video-gen

On the rail as video-gen, kling-o3, kling-o3-pro, wan-2.7, wan-3.0, happy-horse, grok-imagine, flux-3, ltx-2.5-fast, ltx-2.5-pro.

When to use it

Making a clip from a start frame and an optional end frame, from words alone, or from pictures, clips and sounds named in the words, and editing or extending a clip you already have. The toggle on the card picks the side.

When to reach for something else

Each model takes its own inputs and lengths, and the card checks them before Run. For one still turned by a camera move, a tool built around that move is closer.

How to use it

1. Use the card’s Video area. Start and end frames, references, or a clip to edit. The toggle on the card picks which side you are on.
2. Write what you want in Prompt.
3. Set Duration, Resolution, Ratio, Draft first, Generate audio, Direction on the card.
4. Pick a mode on the card: Frames, Subjects.
5. Pick a model in the pill at the top of the card, or keep Seedance 2.5.
6. Press Run.

On the card

Control	What it does	Choices
Video	Start and end frames, references, or a clip to edit. The toggle on the card picks which side you are on.
Prompt	What you want, in your own words. The card sends it with your inputs.	Your words
Duration	How long the clip is, in seconds. On some models a longer clip costs more.	4 s, 5 s, 6 s, 7 s, 8 s, 9 s, 10 s, 11 s, 12 s, 13 s, 14 s, 15 s, 16 s, 17 s, 18 s, 19 s, 20 s, 21 s, 22 s, 23 s, 24 s, 25 s, 26 s, 27 s, 28 s, 29 s, 30 s
Resolution	How big and detailed the result is. Higher looks sharper but costs more and takes longer.	480p, 720p, 1080p
Ratio	The shape of the clip. Adaptive picks one that suits what you dropped in.	Adaptive, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16
Draft first	Renders a 480p draft. Finish the one you like in 1080p, within seven days.	On or off
Generate audio	Render a soundtrack with the clip. Costs nothing extra.	On or off
Direction	Where the new seconds go: after the clip ends, or before it starts.	After the clip, Before it
Run	Starts the run. The number on it is what this run costs in credits, worked out before you press.
×1	Runs the card several times in a row with the same settings.	1 to 99

Modes

Frames

Start from a start frame and, where the model takes one, an end frame; leave them empty to start from words where the model can.

Models in this mode: Kling o3, LTX 2.5, Seedance 2.5, Wan 3.0, Seedance 2.0, FLUX 3, Gemini Omni, H3, Veo 3.1, Runway Gen-4.5, Grok Imagine 1.5, WAN 2.7, Ray 3.2, Happy Horse 1.1.

Subjects

Cast pictures, clips and sounds, then name them in the words.

Models in this mode: Seedance 2.5, Seedance 2.0, Gemini Omni, Wan 3.0, H3, Kling o3 4K, Veo 3.1, Grok Imagine 1.5, Happy Horse 1.1.

Models

The pill at the top of the card picks the model. Seedance 2.5 is the one a new card starts on.

