42) | 3–15 s, up to 4 references | rejected in April, retest planned |
| HappyHorse video-edit | $0.28 delivered (720p) | 3–60 s in, 15 s out | in production |
The change that matters: Recast makes "who becomes whom" an explicit input (several people in the frame). fal's own recommender returns no identity guarantee for any of them. The overview is how to replace a person in a video.
Talking photos and lip sync
| Model | Price per second | Notes | Status |
|---|---|---|---|
| OmniHuman 2.0 | $0.105 | audio 4–300 s; replaces 1.5 ($0.16) | candidate |
| MiniMax H3 Max Lip Sync | $0.05 base | audio up to 15 minutes | candidate |
| Kling AI Avatar v2 | $0.115 Pro / $0.0562 Standard | in use for our onboarding video | active |
| FlashTalk | $0.02 | fixed at 768×448 landscape | not for vertical |
| ElevenLabs Dubbing | $0.01 | voice only, no lip movement | candidate |
Generators
| Model | Price | Notes | Status |
|---|---|---|---|
| MiniMax H3 Max and Turbo | $0.0325 / $0.01625 per second (768P base) | up to 15 s | candidate |
| Wan 3.0 Prime | $0.05 base, about $0.14 at 720p | up to 30 s, video references | candidate |
| FLUX.3 video | $0.085 per second | no face filter | active for face-less scenes |
| Seedance 2.5 | about $0.46 per second at 720p | 30 s, 50 references, no real faces | deprecated by policy |
| LTX 2.5 | $0.01 per undefined unit | open weights | candidate, unquotable |
| Gemini Omni Flash 1.1 | $0.03 base | blocks real people; not offered in the EEA, UK or Switzerland | not usable for faces |
Kling 3.0 is steady: Pro and Standard at $0.14, Turbo Standard at $0.112, native 4K at $0.42 (Pro vs Turbo, 4K).
Motion transfer
Scail 2 lists $0.10 per second against Kling Pro's $0.168 and was preferred in one blind pair of ours, but it outputs 704p at 16 fps; Wan Motion lists $0.06. Kling stays the default (Motion Control).
Images and finishing
| Model | Price | Notes |
|---|---|---|
| Nano Banana 2.1 | about $0.034 for 1K (Google's rate) | replaces Nano Banana 2; up to 14 references |
| GPT Image 2.5 Flare and Sunburst | about $0.05–0.25 | token-billed; masks, transparency |
| Seedream 5.0 | $0.027 / $0.035 / $0.0675 | three tiers; faces untested |
| FLUX.3 image and Klein | $0.048 / $0.009 per megapixel | up to 10 references |
| HY-WU Edit (Tencent) | $0.10 per unit | the one new image face-swap on fal; candidate |
| ByteDance video upscaler | $0.0072 per second | a preset for AI-generated footage; test planned |
What left fal
- Veo. The Veo endpoints show only as retired on fal; one tracker reports a deprecation banner dated 17 November, with no replacement named.
- Sora. OpenAI's API closed on 24 September (alternatives).
- Runway Aleph 2, Act-Two and Hedra Character-3. Searches for them on fal return nothing under any name, so they are available only through their own sites.
- Hailuo 2.3 still exists, but it dates from October 2025 and the H3 line has overtaken it.
What to take from the month
- Check the unit before the price. Base rates, per-megapixel billing, token pricing and "per unit" with no definition are all in this list (the resolution lesson).
- Specific beats general. A model made for replacing a person beats a general generator at it, if it holds a face; that is the open question for Recast and O3.
- Test before you route. Prices are in the schema; identity is not. Every candidate above waits for a paired test on clips we own.
Common questions
What is the biggest change this month?
Models built for specific jobs: Recast and Kling O3 for replacing a person, OmniHuman 2.0 and H3 Max Lip Sync for talking photos.
What left fal?
Veo is retired there, Sora's API closed on 24 September, and Runway and Hedra models are not on fal at all.
Which new models are on sale in the Studio?
None of the candidates. They go on sale only after a test on clips we own.
Why are some prices marked as base rates?
fal lists a base unit for several new models without the resolution multiplier, so the price at 1080P is not published.
Where are the older updates?
The late-August wave is in our autumn round-up, and the summer releases in the summer round-up.
How to do this without code: You do not have to follow each month's releases: new models are tested before they reach the Studio. In AutoHustle this is the autopilot: paste the Reel link, add one photo of yourself and confirm the plan (about 4 credits per second of finished video, so 60 credits for a full 15-second call; analysing the Reel is free). See How to put your face in any Reel and Recreate an Instagram Reel.
Related reading: AI Video Generation Pricing Compared (October 2026): What a Second Really Costs · Higgsfield Pricing 2026 and What You Are Actually Paying For
For developers: key endpoints
| Model | fal endpoint |
|---|---|
| MiniMax H3 Max Recast | minimax/h3-max/recast |
| Kling O3 Pro video-to-video edit | fal-ai/kling-video/o3/pro/video-to-video/edit |
| OmniHuman 2.0 | bytedance/omnihuman/v2 |
| MiniMax H3 Max Lip Sync | minimax/h3-max/lip-sync/image-to-video |
| MiniMax H3 Max image-to-video | minimax/h3-max/image-to-video |
| Wan 3.0 Prime reference-to-video | alibaba/wan-3.0-prime/reference-to-video |
| Nano Banana 2.1 edit | google/nano-banana-2.1/edit |
| GPT Image 2.5 Sunburst edit | openai/gpt-image-2.5/sunburst/edit |
| ElevenLabs Dubbing | fal-ai/elevenlabs/dubbing |
Prices read from fal on 18 October 2026, with third-party figures where fal shows only a base unit.