Midjourney V1 Video Model · image-to-video · 480p and 720p

Midjourney Video: animate one still into a 21 second clip

Midjourney calls its video generator the V1 Video Model and its only workflow Image-to-Video. This page runs one image-led route on it, quotes the official resolution, duration, and GPU time tables, and writes out the settings the model has that this workspace does not.

Open the video workspace
01

What this page actually runs

Upload one frame, write the movement, choose a duration and a resolution, and read the point estimate before you submit. The official model also accepts an ending frame and returns four variations per prompt; those settings are not on this page, and the difference is listed below rather than implied.

LIVE
5s
2s15s

This model does not support reference video input

Results

Creation history

Generated content will appear here
02 / Overview

Midjourney Video specifications

Facts reviewed:

Developer
Midjourney
Official product name
V1 Video Model
Official workflow name
Image-to-Video
Released
18 June 2025
Input
One image as the first frame, with an optional text prompt
Accepted images
A Midjourney generation or an image uploaded from outside Midjourney
Ending frame
Optional; reusing the first frame produces a loop
Default output resolution
480p, described as standard definition
High definition output
720p, on Standard, Pro, and Mega plans, in Fast mode only
Clip length
5 seconds at the start, 4 seconds per extension, 4 extensions maximum, 21 seconds ceiling
Audio
None; no audio parameter and no soundtrack in the output
Video-only parameters
Six: motion low, motion high, raw, loop, end, and batch size
Videos per prompt
Four by default, selectable as one, two, or four
Public API
None published
Route id on this site
midjourney-video, a platform identifier rather than an official model id

Documented by Midjourney, and offered on this page

This page has an unusual shape: every mode switch matches the official model, and the distance sits inside the settings those switches open. The two lists below separate what Midjourney documents from what the workspace here exposes, field by field.

OFFICIAL

Documented for the V1 Video Model

  • Image-to-video from one supplied frame, with the prompt optional rather than required
  • An optional ending frame, which turns a supplied pair into a controlled move, or into a loop when it repeats the first frame
  • An automatic mode that derives the movement from the image, and a manual mode where the prompt carries it
  • Low motion as the default, and high motion for larger camera and subject movement at a documented risk of artefacts
  • A batch of four videos from one prompt by default, selectable as one, two, or four
  • Standard definition at 480p on every plan, and high definition at 720p on Standard, Pro, and Mega in Fast mode only
  • Five seconds at the start, four more seconds per extension, up to four extensions and a 21 second ceiling
  • Six video-only parameters: motion low, motion high, raw, loop, end, and batch size
  • Playback in the browser on hover, and frame-by-frame scrubbing by holding Control or Command while moving the pointer
  • Three download options: a social-media encoded MP4, the raw MP4, and a GIF
LIVE PLATFORM

Offered on this page today

  • Image-to-video from one uploaded frame, plus a first-and-last-frame route that takes two
  • A 2 to 15 second selector with 5 seconds as the default, against an official start of 5 and a ceiling of 21
  • 720p and 1080p in the resolution selector, against an official default of 480p and a high definition tier of 720p
  • Five fixed aspect ratios with 16:9 as the default, where the official model derives its ratio from the starting image
  • No batch control, so one submission yields one clip rather than the official default of four
  • No extension control and no motion-strength control, and no audio switch, which matches the model because it produces no audio

Most of the distance here is in settings rather than switches, which is the reverse of every other video page on this site. The mode switches and the absent audio control match the official model exactly: no text-to-video, no reference-to-video, no soundtrack. Everything that differs sits inside the settings. Duration moves in both directions, with a 2 second floor below the official 5 and a 15 second ceiling below the official 21. Resolution is a swap rather than a gap, because 1080p is offered here and is not an official output, while the official 480p default is absent. Aspect ratio is the one control the official model does not have at all, since the starting frame decides the ratio and the help centre only shows what that implies for pixel dimensions. The batch of four and the four extensions are the two capabilities most often assumed to be present, and neither has an interface on this page.

