AI Model Guide
NGMC integrates AI models across four families — image, video, audio, and chat. Each model has unique capabilities, strengths, and limitations. This guide helps you choose the right model and get the most out of it.
Image Models
Generate and edit still images from text prompts and reference images.
| Model | Speed | Max Resolution | Reference Images | Key Strength |
|---|---|---|---|---|
| GPT Image 2 | ~45s | 4K (3840x3840) | Yes | Prompt-faithful generation + edits, three quality tiers |
| GPT Image 2 Agent | 2–30min | 4K (3840x3840) | Up to 6 (shared) | AI-planned multi-frame sets — storyboards in one run |
| Nano Banana Pro | ~45s | 4K (6336x5504) | Up to 6 | Professional quality, multi-turn editing |
| Nano Banana 2 | ~15s | 4K (4096x4096) | Up to 14 | Most reference images, 0.5K draft mode |
| Nano Banana 2 Lite | ~2s | 1K | Up to 14 | Fastest + cheapest iteration |
| Grok Image Pro | ~15s | 2K | Up to 3 | Multi-image merge, adaptive + phone ratios, 4 outputs/run |
| MAI Image 2.5 | ~30s | ~1.3K | 1 | Photorealism + in-image text rendering |
| Seedream 5.0 Pro | ~30s | 2K | Up to 10 | Sketch/region-guided edits, Layer Workbench |
Video Models
Generate videos from text, images, or references — or edit existing footage.
| Model | Speed | Max Resolution | Duration | Audio | Key Strength |
|---|---|---|---|---|---|
| Veo 3.1 | ~2min | 4K | 4 / 6 / 8s | Native audio + dialogue | Highest quality, video extension |
| Veo 3.1 Fast | ~45s | 4K | 4 / 6 / 8s | Native audio + dialogue | Fast iteration with full resolution tiers |
| Seedance 2.0 | ~2min | 4K | 4–15s (continuous) | Optional native audio | Cinematic quality, mixed image + video + audio references |
| Seedance 2.0 Fast | ~45s | 720p | 4–15s (continuous) | Optional native audio | Fast Seedance iteration |
| Seedance 2.0 Mini | ~30s | 720p | 4–15s (continuous) | Optional native audio | Cheapest Seedance drafts |
| HappyHorse 1.1 | ~1min | 1080p | 3–15s (continuous) | No | 9 reference images, physical realism, seed support |
| Grok Video 1.5 | 1–3min | 1080p | 1–15s (continuous) | Native audio + lip-synced dialogue | 1-second minimum, adaptive aspect |
| Gemini Omni Flash | ~1min | 720p | Up to 10s | Native audio | Video editing + 30 reference images |
Audio Models
Music, sound effects, and voiceover.
| Model | Category | Speed | Duration | Key Strength |
|---|---|---|---|---|
| Lyria 3 Clip | Music | ~30s | 30s fixed | Fast musical iteration, image references |
| Lyria 3 Pro | Music | ~1min | Up to 2min | Full song structure, timestamped control |
| ElevenLabs Music V2 | Music | ~30s | 3s – 10min (exact) | Exact-length tracks, royalty-free |
| ElevenLabs Sound Effects | SFX | ~10s | Up to 30s | Foley/braams/ambience, seamless loop |
| Seed Audio 1.0 | Multimodal | ~30s | Up to 2min | Reference-audio style guidance, image-guided ambience |
| Gemini 3.1 Flash TTS | Voiceover | ~10s | Up to 3min | 30 voices, 2-speaker dialogue, audio tags |
| Qwen3 TTS Flash | Voiceover | ~10s | Up to 3min | 36 voices incl. 8 Chinese dialects |
| Qwen3 TTS Instruct | Voiceover | ~10s | Up to 3min | Natural-language delivery instructions |
Chat Models
AI Chat nodes for writing, planning, and analyzing — with vision and long context.
| Model | Context | Reasoning Effort | Key Strength |
|---|---|---|---|
| Claude Opus 4.6 | 1M tokens | None – High | Screenwriting workhorse, fast + deep |
| Claude Fable 5 | 1M tokens | Low – X-High (always thinking) | Hardest reasoning + long-horizon writing (Max plan) |
| GPT 5.5 | 1M tokens | None – X-High | Widest effort range, structured planning |
All three see connected images (up to 10), read connected documents (up to 100), and stream replies into the canvas conversation.
Referencing Images in Prompts
You can use @-mentions in your prompt to tell the model exactly how to use each reference image.
How it works
Type @ in the prompt textarea to open the reference picker. Select an image, and a mention like @[Image 1] is inserted at that position. When generated, the model sees the actual image inline at that exact point in your text, giving it precise context.
Example:
Add one more ragdoll cat based on @[Image 2], and replace the background
with the scenery from @[Image 1].
The model receives Image 2 right after "based on" and Image 1 right after "scenery from", so it understands which image to use for which purpose.
Without @-mentions
If you don't use @, all reference images are appended after your prompt text. The model still sees them but has to infer how to use each one. For simple cases (single reference, style transfer) this works fine. Use @ when you need the model to distinguish between multiple images.
- Iterating on ideas? Nano Banana 2 Lite (~2s) for images; Seedance Fast/Mini, Veo 3.1 Fast, or Gemini Omni Flash for video.
- Final output quality? Nano Banana Pro or Seedream 5.0 Pro (images); Veo 3.1 or Seedance 2.0 at 4K (video).
- Need many reference images? Nano Banana 2 / Lite take 14; Seedream 5.0 Pro takes 10; Omni Flash takes 30 on the video side.
- Mixed reference inputs (image + video + audio)? Seedance 2.0 accepts all three modalities in the same run.
- Editing existing footage? Gemini Omni Flash is the only video-editing model.
- Need dialogue in video? Veo 3.1 / Fast generate speech from quoted text in your prompt.
- Whole storyboard in one run? GPT Image 2 Agent plans and renders up to 10 coherent frames.
- Voiceover? Gemini TTS (dialogue, tags), Qwen3 TTS (Chinese dialects), or Seed Audio 1.0 (reference-audio style guidance).
- Music to exact length? ElevenLabs Music V2 renders 3s–10min on the second.
- Precise spatial edits? Seedream 5.0 Pro's sketch + coordinate-marker editing.