AI video model guide

Which AI video model should you use?

Every video model aicut sells, compared on rating, sound, German speech and price.

For most videos with sound, use WAN 3.0: it has the highest with-sound arena rating of the models aicut sells. For a product ad use Seedance 2.5, for German speech Grok Imagine, and for silent b-roll MiniMax H3 Max. The cheapest clip is Veo 3.1 Lite. Every model below runs in one aicut account, with the price shown before you generate.

20 models compared
Open the AI Video Generator

Free account. The token price is shown before anything is generated.

Which model for which job

The short answer for the common jobs. Each pick links to its model card below or to the tool that does the job.

Every AI video model compared

Sorted by the Artificial Analysis text-to-video arena Elo on the board with sound, verified November 19, 2026. Kling 3 is the arena's anchor at 1000. MiniMax H3 Max is rated on that board, but aicut sells it without sound. "Not rated" means the arena has no rating yet; those models are placed by our own estimate. The price is one clip at 8 seconds and 720p, or the closest setting the model accepts.

  • WAN 3.0

    Inputs
    text, start frame, end frame, 10 reference images, 5 reference videos, 1 reference audio
    Sound
    On or off
    Speech languages
    No published list
    Length
    2-30s
    Max resolution
    1080p
    Price
    3.2 tokens ($1.92) for 8s at 720p
    Arena Elo
    1174
  • WAN 3.0 Prime

    Inputs
    text, start frame, end frame, 10 reference images, 5 reference videos, 1 reference audio
    Sound
    On or off
    Speech languages
    No published list
    Length
    2-30s
    Max resolution
    1080p
    Price
    4.8 tokens ($2.88) for 8s at 720p
    Arena Elo
    Not rated
  • Seedance 2.5

    Inputs
    text, start frame, end frame, 30 reference images, 10 reference videos, 3 reference audios
    Sound
    On or off
    Speech languages
    Chinese, English
    Length
    4-30s
    Max resolution
    1080p
    Price
    7.68 tokens ($4.61) for 8s at 720p
    Arena Elo
    1156
  • FLUX 3 Max

    Inputs
    text, start frame, end frame, 10 keyframes
    Sound
    On or off
    Speech languages
    No published list
    Length
    5-20s
    Max resolution
    1080p
    Price
    5.6 tokens ($3.36) for 8s at 720p
    Arena Elo
    1152
  • Gemini Omni Flash 1.2

    Inputs
    text, start frame, end frame, 6 reference images
    Sound
    Always on
    Speech languages
    English
    Length
    3-15s
    Max resolution
    4K
    Price
    2 tokens ($1.20) for 8s at 720p
    Arena Elo
    1151
  • MiniMax H3 Max

    Inputs
    text, start frame, end frame
    Sound
    No sound
    Speech languages
    None
    Length
    5-15s
    Max resolution
    1080p
    Price
    2.4 tokens ($1.44) for 8s at 768p
    Arena Elo
    1143
  • Kling 4

    Inputs
    text, start frame, end frame, 10 reference images, 5 reference videos, 1 reference audio
    Sound
    Always on
    Speech languages
    9 languages
    Length
    3-30s
    Max resolution
    1080p
    Price
    4.8 tokens ($2.88) for 8s at 720p
    Arena Elo
    Not rated
  • Seedance 2.0

    Inputs
    text, start frame, end frame, 9 reference images, 3 reference videos
    Sound
    On or off
    Speech languages
    Chinese, English
    Length
    4-15s
    Max resolution
    4K
    Price
    6 tokens ($3.60) for 8s at 720p
    Arena Elo
    1133
  • Seedance 2.5 Flash

    Inputs
    text, start frame, end frame, 10 reference images, 5 reference videos, 3 reference audios
    Sound
    On or off
    Speech languages
    Chinese, English
    Length
    4-30s
    Max resolution
    720p
    Price
    5.12 tokens ($3.07) for 8s at 720p
    Arena Elo
    Not rated
  • MiniMax H3 Max Turbo