Model brief

What the V1 Video Model is, and what it is not

Midjourney Video is not a version of the image model. The company announced its video generator on 18 June 2025 under the name V1 Video Model, describing the release as Version 1 of its Video Model at a point when the image line carried its own separate version numbers. The official help centre keeps that separation intact: the article covering video is titled Video, and it states plainly that every image in a gallery can be animated regardless of which version it was made with. That sentence rules out the most common error in third-party coverage, which is to attach a current image version number to the video generator. There is no video mode inside a numbered image version. Video is its own product with its own number.

The workflow name is just as literal, and it does the work of a specification. Midjourney calls it Image-to-Video, and the name describes the machine: an image goes in, a clip comes out, and the text prompt is decoration layered on top of the frame rather than the thing that decides the picture. There is no text-to-video route and no reference-to-video route, and neither is missing by oversight, because the entire generation is built around feeding the model a first frame. Two dials sit over that frame. The first is the animation mode, where automatic lets the system read a movement out of the image and manual asks for the movement to be written. The second is motion strength, where low is the default and high buys larger movement at a higher chance of a result that looks wrong somewhere in the frame.

Three absences are worth naming because they are routinely assumed to be present. There is no audio in either direction: the six video-only parameters contain nothing audio-related, and no output carries a soundtrack. There is no public API, which matters because a model page on a platform like this one normally implies a service call behind it. And there is no way to describe a shot in words and skip the image, because the first frame is the substrate rather than a reference. The parameter list reinforces the point: video accepts six parameters and ignores the rest, so the aspect ratio, seed, stylisation, and version switches that shape an image have no effect on a clip. Midjourney goes further and removes the image parameters used to create the source picture as soon as video generation begins.

This page runs one route on the model: image-to-video from a single uploaded frame, plus a first-and-last-frame variant that accepts two. That means the four mode switches and the absent audio control line up with the official model exactly, which is unusual among the video pages here, where the more common failure is offering something the model cannot do. Every difference on this page sits inside a setting. The duration selector starts below the official floor, the resolution selector omits the official default and lists a tier the model does not output, and the batch of four and the four extensions the official model performs have no controls here at all. Read the official documentation as the ceiling and the workspace as the floor you actually work on.

03 / Capabilities

What the official documentation establishes

Each point below is read from a Midjourney help centre article or from the launch announcement, rather than from a summary of them.

01

The first frame is the whole specification

A clip inherits its composition before generation starts, because the model animates the image it is handed rather than inventing a scene around a sentence. Midjourney describes the workflow as feeding it an image to serve as the first frame, and the output dimensions follow from that image. Since the ratio cannot be set separately, the frame you choose is also the framing you get, which moves a creative decision earlier in the process than most video tools expect.

02

An ending frame turns a pair of stills into a move

Supplying a second image as the endpoint converts two static frames into a controlled transition, and the model fills the space between them. Leave it out and the generation is held to the starting image alone. Point the endpoint back at the start and the result is a loop, which is why the web interface offers a Loop checkbox and the parameter list carries a loop switch of its own.

03

Motion strength is a trade with a stated cost

Low motion is the default and keeps a clip close to its source, with restrained subject movement and little camera travel. High motion allows the camera and the subject to travel further, and Midjourney states the price of that freedom as a greater chance of artefacts or an abrupt jump somewhere in the clip. That is a documented quality boundary rather than a rule of thumb, and it is the reason a motion test is worth running before a delivery is committed.

04

Four clips per prompt, priced into the same task

A video job returns four variations by default and the count can be set to one, two, or four. The GPU time table follows the split: four standard definition clips cost 8 minutes, two cost 4, and one costs 2. The per-clip cost is therefore flat rather than discounted for volume, which makes the batch size a choice about how many options to see rather than about efficiency.

