01
The same character in every clip
Give the character sheet or product shots as reference images. Each clip starts from the same look, so fewer clips fail review for drift.
Blog · October 14, 2026
FastH3-Omni launch: MiniMax H3 omni reference with audio and first and last frames, accelerated
MiniMax H3 omni reference conditions a clip on images and audio as well as text, but base H3 takes minutes per clip. FastH3-Omni is the accelerated version. FastVideo post-trained it in collaboration with Nuva Lab and Reactor, the two exclusive preview launch partners. The preview is live now, and the open weights ship on October 20, 2026.
What changed
From FastH3 to FastH3-Omni
MiniMax H3 makes video and audio in one pass. FastH3 V1 and V2 made its text-to-video mode fast: 8 model calls instead of 49 (speed numbers). FastH3-Omni does the same for the reference mode.
| FastH3 V1 / V2 | FastH3-Omni | |
|---|---|---|
| Text to video with audio | Yes | Yes |
| Reference images | No | Up to 3 (realtime), up to 9 |
| Audio references | No | Up to 3 |
| First and last frame | No | Yes |
| Clip continuation | No | Yes |
Inputs
MiniMax H3 omni reference inputs in FastH3-Omni
| Input or output | FastH3-Omni |
|---|---|
| Text prompt | Yes, with strong dialogue |
| Reference images | Up to 3 at realtime speed, up to 9 when realtime speed is not needed |
| Audio references | Up to 3, for voices and soundtracks |
| First and last frame | Yes, as keyframe images. A keyframe needs at least one normal reference image in the same request |
| Clip continuation | Latent mode carries motion, camera and sound into the next clip. Frame mode starts the next clip from the last frame |
| Output | Video with synchronized audio, about 5–15 s, 24 fps |
| Aspect ratios | 16:9 (1344×768), 1:1 (768×768), 9:16 (768×1344), 4:3 (1024×768) |
Source: Reactor FastH3-Omni documentation and Reactor launch post. Read October 14, 2026. The open release on October 20 can change these limits.
Use cases
What references let a production team do
02
A real voice and a real track
Give a voice line or a music clip as an audio reference. The video and the sound come from one generation, so you do not sync them later.
03
Shots that cut together
Pin the first or last frame to a storyboard frame or to the end of the previous shot. Ad versions and episodes then join at known frames.
04
Longer scenes
Continue a clip in latent mode to keep motion, camera and sound across a scene that is longer than 15 seconds.
Access
Two exclusive preview partners, three ways to run it
FastH3-Omni has two exclusive preview launch partners: Nuva Lab and Reactor. Reactor runs the hosted realtime API. Nuva Lab is the only partner that offers dedicated deployment, fine-tuning on your data and the Creative Agent end to end.
| Nuva Lab preview | Reactor API preview | Open weights | |
|---|---|---|---|
| Available | Now | Now, until October 20, 2026 | October 20, 2026 |
| Where it runs | A dedicated host that serves only your team | Reactor's hosted realtime API | Your own GPUs |
| Fine-tuning on your data | Yes: we fine-tune and post-train on your references and review criteria | No | You train it with the released code |
| Creative Agent, end to end | Yes: briefs, batches, retries, checks and review | No | No |
| Your data | Stays in your deployment | Sent to the API per request | Stays on your hardware |
The standard Nuva Lab host has eight RTX PRO 6000 GPUs. B200, B300 and GB200 hosts are available on request.
Speed and local
FastH3-Omni speed and local GPU support
Reactor describes FastH3-Omni as a realtime H3. No measured time per clip is public yet. We will add measured numbers after the October 20 release. For the earlier text-to-video FastH3 V2, Hao AI Lab measured 36.5 s for a 5-second 1344×768 clip on one RTX PRO 6000.
FastVideo says the final FastH3-Omni weights ship on October 20, 2026, with training code and support for DGX Spark and RTX GPUs. For VRAM and system RAM today, read MiniMax H3 VRAM and GPU requirements.
FAQ
Questions
What is FastH3-Omni?
FastH3-Omni is a post-trained, accelerated version of MiniMax H3 with omni reference input. It takes text, up to three reference images, up to three audio references, first and last frame keyframes and clip continuation, and it outputs video with synchronized audio. FastVideo post-trained it in collaboration with Nuva Lab and Reactor.
How is FastH3-Omni different from FastH3?
FastH3 V1 and V2 cover text to video with audio. FastH3-Omni adds reference images, audio references, first and last frame keyframes and clip continuation.
When will FastH3-Omni be open?
FastVideo says the final FastH3-Omni weights ship on October 20, 2026, with training code and support for DGX Spark and RTX GPUs.
Can FastH3-Omni take a video as a reference?
The Reactor documentation lists text, image and audio references, keyframes and clip continuation. It does not list video references.
How fast is FastH3-Omni?
Reactor describes FastH3-Omni as a realtime H3. No measured time per clip is public yet. We will add measured numbers after the October 20 release.
Sources and further reading
Read next
Run FastH3-Omni on your own host
Send us your references and clip formats. We'll set up the preview on a dedicated host and test it on your briefs.
Get started · 1 min