    Inputs
    text, start frame, end frame
    Sound
    No sound
    Speech languages
    None
    Length
    5-15s
    Max resolution
    1080p
    Price
    1.6 tokens ($0.96) for 8s at 768p
    Arena Elo
    Not rated
  • Kling 4 Flash

    Inputs
    text, start frame, 10 reference images, 5 reference videos, 1 reference audio
    Sound
    Always on
    Speech languages
    9 languages
    Length
    3-20s
    Max resolution
    720p
    Price
    3.2 tokens ($1.92) for 8s at 720p
    Arena Elo
    Not rated
  • Seedance 2.0 Fast

    Inputs
    text, start frame, end frame, 9 reference images, 3 reference videos
    Sound
    On or off
    Speech languages
    Chinese, English
    Length
    4-15s
    Max resolution
    720p
    Price
    4.8 tokens ($2.88) for 8s at 720p
    Arena Elo
    Not rated
  • Seedance 2.0 Mini

    Inputs
    text, start frame, end frame, 9 reference images, 3 reference videos
    Sound
    On or off
    Speech languages
    Chinese, English
    Length
    4-15s
    Max resolution
    720p
    Price
    3.2 tokens ($1.92) for 8s at 720p
    Arena Elo
    Not rated
  • Grok Imagine

    Inputs
    text, start frame, 7 reference images
    Sound
    Always on
    Speech languages
    No published list
    Length
    1-15s
    Max resolution
    1080p
    Price
    2.4 tokens ($1.44) for 8s at 720p
    Arena Elo
    1051
  • Kling 3

    Inputs
    text, start frame, end frame
    Sound
    On or off
    Speech languages
    Chinese, English, Japanese, Korean, Spanish
    Length
    3-15s
    Max resolution
    1080p
    Price
    6.4 tokens ($3.84) for 8s at 720p
    Arena Elo
    1000 (anchor)
  • Kling 3 Turbo

    Inputs
    text, start frame
    Sound
    No sound
    Speech languages
    None
    Length
    3-15s
    Max resolution
    1080p
    Price
    3.2 tokens ($1.92) for 8s at 720p
    Arena Elo
    Not rated
  • Veo 3.1

    Inputs
    text, start frame, end frame
    Sound
    Always on
    Speech languages
    English
    Length
    4-8s
    Max resolution
    1080p
    Price
    5 tokens ($3.00) for 8s at 720p
    Arena Elo
    963
  • Veo 3.1 Fast

    Inputs
    text, start frame, end frame
    Sound
    Always on
    Speech languages
    English
    Length
    4-8s
    Max resolution
    1080p
    Price
    2 tokens ($1.20) for 8s at 720p
    Arena Elo
    954
  • Veo 3.1 Lite

    Inputs
    text, start frame, end frame
    Sound
    Always on
    Speech languages
    English
    Length
    4-8s
    Max resolution
    1080p
    Price
    1 token ($0.60) for 8s at 720p
    Arena Elo
    949
  • Kling 2.5 Turbo

    Inputs
    text, start frame, end frame
    Sound
    No sound
    Speech languages
    None
    Length
    5-10s
    Max resolution
    1080p
    Price
    3 tokens ($1.80) for 10s at 1080p
    Arena Elo
    Not rated

What each model is good and bad at

From the makers' own documentation and aicut's rulings, in the order of the table.

WAN 3.0

On or off
Arena Elo 1174

3.2 tokens ($1.92) for 8s at 720p

Speech

No published list; the maker shows only English and Chinese examples.

Good at

  • •A quoted line becomes lip-synced speech.
  • •A reference audio drives a dance or a lip-sync to its beat.

Watch out for

  • •Chinese models usually do not speak German.
  • •Speech can come out as burned-in text: describe the speaker and end with 'no on-screen text, no subtitles'.
  • •First and last frames cannot be combined with reference media.

WAN 3.0 Prime

On or off
Not rated

4.8 tokens ($2.88) for 8s at 720p

Speech

No published list; the maker shows only English and Chinese examples.

Good at

  • •WAN 3.0's faster tier: the same quality, made faster.
  • •A quoted line becomes lip-synced speech.
  • •A reference audio drives a dance or a lip-sync to its beat.