05

Duration is a ladder rather than a slider

Generations start at 5 seconds and climb by 4 seconds per extension, up to four extensions, which puts the practical ceiling at 21 seconds. Midjourney states both that ceiling and the fact that an extension consumes the same GPU time as an initial generation. Length is bought in the same currency as quality, and neither is cheap, so duration planning belongs before the first submission rather than after it.

06

Two resolution tiers, and one of them is gated twice

The default output is 480p, described as standard definition. High definition means 720p, and reaching it takes a Standard, Pro, or Mega subscription and Fast mode, because Midjourney states that high definition generation is available on those plans but only in Fast mode. The Relax queue moves in the opposite direction and is restricted to standard definition even on the plans that offer it.

07

The parameter set is closed

Video generation is compatible with video-only parameters. Midjourney lists six and stops there, then adds that image parameters used to create the source picture are removed automatically once video generation begins. A prompt perfected on an image therefore cannot be carried into a clip intact, because most of the settings that shaped it do not exist in that context and are discarded rather than reinterpreted.

08

The published pixel table explains ratios instead of offering them

The help centre gives a table showing how a starting image maps to output dimensions. A square source produces 624 by 624 in standard definition and 960 by 960 in high definition; the widest listed source, 16:9, produces 832 by 464 or 1280 by 720. Midjourney presents this as an illustration of what a source ratio implies, and the wording matters, because the table is not a list of supported ratios and this page does not treat it as one.

04 / Scene ideas

Work that suits a five to twenty-one second clip

Each point below is read from a Midjourney help centre article or from the launch announcement, rather than from a summary of them.

Bringing an existing still to life

The route is built for material that already exists. A finished image, generated or uploaded, becomes the first frame, and the remaining decisions are how much movement to allow and how long to keep it running. That suits a library of stills that needs motion without a reshoot, and it suits images made long before anyone considered animating them.

Loops and ambient cutaways

Pointing the ending frame back at the starting image produces a loop, and the 5 second minimum makes it long enough to sit behind a page, a panel, or a kiosk. The absence of audio is an advantage in that setting rather than a limitation, because the clip drops into an existing sound design without a soundtrack that has to be stripped out first.

Transitions between two supplied frames

A first frame and an ending frame make a controlled move from one composition to another, which is what a product reveal or a scene change needs. The transition is generated rather than cut, so the intermediate frames belong to the model, but both endpoints stay exactly as supplied, which keeps the opening and closing images under direct control.

Testing movement before committing to a shot

Because four variations arrive together and high motion is documented as the riskier setting, a batch makes an inexpensive comparison. Running one prompt at both motion settings and choosing afterwards is the trade off the official documentation describes, done deliberately instead of discovered during delivery.

Extending a shot that already works

An extension inherits the original prompt or accepts an edited one, and each adds 4 seconds to material that already exists rather than starting a fresh clip. For work that must run longer than a single generation but must not look like two different shots, extending is the only route the model offers.

Sequences where the frame ratio is fixed

Because the ratio follows the starting image, a set of frames prepared in one proportion produces a set of clips in that proportion with no per-clip setting to forget. That is a convenience for vertical social formats and a constraint for anything that needs its ratio changed partway through, since a change there means regenerating rather than re-framing.

Storyboards that need motion tests

A drawn or rendered frame can be animated directly, which makes the route usable before any footage exists. The value is in judging whether movement reads at the intended pace and whether the subject survives it, both of which are cheaper to check at this stage than after a shoot has been scheduled.

Judging a clip frame by frame before delivery

The browser player scrubs frame by frame when Control or Command is held while the pointer moves, and the download menu offers a social-media encode alongside the raw file and a GIF. Any job where a single bad frame matters is worth reviewing that way rather than at full speed.

05 / Prompt lab

Prompt starters for the image-to-video route here

Each of these is written for the single image-led path available here, so the prompt assumes an uploaded frame and describes movement rather than appearance. Motion strength is set separately from the prompt, and the ending frame is only used where the example says so.

SCENE 01image to video

Portrait with one small movement

Hold the supplied face and let a single gesture carry the five seconds.

