What you can make
Video
Clips run 2 to 20 seconds at up to 4K, and you can start one from a prompt, a frame, two frames, ten frames or a clip you already have.
Every clip Plnty makes is short. That is a property of the models rather than of Plnty, and it shapes how the video tools are built: you generate a few seconds, then extend, restyle or cut them together. ## How long Two seconds at the floor, twenty at the ceiling, with most models landing between four and fifteen. WAN 2.7 goes shortest. FLUX 3 goes longest and renders picture and synchronized audio in the same pass, up to 20 seconds. Past that, **video-extender** continues an existing clip. It takes the true last frame as the seed and the clip's tail as a motion reference, has Seedance 2.0 generate what happens next, then stitches the two locally with seam-frame cleanup. That buys another 4 to 15 seconds at up to 1080p, and you can run it again. ## How good Most models offer 720p and 1080p. A few sit at 480p, and MiniMax H3 renders at 2K with native stereo audio. For native 4K there is one route: **scene-director**, which runs Kling o3 4K. Cast one to seven reference images, tag each as an Element the model has to keep or an Image it should only take a look from, then direct the shot with `@Element` and `@Image` mentions. :::caution[Two limits that bite] Every cast image must be at least 300x300px, checked before any credits burn, and the error names the slot that failed. Prompt segments cap at 512 characters, so write tight. ::: Anything already rendered can be pushed up afterwards with **video-upscale**, which runs SeedVR to 720p, 1080p, 1440p or 2160p. ## From what The starting point is the real choice, more than the model is. | Start from | Where to go | |---|---| | A prompt on its own | all-video-models, gemini-text-to-video | | One frame | all-video-models, any image-to-video pill | | A first and a last frame | first-last-frame | | Up to ten anchor frames | flux-3, which generates the motion between them | | A clip you already have | video-restyler, aleph-2, video-extender | | A cast of images, clips and sound | minimax-h3-reference, seedance-director | **seedance-director** works differently from the rest. You draw a freehand path over the hero image to say where the camera and the subject move, and that drawing never reaches the model. It feeds a vision model inside the card, which translates your geometry into landmark-referenced motion words and writes the prompt from them. You then edit that prompt before running. ## Afterwards **mmaudio-v2** scores a silent clip, **plnty-fx** grades it, **remove-video-bg** cuts it out with no green screen, and **audio-video-merge** assembles the pieces into a finished MP4 using ffmpeg compiled into your browser.Every clip Plnty makes is short. That is a property of the models rather than of Plnty, and it shapes how the video tools are built: you generate a few seconds, then extend, restyle or cut them together.
How long
Section titled “How long”Two seconds at the floor, twenty at the ceiling, with most models landing between four and fifteen. WAN 2.7 goes shortest. FLUX 3 goes longest and renders picture and synchronized audio in the same pass, up to 20 seconds.
Past that, video-extender continues an existing clip. It takes the true last frame as the seed and the clip’s tail as a motion reference, has Seedance 2.0 generate what happens next, then stitches the two locally with seam-frame cleanup. That buys another 4 to 15 seconds at up to 1080p, and you can run it again.
How good
Section titled “How good”Most models offer 720p and 1080p. A few sit at 480p, and MiniMax H3 renders at 2K with native stereo audio.
For native 4K there is one route: scene-director, which runs Kling o3 4K. Cast
one to seven reference images, tag each as an Element the model has to keep or an
Image it should only take a look from, then direct the shot with @Element and
@Image mentions.
Anything already rendered can be pushed up afterwards with video-upscale, which runs SeedVR to 720p, 1080p, 1440p or 2160p.
From what
Section titled “From what”The starting point is the real choice, more than the model is.
| Start from | Where to go |
|---|---|
| A prompt on its own | all-video-models, gemini-text-to-video |
| One frame | all-video-models, any image-to-video pill |
| A first and a last frame | first-last-frame |
| Up to ten anchor frames | flux-3, which generates the motion between them |
| A clip you already have | video-restyler, aleph-2, video-extender |
| A cast of images, clips and sound | minimax-h3-reference, seedance-director |
seedance-director works differently from the rest. You draw a freehand path over the hero image to say where the camera and the subject move, and that drawing never reaches the model. It feeds a vision model inside the card, which translates your geometry into landmark-referenced motion words and writes the prompt from them. You then edit that prompt before running.
Afterwards
Section titled “Afterwards”mmaudio-v2 scores a silent clip, plnty-fx grades it, remove-video-bg cuts it out with no green screen, and audio-video-merge assembles the pieces into a finished MP4 using ffmpeg compiled into your browser.