Give it a frame you already have and say what moves. Or set a first and last frame and let it fill the gap. No image yet? Describe the shot and start there.
Create
Video generation looks complicated from outside and isn't. Most of the decisions are ones you change after you see the first result.
Upload the image the clip should open on. A storyboard panel, a product shot, a still you generated a minute ago. Add a last frame too, and the clip moves from one to the other. Starting from an image keeps your art direction. Starting from a prompt alone gives you more range, and less control.
Try it in the generator →Camera, subject, or both. "Slow push in", "handheld", "static, subject walks left". The more specific the motion, the less you fight the result.
Try it in the generator →Most clips are either about 5 seconds or about 10. Short renders faster and costs less, which matters while you're still deciding. Set the aspect ratio to match where it's going, and it's more often vertical than widescreen.
Try it in the generator →Change one thing at a time. Most shots land in a handful of attempts, and the ones that take longer are usually exploring range on purpose.
Try it in the generator →Models
No single video model is best at everything. One handles motion well, another holds a face across the clip, another gets you a usable result fast while you're still figuring out the edit. Having them in one place means you switch without moving files between products.
There's no default here. Every clip on Magnific is generated with a model someone chose, and that choice is the biggest lever on the page. Name what you want different, then pick.
| When you need | Reach for | Why |
|---|---|---|
| Sound in the same take | Dialogue, effects and music generated with the picture, in sync. | |
| Several shots in one clip | Connected shots in a single generation, up to 15 seconds on the longest models. | |
| Motion from a reference | Transfers body, face and hand movement from a video onto your character. | |
| Edit footage you already have | Works from image and video references to change a clip and keep its characters consistent. | |
| Fast drafts | Quicker, lower-cost renders to test the motion before the final take. | |
| A talking character | Lip sync, gestures and body movement from one image and an audio clip. |
About a third of all video seconds on Magnific go through one model, and the one most people try is a different one. The rest is spread across more than 150 model and setting combinations.
Showcase
Shorts and internal pieces from our team, made with the same models and tools on this page. Pick one and press play.
More pieces
Workflow
Generating the clip is the part everyone shows. What decides whether it's usable is everything around it.
Before
A storyboard frame, a product shot, a still you generated an hour ago. Most clips on Magnific start from an image someone brought. Video that starts from your own image inherits your art direction instead of reinventing it every time.
After
Upscale it. Extend it past its last frame. Add a voice, score it, cut the background. When a clip gets a second pass here, it's to change a detail, make it bigger or make it longer, and none of the three means exporting.
Around
Through the interface, through the API, or from an assistant with a connection to Magnific. The last one is now bigger than the API for video, and it needs no integration project. Ask for the clip from where you already are.
Trust
The questions that matter before a client project, or before a channel, answered plainly.
What you generate is yours, including for commercial work. No separate license, no per-use fee.
We don't train models on what you generate here.
Content you start from comes from a licensed library rather than scraped material.
Nothing about how a clip is made here restricts where you publish it or what you earn from it. What a platform does with AI-made content is that platform's policy, and the rules change. Ours don't: it's your clip.
Numbers
79%
31.6M video generations on Magnific over one summer, about 61,750 hours of video. The public story of AI video is a sentence turning into a scene. That's not how it's used. It's an image, a note on what moves, and a decision about 5 seconds or 10.
01
1 in 6
Most clips start from a single frame someone brought. 1 in 6 set a first and a last frame. Text alone is about a fifth, and those clips run longest, at 8.1s on average.
What it means
Video on Magnific is a way to move an image you already have, not a way to conjure one. First and last frame at 1 in 6 is a storyboard workflow: two panels, and the model draws the motion between them. Text alone is where people go for range, and they go longer when they do.
What to do with it
Bring the frame. If you have a still you like, the clip will keep what you like about it. Use a prompt alone when you're still looking for the shot, not when you've found it.
02
44%
44% of clips are 5 to 6 seconds, and 27.9% are 8 to 11. By seconds the long band wins, with 38.3% of all video seconds. The 7-second average describes almost nobody.
What it means
People don't choose a duration. They choose short or long. Short is the sketch: cheaper, faster, and the place to find out if the motion works. Long is the shot. Both lengths are presets, and the presets are the product.
What to do with it
Draft at 5 seconds. Once the motion lands, generate the 10. The extender is there for when 10 turns out to be 15.
03
44.6%
9:16 is the most generated format, ahead of 16:9 at 37.9%. About 97% of clips run at 24fps, and native-audio models account for 7.4% of clips.