Model	Good for
Kling o3 Standard	The default start-frame animator at 720p: 3 to 15 seconds, an optional end frame, a seamless loop, hard cuts between prompt lines or Kling planning the cuts itself, cheaper than Pro. It also drives the turntable card.
Kling o3 Pro	Cinematic start-frame animation at 1080p, 3 to 15 seconds, with an optional end frame for a first-to-last interpolation, a seamless-loop option and a shot mode where Kling plans the cuts itself.
Kling o3 4K	A start frame animated into a native 4K clip for a big-screen deliverable, 3 to 15 seconds, with an optional end frame, a seamless-loop option and a shot mode where Kling plans the cuts itself; audio is free at this tier.
Kling o3 Edit	Changing what is in a clip that already exists (the weather, a costume, the setting, the look) from a written instruction, keeping its motion, camera and length, at 720p, 1080p or native 4K.
Kling o3 4K References	Native 4K from a cast of 1 to 7 images tagged as identity elements or look references and directed with @Element and @Image mentions, with optional start and end frames and multi-shot segments, 3 to 15 seconds.
LTX 2.5 Pro	The quality tier of the LTX 2.5 family: picture and synchronized audio at 6, 8 or 10 seconds up to 1080p, with the same named camera moves, spending the extra compute on holding detail and motion together.
LTX 2.5 Fast	Long, high-resolution clips with native audio: 6 to 20 seconds in even steps, up to 4K at up to 50 fps, from a start frame, a start-and-end pair or the prompt alone, with a named camera move (dolly, jib, focus shift) instead of a prompt.
LTX 2.5 Pro, from a sound	The quality tier for sound-driven clips: the same 2 to 20 second, 1080p, audio-length render with more compute spent on holding the picture together, for the take that has to hold up.
LTX 2.5 Fast, from a sound	Picture cut to a piece of sound, 2 to 20 seconds, at 1080p and exactly as long as the audio, from a start frame or a prompt; the default tier of the two.
Seedance 2.5	Casting up to 30 images, 10 clips and 10 sounds named in the prompt, 4 to 30 s up to 1080p with native audio; also a first and last frame or words alone, Edit to re-shoot one clip, and Extend to continue one before or after it.
Seedance 2.0	The first-last-frame default: 4 to 15 seconds interpolating from a first frame to an optional last one, with native audio at no extra cost and up to 4K.
Seedance 2.0 Fast	The cheaper Seedance 2.0 for a first frame moving toward an optional last one: 4 to 15 seconds at 480p or 720p, with native audio at no extra cost, when the motion matters more than the pixels.
Seedance 2.0 References	Weaving several references into one clip: lead with a hero image or video, stack up to 9 images, 3 clips and 3 sounds named in the prompt; 4 to 15 s up to 4K with native audio, or edit or extend one clip.
Wan 3.0	The longest single-pass clips in the product: 2 to 30 seconds with synchronized audio rendered alongside the picture, from a start frame, a start-and-end pair, or the prompt alone, billed per second by resolution.
Wan 3.0 References	Thirty seconds from one cast: up to 10 images, 5 motion clips and 5 sounds named in the prompt, rendered in a single pass with synchronized audio at up to 1080p, and only the prompt is required.
FLUX 3	Picture and synchronized audio in one pass, 5 to 20 seconds at 720p or 1080p, from a start frame, a start-and-end pair, or the prompt alone, with a Draft quality that rehearses cheaply at 720p before a full render.
FLUX 3 Keyframes	Up to 10 images pinned to moments on a timeline, with the motion between them generated, 5 to 20 seconds at 720p or 1080p with synchronized audio. Use it when the shot has to pass through pictures the person already has.
FLUX 3 Edit	In-place edits of a clip from a text instruction: swap a character, rebuild the background, restyle the shot or rewrite a spoken line with lip sync following it, with motion, timing and framing kept. Billed per second of the source clip.
FLUX 3 Extend	Continuing a clip past its final frame with 5 to 20 new seconds at 720p or 1080p, keeping the motion, scene and sound consistent, with a Draft quality for a cheap rehearsal.
Gemini Omni	Prompt-only clips with synced audio, grounded in Gemini world knowledge, 3 to 10 seconds, for when there is no picture to start from.
Gemini Omni	Animating one still into a 3 to 10 second clip with synced audio from a motion prompt, or moving from a first picture to an end picture, the simplest of the four modes.
Gemini Omni References	The flagship reference mode: up to 10 images and 3 short clips bound to the prompt with @Image and @Video mentions, so each reference lands in the clip where you name it; 3 to 10 seconds with synced audio, 360p to 4K.
