Video Models — Sora 2, Veo 3.1, Kling
Video generation models compared Type /video in any chat to turn a prompt or an image into a video. Krater hosts 16 video model families from OpenAI, Google, Kuaishou, ByteDance, Black Forest Labs ...
Video generation models compared
Type /video in any chat to turn a prompt or an image into a video. Krater hosts 16 video model families from OpenAI, Google, Kuaishou, ByteDance, Black Forest Labs and more, all billed per second from the same credit pool. Attach an image to animate it (image-to-video), or use a model’s edit mode to rework an existing clip.
Video generation requires a Pro, Ultra or Max plan (or a seat on a team workspace).
All video models
| Model | Best for | Length | Max res | Audio | Credits/sec |
|---|---|---|---|---|---|
| Sora 2 | OpenAI’s cinematic flagship | 4-12s | auto | always | 25 |
| Veo 3.1 | Premium ads; Lite for drafts | 4-8s | 1080p | optional | 7.5-50 |
| Kling | Motion quality, video editing | 3-15s | auto | optional | 17.5-35 |
| Seedance | ByteDance flagship, many ratios | 4-15s | 1080p | optional | 60.5-75.9 |
| Flux 3 | Native audio, widest ratio set | 5-20s | 1080p | optional | 15-42.5 |
| LTX | Up to 4K, auto duration | 6-20s | 4K | optional | 22.5-30 |
| MiniMax H3 | 4K with reference images | 5-15s | 4K | - | 20 |
| Wan 2.7 | Multi-shot scenes, baked-in music | 2-15s | 1080p | music | 25 |
| Happy Horse | Audio-aware edits and references | 3-15s | 1080p | optional | 35 |
| Vidu Q3 | Fast image animation (Turbo) | 1-16s | 1080p | optional | 8.75-38.5 |
| PixVerse | Cheap flexible clips, transitions | 1-15s | 1080p | optional | 6.25-30 |
| Grok Imagine | Fast drafts, text/image/video input | 2-15s | 1080p | always | 12.5-20 |
| Luma Ray 3.2 | Cinematic grades and camera moves | 5-10s | 1080p | - | 25 |
| Gemini Omni Flash | Synchronized audio realism | 3-10s | 720p | always | 32 |
| Kandinsky 5 | Cheapest quick tests | 5s | 1024p | - | 4 |
| HeyGen | Talking-avatar presenter videos | up to 60s | - | voice | 8.5-25 |
Audio column: “optional” means an on/off toggle in the panel, “always” means the model generates it natively, “music” adds background music, “-” means no audio output.
Most families cover 16:9 and 9:16; PixVerse, Happy Horse, Grok, Flux 3 and MiniMax offer up to nine ratios including 21:9, 3:2 and 9:21.
Quality pills and versions
One picker row can hide several generations and speed tiers:
- Veo 3.1: Lite (Fast), Standard and Pro pills
- Flux 3: Draft (Fast) and full quality (Standard)
- Kling: O3 default; Kling 3.0, 3.0 Turbo, 2.6 and O1 under Versions
- Seedance: 2.0 with Fast/Standard/Pro pills; 2.5, 2.0 Mini and 1.5 under Versions
- LTX: 2.5 (Fast and Pro pills) plus 2.0
- PixVerse: C1 default; V6 under Versions
- Vidu Q3: Standard plus a Turbo variant (image-to-video only)
- Grok Imagine: Standard plus Video 1.5 (Pro)
- Happy Horse: 1.1 (Pro) plus a legacy version
- HeyGen: agent (Standard) plus v5 Digital Twin (Pro)
- Wan: 2.7 plus 2.6; MiniMax: H3 plus Hailuo 2.3
Special modes
- Reference (Seedance, Vidu Q3, MiniMax H3, Happy Horse, Grok Imagine): attach images to keep a subject, product or scene consistent across clips.
- Transform / Edit (Kling O3, Happy Horse, Grok Imagine): restyle or edit an existing video.
- Continue / Extend (Kling O3, PixVerse V6): add seconds to a finished clip.
- Animate Frames (Kling O3): animate between a start and end frame.
- Motion Control (Kling 2.6): copy motion from a reference video; costs 1.6x.
- Kling O3 accepts up to 9 images, 3 videos and 3 audio clips as @-mention references.
- HeyGen is a presenter model: pick an avatar, voice and style and it directs the scene itself. It reserves 510 credits (or 1,500 on v5) up front, then bills the actual seconds used, up to 60.
How video credits work
Cost = clip length x the model’s rate, then multipliers:
- Resolution: higher output costs more. LTX 4K is x3.33, Veo Lite 1080p is x1.67, Vidu Q3 jumps to 38.5 cr/s at 720p and above.
- Audio: on Veo, turning audio on adds roughly 1.5-2x; on most other families it is included in the rate.
- Quality pills and Pro rungs bill the higher rate (Veo Pro 50 cr/s, Seedance Pro ~76 cr/s).
- Reference images: MiniMax H3 gives 5 free then 20 credits each; Grok Video 1.5 charges 2.5 credits per reference.
- Failed generations release the reserved credits automatically.
Choosing a model
| Priority | Choose |
|---|---|
| Cheapest drafts | Kandinsky 5, PixVerse V6 |
| Social / vertical | Any model with 9:16 (nearly all) |
| Maximum quality | Sora 2, Veo 3.1 Pro, Seedance, Kling O3 |
| Longest clips | Flux 3 or LTX (20s), HeyGen (60s) |
| 4K output | LTX, MiniMax H3 |
| Synchronized sound | Flux 3, Gemini Omni Flash, Grok Imagine |
| Silent b-roll | Luma Ray 3.2, MiniMax H3, Kandinsky 5 |
| Talking-head presenter | HeyGen |
| Edit or extend a clip | Kling O3, Happy Horse, Grok Imagine |
Video credits come from the same pool as chat and images, and clips generate in the background so you can keep working. For prompt tips see /video; for the still-image side see Image Models.