What it means
More clips leave vertical than widescreen. Frame rate isn't a decision anyone makes. Sound mostly isn't either: 7 in 10 clips are generated on days with no audio activity at all. Video here is silent by default, and picture and sound are two different jobs.
What to do with it
Set the ratio for the destination before you generate, not in the edit. If the clip needs sound, choose a native-audio model up front or add a voice and a score after. Either works. Deciding late doesn't.
Sound figures are a bracket, not a point. The generator doesn't record audio use on the clip itself; the read comes from the model chosen and same-day use of audio tools.
04
7×
Agents and assistants connected to Magnific bring in 7 times the video of the API. The interface still makes 97.3% of all video; among external agents, Claude leads.
What it means
Agents outrun the API 7 to 1 for video, against about 4 to 1 for images. Almost nobody built an integration for this. They opened a chat that already had Magnific in it and asked for the clip. API callers, by contrast, sit on the short preset: 5.0 seconds per clip against 7.0 overall, which is what a pipeline looks like.
What to do with it
If your work happens in a chat, connect it and ask for the clip there. Build on the API when it's the same clip a thousand times.
Magnific's own agent excluded, the rest renormalized to 100.
05
32.2%
Among individual accounts, one model carries 32.2% of all video seconds. Among Business accounts it's third, at 10.0%, and 34.5% of their video seconds are generated on canvas.
What it means
Individuals converge on one workhorse. Teams spread across models and go to higher resolution, because the clip is going into someone else's timeline and has to survive it. Same tool, different job: exploration on one side, delivery on the other.
What to do with it
If the clip is leaving Magnific for an edit, generate at the resolution the edit needs instead of upscaling later. If several people touch it before it does, put it on a canvas so they can.
06
1 in 3
About a third of video users take a clip further after generating it. Of those second passes, 42% change something in the clip, 37% make it bigger and 11% make it longer. Together they're about 3% of all video generations.
What it means
Three reasons to come back: something in the shot is wrong, the clip needs to be bigger, or it needs to be longer. Everything else is marginal. Most clips are used as they render; the ones that aren't get one operation, not a pipeline.
What to do with it
Get the motion right in the generation. Modify is for the detail that's off, upscale and extend are for when the take is right and the spec isn't.
Plans
Video costs more to generate than images, and the price moves with 3 things: the model, the length, and the resolution. A 5-second draft and a 10-second 1080p delivery aren't the same spend, and they shouldn't be. Every plan draws from the same credit balance, so nothing is locked behind a tier. You're buying volume, not features.
Yes. What you generate is yours to use in client and commercial work, with no separate license and no per-use fee.
From our side, yes: you own the clip and there's no license that limits where it earns. Whether a channel is monetized is YouTube's decision under YouTube's rules for AI-made content, which change. Nothing about generating here works against you in that decision.
Start from images you have the rights to, generate on a platform that doesn't train on your work and tells you what it was trained on, and keep the ownership terms in writing. All three hold here.
You do.
No.
Either, but most people bring an image. About 4 in 5 clips on Magnific start from a frame the user supplied, and 1 in 6 of all clips from a first and last frame. Starting from an image keeps what's in the frame and your art direction. Starting from a prompt gives you range.
Up to the model's limit, shown in the generator when you pick it. In practice clips are either about 5 seconds or about 10; those two lengths are 7 in 10 of everything generated. Draft short, generate the long one when the motion lands, and extend it if 10 isn't enough.
Only if you ask for it. Some models generate native audio; the generator marks them. Most clips are made silent and get a voice or a score added after, or don't need one. Both paths are on the page.
The one that does the thing you can name. Motion, a face that holds, speed, native sound, a specific resolution. There's no default, so every clip is a choice, and the table is built to make it in one pass. About a third of all video seconds here go through one model, and the model most people try first is another, which tells you where to start if you can't name it yet.
Yes. Upscale it, extend it past its last frame, add a voice, score it, or remove the background, without moving the file anywhere. When people come back to a clip it's to modify a detail, upscale it or extend it, in that order.
Both. There's an API for pipelines. And if you work in an assistant like Claude or ChatGPT, connect it to Magnific and ask for the clip there. For video, that route is now about 7 times the size of the API.
Vertical 9:16, widescreen 16:9, square, and the portrait and cinematic ratios in between. Vertical is the most generated format here, ahead of widescreen. Resolution depends on the model; the generator shows the maximum when you pick it.
Start from an image. Say what moves. Change one thing, generate again. That's the whole loop.
