Model catalogue
Models are not tiers of the same thing — they differ in what they accept, how long they can run, and what they cost. You pick one per node, and you can change it between takes without losing what you already made.
The tables below are generated from the live model registry, so they match what the app offers today. Credit figures are a single comparable sample per model (the settings named in each table), not the only price — longer clips, higher resolutions and reference inputs all move the number. The estimate shown on the Generate button is always the one that will be charged.
Video
| Model | Resolutions | Length | Audio | From (credits) |
|---|---|---|---|---|
| Seedance 2.0 | 480p, 720p, 1080p, 4k | 4–15s | Yes | 60 |
| Seedance 2.5 | 480p, 720p, 1080p | 4–30s | Yes | 88 |
| Seedance 2.0 Fast | 480p, 720p | 4–15s | Yes | 48 |
| Seedance 2.0 Mini | 480p, 720p | 4–15s | Yes | 30 |
| Kling 3.0 | 720p, 1080p, 4k | 3–15s | Yes | 61 |
| Kling 3.0 Turbo | 720p, 1080p | 3–15s | No | 54 |
| MiniMax H3 | 720p, 2k | 5–15s | No | 80 |
| FLUX.3 Video | 720p, 1080p | 5–20s | Yes | 153 |
| HappyHorse 1.1 | 720p, 1080p | 3–15s | No | 59 |
| Grok Imagine | 480p, 720p | 1–15s | No | 11 |
| Grok Imagine 1.5 | 480p, 720p | 1–15s | No | 17 |
| Wan 3.0 | 480p, 720p, 1080p | 2–30s | Yes | 20 |
| Wan 3.0 Prime | 480p, 720p, 1080p | 2–30s | Yes | 31 |
| Hailuo 2.3 | 768p, 1080p | 6–10s | No | 34 |
| Hailuo 2.3 Fast | 768p, 1080p | 6–10s | No | 23 |
| Pika 2.5 | 720p, 1080p | 5–10s | No | 40 |
| Veo 3.1 Lite | 720p, 1080p | 4–8s | Yes | 40 |
| Gemini Omni Flash | 720p | 3–10s | No | 61 |
| Veo 3.1 | 720p, 1080p, 4k | 4–8s | Yes | 320 |
| Veo 3.1 Fast | 720p, 1080p, 4k | 4–8s | Yes | 80 |
Video models differ in three ways that matter in practice:
- Length. Some take 4 seconds, some take 15. A cut longer than the model's window has to be two generations.
- Audio. Models marked Yes can generate sound with the picture. The rest are silent, and you add sound in Audio AI.
- References. Character sheets and location plates connected on the Canvas travel into the generation. How many fit depends on the model.
Image
| Model | Prompt limit | 1K · medium · 16:9 (credits) |
|---|---|---|
| GPT Image 2 | not declared | 5 |
| Nano Banana 2 | not declared | 14 |
| Seedream 4.5 | not declared | 8 |
| Seedream 5.0 Pro | not declared | 9 |
| Seedream 5.0 Lite | not declared | 7 |
| Grok Imagine | not declared | 5 |
| Grok Imagine Quality | not declared | 11 |
| Recraft 4.1 | 10,000 chars | 9 |
| Nano Banana Pro | not declared | 28 |
| GPT Image 1 | not declared | 14 |
| Imported | not declared | 1 |
Image models are used for character sheets, location plates and style frames — the reference material that keeps later shots consistent.
Audio
| Model | Kind | Length | 10s (credits) |
|---|---|---|---|
| Seed Audio 1.0 | Sound effects | model decides | 5 |
| Kling Audio (text) | Sound effects | 3–10s | 6 |
| ElevenLabs SFX v2 | Sound effects | 1–30s | 5 |
| Sonilo SFX | Sound effects | 1–180s | 3 |
| Pika SFX | Sound effects | 1–20s | 1 |
| Lyria 2 SFX | Sound effects | model decides | 10 |
| Kling Audio (V2A) | Video to audio | model decides | 6 |
| Sonilo SFX (video) | Video to audio | model decides | 15 |
| Seed Audio 1.0 | Voice | model decides | 5 |
| Seed Audio Score | Music | model decides | 5 |
| Lyria 3 | Music | model decides | 7 |
| Lyria 3 Pro | Music | model decides | 13 |
| MiniMax Music 3.0 | Music | model decides | 18 |
| Pika Music | Music | 10–360s | 1 |
| ElevenLabs Music | Music | 3–600s | 6 |
| Pika Speech | Voice | model decides | 1 |
| ElevenLabs v2 | Voice | model decides | 3 |
| ElevenLabs Turbo | Voice | model decides | 2 |
| MiniMax Speech HD | Voice | model decides | 3 |
| MiniMax Speech Turbo | Voice | model decides | 2 |
| ElevenLabs Voice Changer | Voice | model decides | 5 |
| ElevenLabs Voice Isolation | Voice | model decides | 5 |
| ElevenLabs Scribe | Voice | model decides | 1 |
Audio generation is a separate step applied to a range on the timeline, not a property of the video node. See Audio AI.
Deprecated models
Models that have been retired stay readable on old takes — a take you generated last month keeps its model, prompt and settings — but they no longer appear in the picker. Nothing you already made is affected.
Last updated 2026-08-25. Prices, model specs and credit costs on this page are read from the live catalogue at build time.