Watch out for

  • •Chinese models usually do not speak German.
  • •Speech can come out as burned-in text: describe the speaker and end with 'no on-screen text, no subtitles'.
  • •First and last frames cannot be combined with reference media.

Seedance 2.5

On or off
Arena Elo 1156

7.68 tokens ($4.61) for 8s at 720p

Speech

Speaks mainly Chinese and English.

Good at

  • •Up to 30 seconds.
  • •Takes up to 30 reference images, 10 reference videos and 3 reference audios.

Watch out for

  • •German speech comes out garbled.
  • •Refuses reference photos of real people (BytePlus privacy rule).

FLUX 3 Max

On or off
Arena Elo 1152

5.6 tokens ($3.36) for 8s at 720p

Speech

Many languages and accents, switching inside one clip; no published list.

Good at

  • •Pins up to 10 pictures to moments in the clip (keyframes).
  • •Switches language inside one clip.

Watch out for

  • •Wide or uneven gaps between keyframes give unpredictable motion.
  • •A spoken line can show up as on-screen text: give it a visible speaker or a voiceover cue.
  • •'Quiet room tone' in a prompt can turn into static.

Gemini Omni Flash 1.2

Always on
Arena Elo 1151

2 tokens ($1.20) for 8s at 720p

Speech

English is fully supported; Google has not evaluated other languages.

Good at

  • •Up to 6 subject reference images: at most 3 per character, 2 characters.
  • •Sound in every clip.
  • •Renders up to 4K.

Watch out for

  • •Cuts between shots by default: ask for a single continuous shot.
  • •Blocks photos of minors and of recognisable people in the EEA, Switzerland and the UK.
  • •Takes no audio input.

MiniMax H3 Max

No sound
Arena Elo 1143

2.4 tokens ($1.44) for 8s at 768p

Watch out for

  • •Makes no sound at all.
  • •Visual detail can improve (MiniMax's own note).
  • •No 720p: 480p, 768p or 1080p.

Kling 4

Always on
Not rated

4.8 tokens ($2.88) for 8s at 720p

Speech

Chinese, English, Japanese, Korean, Spanish, Portuguese, German, French, Hindi. Plus Cantonese and Sichuanese, and US, UK and Indian English accents.

Good at

  • •Good German speech.
  • •Up to 30 seconds, split into shots by the model itself.
  • •Takes up to 10 reference images, 5 reference videos and a reference audio.
  • •Stereo sound in every clip.

Watch out for

  • •Splits a clip into shots by default: give timestamped events to control the cuts.
  • •A copyright check can refuse a generation.

Seedance 2.0

On or off
Arena Elo 1133

6 tokens ($3.60) for 8s at 720p

Speech

Speaks mainly Chinese and English.

Good at

  • •Renders up to 4K.
  • •Takes up to 9 reference images and 3 reference videos.

Watch out for

  • •German speech comes out garbled.
  • •Refuses reference photos of real people (BytePlus privacy rule).
  • •4K comes as 10-bit H.265, which some players cannot play.

Seedance 2.5 Flash

On or off
Not rated

5.12 tokens ($3.07) for 8s at 720p

Speech

Speaks mainly Chinese and English.

Good at

  • •About twice as fast as Seedance 2.5 (the maker's figure).
  • •Up to 30 seconds.

Watch out for

  • •German speech comes out garbled.
  • •Refuses reference photos of real people (BytePlus privacy rule).
  • •One step under Seedance 2.5.
  • •480p or 720p only.

MiniMax H3 Max Turbo

No sound
Not rated

1.6 tokens ($0.96) for 8s at 768p

Watch out for

  • •Makes no sound at all.
  • •One step under MiniMax H3 Max.
  • •No 720p.

Kling 4 Flash

Always on
Not rated

3.2 tokens ($1.92) for 8s at 720p

Speech

Chinese, English, Japanese, Korean, Spanish, Portuguese, German, French, Hindi. Plus Cantonese and Sichuanese, and US, UK and Indian English accents.

