# Model Capability Matrix

Source: [Model Capability Matrix](<https://pr-760-preview.chatcut.dev/docs/model-capability-matrix>)

Documentation index: [llms.txt](<https://pr-760-preview.chatcut.dev/docs/llms.txt>)

Last updated: 2026-11-01

This matrix is the dense reference for the video models shown in ChatCut’s current selector. Recheck the visible **Video settings** before sending because provider and rollout limits can change.

| Model | Generate inputs | Reference limits | Edit | Extend | Output tiers |
| --- | --- | --- | --- | --- | --- |
| **Seedance 2.5** | Text, first frame, first + last frame, or references | Up to 30 images, 10 videos, and 10 audio files | One 4–30s source video plus an edit prompt | One source video plus an extension prompt | 480p, 720p, 1080p as offered |
| **Seedance 2.0** | Text, first frame, first + last frame, or references | Up to 9 images, 3 videos, and 3 audio files (9 files total in Video Gen); audio needs an image or video | No | No | 480p, 720p, 1080p, or native 4K |
| **Seedance 2.0 Fast** | Same input structure as Seedance 2.0 | Same limits as Seedance 2.0 | No | No | 480p, 720p |
| **Seedance 2.0 mini** | Same input structure as Seedance 2.0 | Same limits as Seedance 2.0 | No | No | 480p, 720p |
| **Gemini Omni** | Text, first frame, first + last frame, or up to 3 image references | Image inputs only | One project-library video up to about 10s | One project-library video up to about 10s plus an extension prompt | 360p, 720p, 1080p, or 4K (1080p and 4K are upscaled) |
| **MiniMax H3** | Text, first frame, last frame, first + last frame, or references | Up to 9 images, 3 videos, and 3 audio files; 12 files total | No | No | 768p or 2K |
| **MiniMax H3 Max** | Text, first frame, or first + last frame | No reference media; a last frame requires a first frame | No | No | 480p or 768p |
| **Kling 3.0** | Text, first frame, last frame, first + last frame, or reference images; frames and images combine | Up to 7 images in all, frames included; no reference video or audio | No | No | 720p or 1080p |
| **Kling 4.0** | Text, first frame, last frame, first + last frame, or references; frames and references combine | Up to 10 images in all, frames included, plus up to 5 videos (3–30s each, 30s total); no audio | No | No | 720p or 1080p |

Rules to remember:

-   A last frame requires a first frame, except on MiniMax H3, Kling 4.0, and Kling 3.0 without reference images.
-   Gemini Omni first-frame and multi-image-reference modes are mutually exclusive.
-   Seedance 2.5 Edit and Extend are distinct modes with their own source slots.
-   MiniMax H3 frame inputs and multimodal references are mutually exclusive.
-   Seedance 2.0 is the only model with native 4K output. 4K clips are 10-bit HEVC, take noticeably longer, and cost about twice as much as 1080p. Seedance 2.5 stops at 1080p, and Seedance 2.0 Fast and mini at 720p.
-   MiniMax H3 Max always requires a prompt and does not support reference media, editing, extension, or 2K output.
-   Kling 3.0 makes 3–15s clips in whole seconds at 16:9, 9:16, or 1:1; with frames and no reference images, the output follows the frame image. Its prompt is limited to 2,500 characters and has no negative prompt, so write exclusions into the prompt.
-   Kling 4.0 makes clips of 3–30s at 21:9, 16:9, 9:16, or 1:1; frames make the ratio follow the input, and text-only clips default to 16:9. Reference images must be JPG or PNG, at least 300 px per side, with a width:height ratio between 2:5 and 5:2. Reference videos must be MP4 or MOV, 700–4553 px per side, at 24–60 fps; they guide motion, style, or characters, and the output is a new 24 fps clip. The prompt can run up to 26,000 English characters or 11,000 in other languages. Refer to inputs in the prompt as @first\_frame, @last\_frame, @image\_1, @image\_2, @video\_1, and so on.
-   Kling 3.0 and Kling 4.0 always generate native audio and take no audio references; describe dialogue, sound effects, and music in the prompt.
-   Switching model or mode can remove incompatible attachments.
-   The selected duration, resolution, and supported video inputs affect the credit estimate.

For model selection, media inputs, and the exact controls to use, see [AI Video Generation](https://pr-760-preview.chatcut.dev/docs/video-generation-models).