Gemini Omni Edit	Conversational edits of a clip in place from a simple instruction, keeping everything else as it was, with synced audio; the card opens on this mode.
H3 Max	The default of the MiniMax card: closer prompt adherence than the base model, native stereo audio, 480p, 768p or a refined 1080p, 5 to 15 seconds, from a start frame, an end frame, both or text alone, with an optional soundtrack.
H3	Clips with native stereo audio, 5 to 15 seconds, from a start frame, an end frame, both or the prompt alone, at 480p or 768p or upscaled to 2K and 4K, with an optional soundtrack of your own; the high-resolution finish beside H3 Max.
H3 References	Casting up to 9 images, 3 motion clips and 3 audio clips and directing them by name in the prompt, rendered at 2K with native stereo audio, 5 to 15 seconds.
H3 Max References	Casting up to 9 images, 3 motion clips and 3 sound clips and directing them by name in the prompt, on the retuned H3 Max that follows the words more closely, at 480p, 768p or a refined 1080p with native stereo audio, 5 to 15 seconds.
H3 Max Extend	Making a clip run longer: H3 Max reads the whole clip and adds 5 to 15 seconds that carry its motion and sound forward, returning the source and the continuation as one clip at 480p, 768p, 1080p or 2K.
H3 Max Insert	Adding a new scene inside a clip: H3 Max leaves the footage at one point, plays a new scene of 5 to 13 seconds guided by words and reference pictures or clips, and comes back to the footage at a second point, all as one clip with its sound.
Veo 3.1 Lite	Cheap Veo drafts and quick shots: 4, 6 or 8 seconds at 720p (8 seconds at 1080p) from words, a first frame or a first and last frame, with sound always included.
Veo 3.1 Fast	Short first-to-last-frame transitions of 4, 6 or 8 seconds at 720p or 1080p, with native audio on by default, in 16:9 or 9:16.
Veo 3.1	Premium 4, 6 or 8 second clips with native sound from words, a first frame or a first and last frame, and the only Veo that keeps up to three reference pictures (a person, a character, a product) consistent in the clip.
Veo 3.1 Fast Extend	Continuing a landscape or portrait clip by exactly 7 seconds at 720p, with sound made to match, returned joined to the clip as one; a voice in the last second of the clip keeps talking.
Runway Gen-4.5	Cinematic clips written in shot language (a dolly, a rack focus, handheld energy) from a prompt or a start frame, 2 to 10 seconds, with seven shapes including 2.35:1 scope.
Grok Imagine 1.5	A clip with its own sound from words alone or from a start frame, 1 to 15 seconds at up to 1080p, prompted in camera language; from words it takes seven shapes, from a picture it keeps the picture shape.
Grok Imagine 1.5 References	Keeping up to seven people, objects or places from pictures in one clip and directing them by name (Image 1, Image 2), 1 to 15 seconds at 480p or 720p, with sound made in the same pass.
Grok Imagine Edit	Changing what is in a short clip (a colour, the weather, a costume, the look) from a written instruction, keeping its length and shape, at 480p or 720p.
Grok Imagine Extend	Making a clip of 2 to 15 seconds run longer: it adds 2 to 10 seconds that carry on from the last frame, returning the clip and the new footage joined as one.
WAN 2.7	Start-frame animation from 2 to 15 seconds at 720p or 1080p with an optional end frame, an optional driving sound that sets the timing and lip-sync, a negative prompt and automatic expansion of short prompts.
WAN 2.7 Extend	Making a short clip run longer: WAN 2.7 reads a clip of 2 to 10 seconds and adds 2 to 13 seconds that carry its motion forward, returning the clip and the continuation as one, 15 seconds at most, at 720p or 1080p.
Ray 3.2	Five-second clips from a start frame and an optional last frame, or 5 and 10 second clips from words alone, with native HDR at 720p or 1080p, a seamless loop option and six aspect ratios.
Ray 3.2 Extend	Making a Ray 3.2 clip that was made in Plnty longer: five more seconds that continue it forward, or that lead into it backward, at 540p, 720p or 1080p.
Happy Horse 1.1	A clip from a start frame or from words alone, 3 to 15 seconds at 720p or 1080p, with nine shapes from 21:9 to 9:21 when there is no picture and a seed for repeatable takes.
Happy Horse 1.1 Characters	Keeping up to nine people or characters from pictures in one clip and directing them by name (Image 1, Image 2), 3 to 15 seconds at 720p or 1080p, in nine shapes.
Happy Horse Edit	Restyling a clip or swapping something in it from words and up to five pictures, keeping its shape and length, at 720p or 1080p, with the choice of keeping the clip sound or letting the model make new sound.

