Correction, 2026-10-02: the page did not say that 8frame has no tool that re-syncs the mouth on footage you already have and no dubbing or voice cloning (its lip-sync tools make a new clip from a still portrait and an audio track), and it described the Video Editor mode only for captions. Both are covered below, with what the Video Editor and the canvas Montage node can do to a clip without a model. It also did not say that the Montage node's output has no sound: it drops the audio of the footage it joins.
Correction, 2026-10-01: An earlier version said five models on 8frame work from a video you already have. It left out Kling O1 Video, Kling 2.6 Pro Motion Control, Wan 2.2 video-to-video, ID-V2V Restyle, Wan 3.0 Prime with a reference video and Runway Act Two. It also said Gemini Omni Flash can extend uploaded video on 8frame (its edit mode here returns a clip as long as the one you give it), that Seedance takes up to ten reference clips (the canvas node sends one), and that 8frame cannot add captions (the Video Editor mode has caption layers). Those are fixed below, and the Wan 2.7 VideoEdit input limit (2 to 10 seconds) is filled in. It also said every model refuses a clip over its limit; not every model checks the length before it runs.
A dozen models on the 8frame canvas can work from a video you already have instead of starting from text or a still. They differ in two things that decide the choice: how long an input clip they accept, and whether they bill on the clip you give them or the clip you get back. The cheapest edit of a 5-second clip is Grok Imagine at 41 credits at 480p; the longest input is Happy Horse, which accepts up to 60 seconds and bills on at most 15; Gemini Omni Flash edits clips of 3 to 10 seconds, and Google says editing uploaded video is not available to users in the EEA, Switzerland or the UK.
TL;DR
| Model on 8frame | What it does to a clip | Input limit | Price |
|---|---|---|---|
| Grok Imagine, edit | Changes an existing clip, with audio | up to 15 s | 41 (480p) / 54 (720p) for a 5 s clip |
| Grok Imagine, extend | Adds 2 to 10 seconds | up to 15 s | 54 for 5 s added to a 5 s clip (720p) |
| Wan 2.7 VideoEdit | Edits a clip from a text instruction into a 2 to 10 second result, 720p or 1080p | 2 to 10 s | 68 for 5 s of output |
| Wan 2.2 video-to-video | Turns a clip into a new one from a prompt, 720p | 50 per 5 s of output | |
| Gemini Omni Flash, edit | Edits uploaded video, with audio; the result is as long as the input | 3 to 10 s, 100 MB | 85 for 5 s, 169 for 10 s |
| Happy Horse, video edit | Edits with up to 5 reference images | 3 to 60 s, 100 MB | 189 (720p) / 378 (1080p) per 5 s, billed on at most 15 s |
| Kling O1 Video | Edits a clip with up to 4 elements or reference images, or uses the clip alone as a reference for a new shot | 3 to 10 s, 100 MB | 109 for 5 s, 218 for 10 s |
| ID-V2V Restyle / Relight | Carries an edited first frame through the clip | up to 15 s, 100 MB | 27 per second: 135 for 5 s |
| Kling 2.6 Pro Motion Control | Moves a character image with the motion of your clip | 3 to 30 s, 100 MB | 95 for a 5 s clip, 189 for 10 s, 567 for 30 s |
| Seedance 2.5, reference video | Uses a clip as a reference for a new one | one clip on the canvas, up to 30 s | 291 (480p) / 654 (720p) for 5 s |
| Seedance 2.0, reference video | Uses a clip as a reference for a new one | one clip on the canvas, up to 15 s | 60 (480p) / 132 (720p) for 5 s |
| Wan 3.0 Prime, reference video | Uses a clip as a reference for a new one | 46 / 95 / 189 for 5 s (480p / 720p / 1080p) |
How each one bills
Grok Imagine charges for the output seconds plus a small amount per second of input. Editing has no separate duration: the result tracks the input clip, so a 5-second input costs 41 credits at 480p or 54 at 720p. Extending is billed at the 720p rate only.
Wan 2.7 VideoEdit charges per second of output, about 13.5 credits a second, so 5 seconds is 68 and 10 seconds is 135. It defaults to 1080p, and 720p costs the same. Wan 2.2 video-to-video is 50 credits per 5 seconds of output (5, 10 or 15 seconds) at 720p.
Gemini Omni Flash prices an edit off the length of the input, clamped to 3 to 10 seconds, at the same rate as a normal generation: 85 credits for 5 seconds, 135 for 8, 169 for 10. Audio is always generated. On 8frame its edit mode returns a clip as long as the input, so it edits but does not extend. Google's Gemini API documentation says editing or extending uploaded video is not available to users in the EEA, Switzerland or the UK; that note is about Google's own API, and we have not tested whether it applies when 8frame calls the model.
Happy Horse accepts long inputs, up to 60 seconds, but bills on the shorter of the input and 15 seconds, at 189 credits per 5 seconds at 720p and 378 at 1080p. A 15-second edit is 567 at 720p.
Kling O1 Video is 109 credits for 5 seconds and 218 for 10. With elements or reference images (up to 4 in total) it edits the clip you give it; with the clip alone it uses it as a reference for a new shot.
ID-V2V restyles or relights a clip: you give it the source clip and a first frame you have already edited, and it carries that look through the clip. It bills 27 credits per second of the source clip, 135 for 5 seconds and 405 for 15; a longer clip is refused. 480p and 720p cost the same, and Relight costs the same as Restyle.
