AI video models
for Dimly music videos
The video models behind Dimly music videos, the image models that draw their keyframes, and guides to models we cover but do not run: what each does, where it shines, and what it costs.
MiniMax H3
MiniMax
Reference-driven video with native stereo sound. Your track drives the picture, 768P up to 4K.
Seedance 2.5
ByteDance
First-and-last-frame moves up to thirty seconds, audio-driven at 480p/720p — your song muxed back intact.
Kling 4.0
Kuaishou
Kuaishou's long-take video model: up to thirty seconds in one pass, ten keyframes and Omni Reference with up to fifteen inputs.

MiniMax H3 Image
MiniMax
MiniMax's open-weights image model: up to sixteen reference images, native 4K and Chinese or English lettering, in about twenty seconds.

Seedream 5.5
ByteDance Seed
ByteDance Seed's image model: natural skin and light, edits that leave the rest of the frame alone, and dense small text that stays legible.
AI video models for music videos
Dimly directs a music video shot by shot: an image model draws each shot’s keyframe, and a video model animates it. This page lists the models behind that pipeline. Every model page has specs, sample prompts and samples.
Pick by job, not by leaderboard. The table compares what matters for a song: clip length, resolution, whether sound comes out, and whether the model is in the Dimly picker today.
Compared at a glance
| Model | Provider | Clip length | Resolution | Sound | In the Dimly picker |
|---|---|---|---|---|---|
| MiniMax H3 | MiniMax | 4 to 15 s | 768P to 4K | Native stereo | Yes |
| Seedance 2.5 | ByteDance | 4 to 30 s | 480p / 720p | Audio in, silent out; your song is added back | Yes |
| Kling 4.0 | Kuaishou | 3 to 30 s | 720p / 1080p | Native audio; your song is added back | Yes |
| MiniMax H3 Image | MiniMax | Still image | 1K / 2K / 4K | None (draws keyframes) | No; a studio image-model option |
| Seedream 5.5 | ByteDance Seed | Still image | 1K / 1.5K / 2K | None (draws keyframes) | No; a studio image-model option |
Which model for which job
When the song has to drive the picture
Reference-driven video with native stereo sound, from 768P pilots up to 4K. The one to open when the track, not a prompt, sets the pace.
See specs and promptsWhen you want a long move between two exact frames
First-and-last-frame video from four to thirty seconds, audio-driven at 480p or 720p. Your song is muxed back in intact.
See specs and promptsWhen you want one long take with a steady cast
Up to thirty seconds in a single pass from a first and last frame, at 720p or 1080p. It takes no song reference, so use it for shots nobody has to lip-sync.
See specs and promptsWhen the keyframes need one cast across many frames
An image model, not a video model: up to sixteen reference images and native 4K, fast and low-cost. Dimly can draw its sheets and keyframes with it.
See specs and promptsWhen a close-up or a fix has to look photographed
An image model with natural skin and light and edits that leave the rest of the frame alone. Dimly can draw its sheets and keyframes with it.
See specs and prompts