View prompt
Keep the uploaded person unchanged in face, hair, wardrobe, and the direction of the existing light. Add one slow turn of the head toward camera and a single breath, nothing else. Do not move the camera. Keep the background layout and the shadow on the wall exactly as supplied. No added objects, no captions, no text.
SCENE 02image to video

Product still, brought to life

Use one supplied studio frame and move only what has to move.

View prompt
Preserve the uploaded product shape, the label position, and the reflection on the surface. The camera does not move at all. A single hand enters from the right, lifts the lid, and withdraws. Keep the label legible and unchanged and keep the highlight along the left edge. Add nothing to the background and change no colour in the frame.
SCENE 03image to video

Environment drift on a locked-off frame

Check how the model handles atmosphere when nothing in the camera moves.

View prompt
Keep the uploaded street frame identical in perspective, colour, and signage. The camera stays fixed. Fog thickens from the far end toward the viewer, the nearest lamp flickers once and steadies, and one cyclist passes through the middle distance from left to right. Hold the geometry of the buildings unchanged for the whole clip.
SCENE 04image to video

Fabric and material study

A movement test where drift and weight are easy to judge.

View prompt
Keep the uploaded fabric unchanged in colour, weave, and the fold pattern already present. A steady draught lifts the near edge of the cloth and lets it fall back, then lifts it a second time more gently. Keep the seams and the stitching identical throughout. The camera drifts slowly to the right with no cut.
SCENE 05image to video

Interior with a forward walk-through

Use one supplied frame of a space and let the camera do the describing.

View prompt
Preserve the uploaded interior exactly in proportion, materials, and the position of every source of light. Move the camera in a steady forward push along the centre line, with no pan and no cut. Dust drifts through the shaft of light from the left window and a curtain settles at the far end. Keep the floor line and the ceiling height unchanged for the whole clip.
SCENE 06start end to video

Two supplied frames joined by a move

A first frame and an ending frame, with the model filling what sits between them.

View prompt
Treat the uploaded first frame and the uploaded ending frame as the two ends of one continuous move. Keep the subject, the wardrobe, and the direction of light identical in both. Travel between them without a cut, holding the horizon level and the scale of the subject constant. Do not introduce new objects, and do not change the colour of the scene between the two endpoints.
LIVE PRICING

How the V1 Video Model spends GPU time

Midjourney does not price video by the second, by the clip, or through an API. A subscription buys Fast GPU minutes, and a video task draws on them at documented rates: a batch of four standard definition clips costs 8 minutes, and the same batch in high definition costs 26. Two clips cost 4 minutes and 13, and a single clip costs 2 minutes and 7. Extending a clip costs the same as an initial generation, so a longer sequence is billed like several separate tasks. Fast GPU time is included at 3.3 hours on the 10 dollar plan, 15 hours on the 30 dollar plan, 30 hours on the 60 dollar plan, and 60 hours on the 120 dollar plan, with additional time sold at 4 dollars per hour on every tier. There is no free trial through the website or Discord. This site does not bill in dollars; the workspace quotes platform points, and the live estimate before submission is the price that applies here.

Open the workspace to see the current estimate
01

Platform points

The workspace quotes points before you submit. Points are set by this site and are not the official GPU time figures.

02

Resolution and duration on this route

The official model outputs 480p by default and 720p on Standard, Pro, and Mega in Fast mode only, starting at 5 seconds and reaching 21 after four extensions. This page exposes 720p and 1080p from 2 to 15 seconds, so it lists a resolution that is not an official output, omits the official default, and stops six seconds short of the official ceiling.

06 / Workflow

Running one generation end to end

  1. 01

    Start from a frame whose proportions are already decided

    Because the output ratio follows the starting image, the composition decision happens before the workspace is opened. Upload the frame at the proportions you intend to deliver, since changing your mind later means generating again rather than re-framing, and the environment behind the subject decides how much room the movement has to work in.