Kling o3 Standard

Not the one for: Audio is off by default and adds cost when on; the start frame must be at least 300x300, and there is no text-to-video route.

This model has no settings behind the settings button.

Kling o3 Pro

Not the one for: Audio is off by default and costs extra, it needs a start frame and there is no text-to-video route; for a native 4K clip use Kling o3 4K.

This model has no settings behind the settings button.

Kling o3 4K

Not the one for: Four to five times the Standard price per second, so it is the finish rather than the draft; it needs a start frame and has no text-to-video route.

This model has no settings behind the settings button.

Kling o3 Edit

Not the one for: It edits and does not extend or re-time: the result is as long as the source, which must be a 3 to 15 second MP4 or MOV at least 720 pixels across, and 4K costs about two and a half times 1080p per second.

This model has no settings behind the settings button.

Kling o3 4K References

Not the one for: Every cast image must be at least 300x300 and each prompt segment caps at 512 characters; only 16:9, 9:16 and 1:1 shapes, and audio is off by default.

This model has no settings behind the settings button.

LTX 2.5 Pro

Not the one for: Capped at 10 seconds and 1080p with no 48 fps, and only 16:9, 9:16 or auto shapes; 20 seconds or 4K belongs to the fast tier.

This model has no settings behind the settings button.

LTX 2.5 Fast

Not the one for: Only 16:9, 9:16 or auto shapes, no square or 21:9, and durations are even numbers only; the quality tier spends more compute on holding detail together if the fast render looks loose.

This model has no settings behind the settings button.

LTX 2.5 Pro, from a sound

Not the one for: Costs more per second of audio than the fast tier for the same shape and limits; it still needs a start frame or a prompt, and the clip length is always the audio length.

This model has no settings behind the settings button.

LTX 2.5 Fast, from a sound

Not the one for: It needs a start frame or a prompt on top of the audio, since the sound decides length and pacing but not content; there is no duration, resolution or camera-move setting, and shapes are 16:9, 9:16 or auto.

This model has no settings behind the settings button.

Seedance 2.5

Not the one for: No 4K (that stays with Seedance 2.0), frames and references cannot be mixed in one run, reference-clip seconds are billed on top of the output, and the multicam edit refuses clips whose people look photoreal.

This model has no settings behind the settings button.

Seedance 2.0

Not the one for: A single start frame plus an optional end frame only, no reference cast; 4K costs about five times 720p and renders one at a time across the account.

This model has no settings behind the settings button.

Seedance 2.0 Fast

Not the one for: Stops at 720p, so a 1080p or 4K deliverable belongs to the standard Seedance 2.0; frames only, never a reference cast, and a last frame always needs a first.

This model has no settings behind the settings button.

Seedance 2.0 References

Not the one for: It needs a hero image or a hero video; clips and sounds run 15 seconds at most, sounds as MP3 or WAV. For a song keep audio off, since generated audio can be refused as copyrighted, and reference-clip seconds bill on top.

This model has no settings behind the settings button.

Wan 3.0

Not the one for: Duration comes in fixed steps (2, 5, 10, 15, 20, 30) and the aspect list has no 21:9; a 30-second 1080p run is the largest single click in the product, so keep long takes at 720p unless the delivery needs more.

This model has no settings behind the settings button.

Wan 3.0 References

Not the one for: Reference clips and sounds each total 15 seconds at most across their slots, the aspect list has no 21:9, and 1080p is the ceiling.

This model has no settings behind the settings button.

FLUX 3

Not the one for: Draft is 720p only and its saving only shows from the second take; the shortest clip is 5 seconds, and nothing runs past 20.

This model has no settings behind the settings button.

FLUX 3 Keyframes

Not the one for: It generates motion between anchors, it does not morph one picture into another: two unrelated anchors are usually answered with a cut, because a cut satisfies both exactly. One anchor alone has nothing to move between.

This model has no settings behind the settings button.

FLUX 3 Edit

Not the one for: The clip must be an MP4 under 15 seconds and the result is exactly as long as the input; it cannot lengthen a clip and takes no style image.

This model has no settings behind the settings button.

FLUX 3 Extend

Not the one for: The source clip must be an MP4 under 15 seconds; it is priced higher per second than generating from frames, and Draft is 720p only.

This model has no settings behind the settings button.

Gemini Omni

Not the one for: It reads no images or video, so a dropped still is ignored; only 16:9 or 9:16, and nothing past 10 seconds.

This model has no settings behind the settings button.

Gemini Omni

Not the one for: One start picture and an optional end picture, with no extra references; 16:9 or 9:16 shapes, and 10 seconds at most.

This model has no settings behind the settings button.

Gemini Omni References

Not the one for: Clips stop at 10 seconds in 16:9 or 9:16 only; a reference clip may run 3 seconds at most and its sound is ignored, and 4K costs three times the 720p rate per second.

This model has no settings behind the settings button.

Gemini Omni Edit

Not the one for: Unavailable for uploaded clips in the EEA, Switzerland and the UK; it ignores aspect ratio and duration, so the result is the length of the clip you dropped.

This model has no settings behind the settings button.

H3 Max

Not the one for: Tops out at 1080p with no 2K or 4K upscale, so it is not the pick for a big-screen finish; the aspect setting only applies when starting from text.

This model has no settings behind the settings button.

H3

Not the one for: 2K and 4K are upscales of a 768p render, not native detail; the aspect setting applies only when starting from text, and a soundtrack must run at least 2 seconds and weigh 15 MB at most.