Good at

  • •Good German speech, the same languages as Kling 4.
  • •Takes up to 10 reference images, 5 reference videos and a reference audio.

Watch out for

  • •One step under Kling 4.
  • •720p only, up to 20 seconds.

Seedance 2.0 Fast

On or off
Not rated

4.8 tokens ($2.88) for 8s at 720p

Speech

Speaks mainly Chinese and English.

Good at

  • •Takes up to 9 reference images and 3 reference videos.

Watch out for

  • •German speech comes out garbled.
  • •Refuses reference photos of real people (BytePlus privacy rule).
  • •480p or 720p only.

Seedance 2.0 Mini

On or off
Not rated

3.2 tokens ($1.92) for 8s at 720p

Speech

Speaks mainly Chinese and English.

Good at

  • •Takes up to 9 reference images and 3 reference videos.

Watch out for

  • •German speech comes out garbled.
  • •Refuses reference photos of real people (BytePlus privacy rule).
  • •Two steps under Seedance 2.0.
  • •480p or 720p only.

Grok Imagine

Always on
Arena Elo 1051

2.4 tokens ($1.44) for 8s at 720p

Speech

No published language list.

Good at

  • •Good German speech.
  • •Sound in every clip, music included.
  • •Takes up to 7 reference images.

Watch out for

  • •Reference images guide the subject but do not lock the first frame: use a start frame for an exact look.
  • •More than one reference image caps the clip at 720p.
  • •Sound cannot be turned off.

Kling 3

On or off
Arena Elo 1000

6.4 tokens ($3.84) for 8s at 720p

Speech

Chinese, English, Japanese, Korean, Spanish. German is not on the list.

Good at

  • •Several shots in one clip.

Watch out for

  • •German is not on its speech list.
  • •Can mix up speakers: keep each name, line and delivery together.

Kling 3 Turbo

No sound
Not rated

3.2 tokens ($1.92) for 8s at 720p

Watch out for

  • •Makes no sound at all.
  • •One shot per clip.
  • •One step under Kling 3.

Veo 3.1

Always on
Arena Elo 963

5 tokens ($3.00) for 8s at 720p

Speech

English is fully supported; Google has not evaluated other languages.

Good at

  • •Makes sound in every clip.

Watch out for

  • •Weak on every angle next to the other models here: realism, dialogue and prompt following.
  • •8 seconds at most.
  • •Takes no reference images.

Veo 3.1 Fast

Always on
Arena Elo 954

2 tokens ($1.20) for 8s at 720p

Speech

English is fully supported; Google has not evaluated other languages.

Good at

  • •Makes sound in every clip.
  • •Takes up to 3 reference images through the MCP and the CLI.

Watch out for

  • •Weak on every angle next to the other models here: realism, dialogue and prompt following.
  • •8 seconds at most.

Veo 3.1 Lite

Always on
Arena Elo 949

1 token ($0.60) for 8s at 720p

Speech

English is fully supported; Google has not evaluated other languages.

Good at

  • •Makes sound in every clip.

Watch out for

  • •Weak on every angle next to the other models here: realism, dialogue and prompt following.
  • •8 seconds at most.
  • •Takes no reference images.

Kling 2.5 Turbo

No sound
Not rated

3 tokens ($1.80) for 10s at 1080p

Watch out for

  • •Makes no sound at all.
  • •5 or 10 seconds only.
  • •One step under Kling 3 Turbo.

Tools for the jobs a prompt cannot do

Editing a video you have, copying motion and lip-syncing to your own audio each have their own tool.

Edit a video I already have

Kling O3

Good at

  • •Edits a video you already have, keeping its shot, motion, length and audio by default.
  • •Takes up to 4 reference images or elements.

Watch out for

  • •German is not on its speech list.

Copy the motion of a video onto my character

Kling 3.0 Motion Control

Good at

  • •Copies the motion of a clip up to 30 seconds long onto a character.

Watch out for

  • •Refuses a picture or a clip without a complete upper body ('No complete upper body detected') or with more than one person.