  2. 02

    Choose between automatic and manual movement

    Automatic lets the system read a movement out of the image and act on it. Manual asks for the movement in the prompt, which is the setting to use when the shot has to do something specific, such as one object travelling in one direction while the camera holds still. Writing the direction, the pace, and what must not change is more useful than describing the mood.

  3. 03

    Set motion strength deliberately

    Low motion is the default and the safer setting, keeping the clip close to its source frame with restrained movement. High motion allows larger camera and subject movement and carries a documented risk of artefacts or an abrupt jump. When a shot matters, running the same prompt at both settings and comparing the results is cheaper than regenerating after a rejection.

  4. 04

    Keep duration at five seconds or above

    The selector on this page starts at 2 seconds, below the official floor, and stops at 15, below the 21 seconds that four extensions would reach. Keep the value at or above 5 for a setting the model actually accepts, and plan the rest of the sequence knowing that this page has no extension control and no batch control.

  5. 05

    Review the clip before you accept it

    A clip is short enough to watch in full, and the browser player scrubs frame by frame when Control or Command is held while the pointer moves, which is how a bad frame gets found before someone else finds it. The download menu offers the raw file, a social-media encode, and a GIF, so the version you keep and the version you post do not have to be the same file.

Review before publishing

Ceilings, subscription terms, and what this page does not offer

The usual failure modes apply here as much as anywhere. Identity drifts across a clip, geometry shifts, a subject changes shape partway through, and text that reads cleanly in a still turns illegible once it moves. High motion raises those risks rather than lowering them, which is why the documentation frames it as the more artefact-prone setting and why low motion is the default. Generate at the smallest batch your judgement permits, watch the entire clip rather than its first second, and keep the prompts that worked, because the model reads a movement description literally and the same words are worth reusing.

The distance between the official model and this page is unusual in one respect and ordinary in another. It is ordinary because the settings differ: duration runs from 2 to 15 seconds here against an official start of 5 and a ceiling of 21, and the resolution selector lists 720p and 1080p against an official 480p default and a 720p high definition tier. It is unusual because the switches do not differ at all. Image-to-video and a first-and-last-frame route are present on both sides, text-to-video and reference-to-video are absent on both, and there is no audio control here because there is no audio capability there.

Subscription terms are part of this product rather than small print attached to it. High definition video at 720p requires a Standard, Pro, or Mega plan and Fast mode, so the resolution ladder is also a plan ladder. Relax mode for video is limited to Pro and Mega and to standard definition. There is no free trial through either the website or Discord. Ownership of what you create is stated plainly, with the qualification that a company earning more than 1,000,000 US dollars a year has to hold a Pro or Mega subscription. The plans article adds its own warning that availability and features can change, which is a reason to read the current terms rather than a dated summary of them.

Two boundaries sit outside the model itself and are worth stating anyway. The first is that there is no public API for video generation, so any page claiming an official integration is describing something that does not exist; this page runs its own route and says so. The second is the responsibility attached to uploading an image from outside Midjourney. The uploader has to hold the rights to it, using it to degrade or sexualise real people is prohibited and can end an account without a refund, and a job that the filter blocks does not consume GPU time. That last point is worth knowing before assuming a failed generation was charged for.

Sources

Official sources and how these facts were checked

Specifications, parameters, limits, plan terms, and GPU time figures come from Midjourney help centre articles and the launch announcement, read at the source rather than through third-party summaries. The help centre blocks automated readers, so every article was read in a browser; none of them requires a login, and nothing here is behind a paywall. Platform controls, points, and availability come from this site. Where the two differ, the difference is written out instead of smoothed over.