Kling 2.6 Pro Motion Control takes a character image and a clip, and bills on the length of the clip at 18.9 credits a second: 95 for 5 seconds, 189 for 10, 567 for 30. The clip can run 3 to 30 seconds (10 in its image-orientation mode). Runway Act Two does a similar job, driving a character from a performance video at 8 credits per second, 3 to 30 seconds.
Seedance is not an editor in the strict sense: it uses your clip as a reference for a new generation, for when you want to carry a motion or a look into a new shot. Reference video is its expensive tier: 654 credits for 5 seconds at 720p on Seedance 2.5 against 157 without video, and 291 at 480p. Seedance 2.0 charges 132 at 720p and 60 at 480p. The canvas node takes one reference clip; the model accepts clips of up to 30 seconds on 2.5 and 15 on 2.0.
Wan 3.0 Prime takes a reference video on the canvas at its normal price: 46, 95 or 189 credits for 5 seconds at 480p, 720p or 1080p. Base Wan 3.0 takes no video on the canvas; through the MCP connector a clip sent to it selects its reference-to-video mode, 34 / 68 / 135 for 5 seconds.
Choosing
- Short tweak to a short clip, cheaply: Grok Imagine edit. At 720p, Wan 2.2 video-to-video (50 for 5 seconds) is a few credits under Grok's 54.
- A clip that needs to be longer: Grok Imagine extend, which adds 2 to 10 seconds.
- A long source clip: Happy Horse, which takes up to 60 seconds.
- A new look on the same footage: ID-V2V from an edited first frame, or Wan 2.7 VideoEdit from a text instruction.
- A character that moves the way someone moves in your clip: Kling 2.6 Pro Motion Control or Runway Act Two.
- Carrying a movement or a style into a new shot: Seedance or Wan 3.0 Prime with a reference video.
- A 1080p result from an edit: Wan 2.7 VideoEdit or Happy Horse at 1080p.
Upload formats and size limits for each input are in upload limits by model. Not every model checks a clip's length before it runs, so cut the clip to the model's limit before you upload it.
Edits without a model
Some changes need no AI model. 8frame's Video Editor mode, separate from the canvas, has a timeline with video, text, caption and sound layers, a filter preset menu (Retro, Vintage, Film Noir, Cinematic and others) and a brightness slider per clip; it has no curves, colour wheels or LUTs. Its "Generate captions" turns a script you paste into a caption layer timed at an estimated 160 words per minute; it does not transcribe the clip's audio and does not export SRT. On the canvas, the Montage node joins 2 to 10 clips with transitions for 10 credits and trims each clip to a length you set (1 to 15 seconds, 5 by default), but it joins the picture only: the merged file has no sound, so the dialogue or music in your footage is dropped. For footage with sound, cut it on the Video Editor timeline or in your own editor.
What editing will not do on 8frame
None of these models upscales video, removes a video background or translates speech, and 8frame has no other tool for those: its upscalers (Topaz, Clarity, BRIA) work on images only. There is also no tool that re-syncs the mouth on footage you already have, and no dubbing or voice cloning; 8frame's lip-sync tools make a new clip from a still portrait and an audio track. For resolution, pick the output resolution when you edit, since nothing enlarges the video afterwards.
FAQ
Can AI edit an existing video? Yes. On 8frame, Grok Imagine, Wan 2.7 VideoEdit, Wan 2.2 video-to-video, Gemini Omni Flash, Happy Horse, Kling O1 Video and ID-V2V change a clip you upload. Seedance and Wan 3.0 Prime use a clip as a reference for a new one, and Kling 2.6 Pro Motion Control and Runway Act Two move a character with the motion in a clip. For cuts, text, captions and filters without a model, use the Video Editor mode.
What is the cheapest way to edit a video with AI? Grok Imagine edit: 41 credits for a 5-second clip at 480p, 54 at 720p, audio included. At 720p, Wan 2.2 video-to-video is 50 for 5 seconds.
How long can the input video be? Up to 10 seconds on Gemini Omni Flash, Wan 2.7 VideoEdit and Kling O1 Video, 15 on Grok Imagine and ID-V2V, 30 on Kling 2.6 Pro Motion Control and 60 on Happy Horse (billed on at most 15). On the canvas, Seedance takes one reference clip, up to 30 seconds on 2.5.
Can I extend a video with AI? Yes: Grok Imagine extend adds 2 to 10 seconds to a clip of up to 15 seconds, 54 credits for 5 seconds added to a 5-second clip. Gemini Omni Flash's edit mode on 8frame returns a clip as long as the input, so it does not extend.
Can AI fix the lip sync or dub an existing video on 8frame? No. 8frame has no tool that re-syncs the mouth on footage you already have, and no translation, dubbing or voice-cloning tool. Its lip-sync tools make a new clip from a still portrait and an audio track.
Sources
- Inputs, input limits and modes: the 8frame app, checked 2026-10-02
- What each canvas node sends, defaults, the Video Editor layers, caption timing and filter presets, and the Montage node: the 8frame app, checked 2026-10-02
- No lip re-sync, dubbing, voice cloning or video upscaling tool: the 8frame app, checked 2026-10-02
- Prices: the 8frame app, checked 2026-10-02
- Base Wan 3.0 with a clip through MCP: 8frame MCP tool definitions, read 2026-10-01
- Gemini Omni regional availability and the 10-second upload limit: Google, Gemini API docs, Gemini Omni, checked 2026-10-02
Pick the editor by the length of the clip you have.
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.