This model has no settings behind the settings button.

H3 References

Not the one for: Audio cannot be the only reference, reference-clip seconds are billed on top of the output, and 2K is the only resolution.

This model has no settings behind the settings button.

H3 Max References

Not the one for: A sound cannot be the only reference and 12 files is the most it takes; references past a small allowance bill on top of the clip, a motion clip at 768p or 1080p most of all, and there is no 2K or 4K.

This model has no settings behind the settings button.

H3 Max Extend

Not the one for: It continues what is already happening, so it cannot start a new scene, and the whole clip comes back at the chosen resolution, so a sharp source extended at 480p loses detail in its own footage too.

This model has no settings behind the settings button.

H3 Max Insert

Not the one for: It keeps the clip’s own look and returns to its footage, so it cannot restyle the whole clip, and it keeps the first 1.6 seconds and the last 1.5 seconds as they are.

This model has no settings behind the settings button.

Veo 3.1 Lite

Not the one for: No reference pictures, no 4K and no way to turn the sound off; 1080p makes 8 second clips only, and it is a step down in quality from Veo 3.1 Fast and Veo 3.1.

This model has no settings behind the settings button.

Veo 3.1 Fast

Not the one for: Capped at 8 seconds and two aspect ratios; audio adds cost, and the source frames work best at 1280x720 or 720x1280.

This model has no settings behind the settings button.

Veo 3.1

Not the one for: The dearest Veo: sound doubles the price per second. 1080p and reference pictures make 8 second clips only, references also force 16:9 and cannot take an end frame, and there is no 4K here.

This model has no settings behind the settings button.

Veo 3.1 Fast Extend

Not the one for: Every run adds exactly 7 seconds at 720p, the clip must be 16:9 or 9:16, and the model is tuned for continuing clips Veo itself made.

This model has no settings behind the settings button.

Runway Gen-4.5

Not the one for: No audio option, no end frame, no resolution choice, and nothing past 10 seconds.

This model has no settings behind the settings button.

Grok Imagine 1.5

Not the one for: The sound cannot be switched off, there is no end frame, and a start frame fixes the shape; the source suggests 5 to 8 seconds for the steadiest motion.

This model has no settings behind the settings button.

Grok Imagine 1.5 References

Not the one for: It stops at 720p, the pictures guide the look rather than pin a first frame, it cannot speak in a chosen voice here, and each picture adds a little to the price.

This model has no settings behind the settings button.

Grok Imagine Edit

Not the one for: Only the first 8 seconds of a clip are used and it is shrunk to about 480p before the edit, so fine detail softens; it cannot lengthen or re-time the clip.

This model has no settings behind the settings button.

Grok Imagine Extend

Not the one for: The clip must be an MP4 of 2 to 15 seconds, the result stays at the source resolution up to 720p, and the source seconds are billed as well as the new ones.

This model has no settings behind the settings button.

WAN 2.7

Not the one for: No text-to-video route and no aspect-ratio setting, since the clip takes the shape of the start frame; a driving sound longer than the clip is cut to the clip, and a shorter one leaves the rest silent.

This model has no settings behind the settings button.

WAN 2.7 Extend

Not the one for: The clip and what it adds stay within 15 seconds, so a 10-second clip gains 5 at most, and the whole result is billed, the source seconds included; it takes no sound of your own and no end frame.

This model has no settings behind the settings button.

Ray 3.2

Not the one for: Ten seconds works only from words alone, without HDR or Loop; a clip anchored on a frame is 5 seconds. Loop takes no end frame and no HDR, there is no audio option, and HDR doubles the cost.

This model has no settings behind the settings button.

Ray 3.2 Extend

Not the one for: It only extends Ray 3.2 clips made in Plnty, never an upload or another model’s clip; it adds five seconds per run, with no HDR and no sound.

This model has no settings behind the settings button.

Happy Horse 1.1

Not the one for: No end frame and no sound of your own; with a start frame the clip takes the picture shape, and the picture must be at least 300 pixels on each side.

This model has no settings behind the settings button.

Happy Horse 1.1 Characters

Not the one for: Each picture needs a shortest side of 400 pixels, the pictures set who appears rather than the first frame, and there is no sound of your own.

This model has no settings behind the settings button.

Happy Horse Edit

Not the one for: Only the first 15 seconds of a clip come back, and both the clip and the result are billed per second, so it costs about twice a new clip of the same length.

This model has no settings behind the settings button.