07 / FAQ

Midjourney Video questions, answered

Is Midjourney Video the same thing as the current image version?
No. Midjourney announced the video generator as V1 Video Model, and the workflow is called Image-to-Video. The help centre states that every image in a gallery can be animated, regardless of which version it was made with, which is the clearest official sign that the two lines are separate. Writing a current image version number next to the video model is the most common error in third-party coverage, and this page avoids it by naming no image version at all.
Can it generate a video from text alone?
No. The workflow is image-to-video, and the first frame is the input the model animates. A prompt can accompany that frame and shape the movement inside it, but there is no route where a text description produces a clip without an image feeding the first frame.
How long can a clip be?
Five seconds at the start, with four more seconds added per extension and a maximum of four extensions, which puts the ceiling at 21 seconds. Extending consumes the same GPU time as an initial generation, so length is not cheaper than a fresh clip. The selector on this page reaches 15 seconds and offers no extension control.
What resolution does it output?
Standard definition at 480p by default, and high definition at 720p on the Standard, Pro, and Mega plans in Fast mode. High definition is therefore gated twice, by plan and by mode, and Relax mode for video is limited to standard definition. This page lists 720p and 1080p, so it offers a tier the model does not output and omits the official default.
How many videos does one prompt produce?
Four by default, and the count can be set to one, two, or four. The GPU time table follows the same split: four standard definition clips cost 8 minutes, two cost 4, and one costs 2, with high definition at 26, 13, and 7 minutes. This page produces one clip per submission and exposes no batch control.
Does the video have sound?
No. There is no audio parameter among the six video-only parameters, and no output carries a soundtrack. Midjourney adds that image parameters used to create the source picture are removed automatically once video generation begins, so the parameter set behind a clip is smaller than the set behind the image it came from.
Can I control the aspect ratio?
Not directly. The ratio follows the starting image, with Midjourney noting that it may adjust it in some cases. The help centre publishes a table showing what a source ratio implies for output dimensions: a square source produces 624 by 624 in standard definition and 960 by 960 in high definition, while a 16:9 source produces 832 by 464 or 1280 by 720. That table illustrates the relationship rather than listing supported ratios.
Which parameters work with video?
Six, and only six: motion set to low or high, raw, loop, end, and the batch size. Everything the model accepts for images, including the version switch and the aspect ratio, falls outside that set and has no effect on a clip. Midjourney makes the boundary explicit by removing the image parameters of the source picture when it starts generating video from it.
Is there a free trial?
Not through the website or Discord. Midjourney states that limited free trial access is currently available on the niji journey app instead. Fast GPU time is included with each paid tier, from 3.3 hours on the 10 dollar plan up to 60 hours on the 120 dollar plan, and additional time is sold at 4 dollars per hour on every tier.
How is a video priced?
By GPU time drawn from a subscription, not by the second and not per clip. A batch of four standard definition videos consumes 8 minutes and the same batch in high definition consumes 26. There is no published per-second rate and no API billing, so any figure expressed as a price per second comes from a reseller rather than from Midjourney, and this page does not quote it.
Does this page use an official API?
No, and the distinction matters because Midjourney publishes no public API for video. This page is an independent creation interface running its own route. It is not affiliated with Midjourney and does not claim an integration that does not exist.
Can I animate an image I made outside Midjourney?
Yes. The documentation covers animating images uploaded from outside Midjourney. Responsibility for doing so sits with the uploader, who has to hold the rights to the image, must not use it to degrade or sexualise real people, and faces account termination without a refund for violations. A job that the filter blocks does not consume GPU time.
Can I get a loop instead of a one-way move?
Yes, by making the ending frame the same image as the starting frame. The web interface exposes this as a Loop checkbox and the parameter list carries a loop switch, and the result returns to its own first frame. Leaving the ending frame unused holds the generation to the starting image alone, which produces movement without a return.

Other video routes on this platform

01

Seedance 2.5

A closed flagship route with longer clips and 10-bit 1080p.

02

Hailuo H3

An open-weight route that generates native stereo audio in the same pass.

03

Kling 3.0 Turbo

A text-and-image route with a different provider boundary.

04

Image to Video

Open the general image-led video workspace with model selection.

Render your next scene

Animate an image with Midjourney Video

Return to the live workspace, upload one frame at the ratio you intend to deliver, keep the duration at five seconds or above, and review the current point estimate before you generate.

Open the video workspace