Kling Motion Control

Good at

  • •The cheaper way to copy a video's motion onto a character.

Watch out for

  • •Refuses a picture or a clip without a complete upper body ('No complete upper body detected') or with more than one person.

Lip-sync a photo to my own audio

Kling AI Avatar 2.0

Good at

  • •Lip-syncs a photo to your own audio.

Which AI video model speaks German?

Kling 4, Kling 4 Flash, and Grok Imagine speak German well. Seedance models try, but their German comes out garbled. Veo and Gemini Omni officially support English only, and Chinese models such as WAN usually do not speak German.

Speaks German well
Kling 4, Kling 4 Flash, Grok Imagine
German comes out garbled
Seedance 2.5, Seedance 2.0, Seedance 2.5 Flash, Seedance 2.0 Fast, Seedance 2.0 Mini
German is not on the maker's list
Kling 3
Does not speak German
WAN 3.0, WAN 3.0 Prime, Gemini Omni Flash 1.2, Veo 3.1, Veo 3.1 Fast, Veo 3.1 Lite
Not tested in German
FLUX 3 Max
Makes no speech
MiniMax H3 Max, MiniMax H3 Max Turbo, Kling 3 Turbo, Kling 2.5 Turbo

Why did my AI video come out wrong?

The common problems, why they happen and what to change. Notes name the models a fix applies to.

Speech in the wrong language, garbled German or a bad accent

Why

  • •The model does not speak the language.
  • •Chinese models usually do not speak German.

What to change

  • •For German speech pick Grok Imagine or Kling 4, never a Seedance model.

No sound, no voice, a mute video

Why

  • •The model makes no sound.
  • •The sound switch was off.
  • •No visible speaker and no quoted line, so the model treats the speech as ambience.

What to change

  • •Pick a model with sound.
  • •Write dialogue in quotes attached to a visible speaker; for a narrator off screen add a voiceover cue.
  • MiniMax H3 Max, MiniMax H3 Max Turbo, Kling 3 Turbo, and Kling 2.5 Turbo: Makes no sound at all: move to a model with sound.
  • Seedance 2.0, Seedance 2.0 Fast, Seedance 2.0 Mini, Seedance 2.5, Seedance 2.5 Flash, WAN 3.0, WAN 3.0 Prime, Kling 3, and FLUX 3 Max: Sound is a switch: check it was on.

Unwanted background music

What to change

  • •Write 'no music'.
  • Kling 4 and Kling 4 Flash: The only negatives Kling recommends are 'no subtitles, no background music'.
  • FLUX 3 Max: 'No music' removes music only; avoid 'quiet room tone', which can turn into static.

Burned-in subtitles, misspelled or gibberish text

Why

  • •Speech gets rendered as captions, or the model invents signs and labels.

What to change

  • •Add 'no on-screen text, no subtitles' and tie each line to a visible speaker.
  • •For text you do want, quote the exact words and say where they sit.
  • •Otherwise render without text and add the captions in the aicut editor.

It ignored my prompt, my reference or my product

What to change

  • •Give every reference one explicit role: 'Image 1 is the product, keep its logo'.
  • •For an exact look, use the picture as the start frame instead of a reference.
  • Seedance 2.5, Kling 4, WAN 3.0, and Gemini Omni Flash 1.2: Takes reference images: use it when a product must appear.
  • Grok Imagine: Reference images guide the subject without locking the first frame.

The character or the face changes

Why

  • •A single clip does not lock a face across clips.
  • •Pronouns instead of names in a prompt with several people.

What to change

  • •For several clips with the same people, use AI Video Story with a saved cast.
  • •Use the same label for a person every time.

Unwanted cuts, a split screen, a line down the middle

Why

  • •The model cuts on its own, or the prompt describes more than one shot.

What to change

  • •Describe one shot per clip with no 'cut to'; write 'single continuous shot, no scene cuts' and name a static camera.
  • Gemini Omni Flash 1.2: Cuts between shots by default.
  • Kling 4 and Kling 4 Flash: Splits shots by default: give timestamped events for control.

