Echo-3 and Genie 4 were both announced on 27 October 2026. Both are called world models. They do not take the same input, and they do not return the same kind of file. Neither one is a face-swap tool. Neither one is for sale on HiVideoStudio.
The short pages are Echo-3 and Genie 4.
Echo-3 rebuilds a place you filmed
The PlayCanvas post that afternoon says a clip was generated by Echo-3 and rendered by the open-source PlayCanvas Engine. It had 121 likes when checked on 2026-10-29. The post quotes David Novotny, who wrote that a quick phone video of a room became a 3DGS model, a 3D Gaussian splat, and that the house was then restyled. That post had 2,439 likes on the same check, and it links SpAItial's announcement.
SpAItial's blog the same day calls Echo-3 a physically grounded world model. A few minutes of iPhone capture become a 3D environment at real-world scale. Surfaces the camera missed are filled in to match what was seen. The post offers an iOS app and an enterprise API, and it says Android is not ready. It does not publish weights.
"Open source" in the PlayCanvas sentence refers to the engine. The PlayCanvas Engine repository is public and states an MIT license, including support for Gaussian splats. That license is not a license for Echo-3.
A note in a keyword list called Echo-3 an open-source omni world model in the same class as Genie 4. The posts checked for this guide do not show a weight release or a prompt-steered video generator under the Echo-3 name. Echo-WM, arXiv:2608.23189 from August 2026, is a different model and is not this announcement.
Genie 4 simulates a place and lets you steer it
Google DeepMind's post the same day, by Raia Hadsell, Diego Rivas, and Hang Qi, describes Genie 4 as a simulation that continues a scene: how the world answers, how it sounds, how people move and speak. Keyboard controls navigate. A text prompt can change weather, camera, or who is in the frame. Audio, including ambient sound, music, and dialogue, is generated with the picture. The preview started with trusted testers. HiVideoStudio is not affiliated with Google or DeepMind.
TechCrunch reported a consistency window of up to two minutes, double the Genie 3 limit in that article. DeepMind's own limitation section says detail and object permanence can drift on longer runs, and that audio accuracy, readable text, and spatial accuracy are unfinished. This guide does not pick a winner between those two sentences. They are both in the sources, and they answer different questions: what a briefing said about a short session, and what the research post still lists as open.
Genie 4's post does not say "record your room and measure it." Echo-3's posts do not say "type a prompt and walk a generated world with generated dialogue."
Side by side
- Input. Echo-3, in these posts, wants a capture of a real space. Genie 4, in the research post, wants a prompt and then keys or further prompts.
- Output. Echo-3's cited demo is a splat of that space, viewable in PlayCanvas, with a restyle of the same space. Genie 4's cited output is a running simulation with its own audio.
- Control. Echo-3's posts talk about looking around a reconstruction. Genie 4's post talks about changing the simulation while it runs.
- Sound. Genie 4's post says the model generates the soundtrack. The Echo-3 posts checked here do not describe generated dialogue.
- Access. Echo-3's blog points at an iOS app and an API for companies. Genie 4's post points at trusted testers, then Project Genie and Gemini API partners. This site is not either door.
- Face swap. Neither announcement is a head swap or a character swap. That job, in the same week's open-source posts, is Henge, which is also not for sale here.
What a photo effect is, next to both
A HiVideoStudio effect does not rebuild a room and does not steer a simulation. Survival Horror AI is one photo, 15 seconds, 50 credits. AI Zombie is one photo, 15 seconds, 45 credits. Ozil Goal AI puts one photo into a fixed goal move, 50 credits. Thriller Dance is 50 credits and silent. The lane those pages sit in is written up in world models and AI video effects.
How to choose between the two names
- If the source is a phone video of a room you can walk back into as a splat, read Echo-3. Do not expect a face swap.
- If the source is a prompt and the result is a world that changes when you type again, read Genie 4. Do not expect a scan of your apartment.
- If the source is one photo and a 15-second template, ignore both names and open the effect that matches the scene.
- If the source is a guide video plus a replacement person, read the Henge page. That is a third tool.
Quick answers
Which one reconstructs a real room? Echo-3, in the 27 October 2026 posts: a phone capture becomes a metric-scale 3D environment, shown in one demo as a Gaussian splat.
Which one generates audio and takes steering prompts? Genie 4, in DeepMind's post the same day. Echo-3's posts do not describe that control loop.
Are either of them face-swap tools? No. Echo-3 rebuilds a captured room, and Genie 4 simulates a world. Neither announcement pastes a face onto a performance.
Can I render either one here? No. Both model pages are introductions. They are not for sale.
Sources
PlayCanvas and David Novotny, 27 October 2026. SpAItial, Echo-3. PlayCanvas Engine. DeepMind, Genie 4. TechCrunch, 27 October 2026. Like counts checked 2026-10-29. Last checked 2026-10-29.