The wrong person speaks, lines merge, the listener's mouth moves

Why

  • •Lines that are not tied to one visible speaker.

What to change

  • •Attribute each line to one visible character.
  • Kling 3, Kling O3, Kling 4, and Kling 4 Flash: Keep name, line and delivery together, put the action before the line, and write 'Immediately' at each switch of speaker.

Blocked or failed by the content filter

Why

  • •The maker's content filter refused the prompt or a picture.

What to change

  • •Use a generated character or an AI Video Story cast portrait instead of a photo of a real person.
  • Seedance 2.0, Seedance 2.0 Fast, Seedance 2.0 Mini, Seedance 2.5, and Seedance 2.5 Flash: Refuses reference photos of real people (BytePlus privacy rule) and checks the output for copyright.
  • Gemini Omni Flash 1.2: Blocks photos of minors and of recognisable people in the EEA, Switzerland and the UK.
  • Kling 4: A copyright check can refuse the generation.

Too short, the wrong length

Why

  • •The length was not set, or the model's maximum is shorter than asked.

What to change

  • •Set the length before generating, within the model's published range.
  • Veo 3.1, Veo 3.1 Fast, and Veo 3.1 Lite: 8 seconds at most.
  • Seedance 2.5, Seedance 2.5 Flash, Kling 4, WAN 3.0, and WAN 3.0 Prime: Up to 30 seconds.

Motion control errors

Why

  • •The picture or the clip lacks a visible head, body and limbs.
  • •More than one person in the clip.

What to change

  • •One person per clip.
  • Kling Motion Control and Kling 3.0 Motion Control: Copies motion only: a photo that should lip-sync to audio belongs in Kling AI Avatar 2.0.

Looks like AI, low detail

Why

  • •Usually the model.

What to change

  • •Move to a stronger model: Veo ranks lowest of the models aicut sells.
  • Seedance 2.0: 4K comes as 10-bit H.265, which some players cannot play.

AI video model FAQ

What is the best AI video model in 2026?

On the Artificial Analysis text-to-video arena with sound, WAN 3.0 has the highest rating of the models aicut sells (verified November 19, 2026). The best model still depends on the job: Seedance 2.5 for product ads, Grok Imagine for German speech and MiniMax H3 Max for silent b-roll.

Which AI video model speaks German?

Kling 4, Kling 4 Flash, and Grok Imagine speak German well. Seedance models try, but their German comes out garbled. Veo and Gemini Omni officially support English only, and Chinese models such as WAN usually do not speak German.

Which AI video models make sound?

Always with sound: Gemini Omni Flash 1.2, Kling 4, Kling 4 Flash, Grok Imagine, Veo 3.1, Veo 3.1 Fast, and Veo 3.1 Lite. Sound you can switch on or off: WAN 3.0, WAN 3.0 Prime, Seedance 2.5, FLUX 3 Max, Seedance 2.0, Seedance 2.5 Flash, Seedance 2.0 Fast, Seedance 2.0 Mini, and Kling 3. No sound at all: MiniMax H3 Max, MiniMax H3 Max Turbo, Kling 3 Turbo, and Kling 2.5 Turbo.

What is the cheapest AI video model?

Veo 3.1 Lite, at 1 token ($0.60) for 8s at 720p. Prices on this page are at the pay-as-you-go token rate.

Why does my AI video have no sound?

Either the model makes no sound, its sound switch was off, or the prompt had no visible speaker and no quoted line, so the model treated the speech as background. Pick a model with sound and write each line in quotes next to the person who says it.

How are the models rated?

By the Artificial Analysis video arena, where people compare two clips without knowing the models. This page uses the text-to-video board with sound, verified November 19, 2026. Kling 3 is the arena's anchor at 1000. Models the arena has not rated are shown as "not rated" and placed by our own estimate.

Can I use all of these models with one account?

Yes. Every model on this page runs in aicut's AI Video Generator on one token balance, and the price of a clip is shown before you generate it.

Pick a model and generate

Every model on this page is in the AI Video Generator, on one token balance.

Open the AI Video Generator

Diese Seite auf Deutsch