Models
Every image, video, and sound model on the canvas, and which one to reach for.
Chromatic puts the image and video models worth using on one plan. You don't have to learn how each one wants to be prompted. Light already knows, so your team picks the look and Light writes the prompt.
Every model below is in the model menu on the canvas. A few have a longer name in the menu, like ByteDance SeedDream 5 Lite. Linked names have their own page with examples.
You'll find upscale and background removal on the image menu. See Image tools.
Image models
| Model | Use it when |
|---|---|
| Nano Banana | You want a still, or an edit with reference images. |
| Nano Banana Pro | You want 1K, 2K, or 4K, and optional web search. |
| Nano Banana 2 | You want the same, plus 0.5K and a 21:9 frame. |
| Nano Banana 2.1 | You want up to 14 reference images and a thinking level. |
| Midjourney V7 | You want a four-image grid. It takes up to 10 references and prompts up to 1024 characters. |
| Midjourney V8.1 | You want the newer four-image grid. It has the same reference cap. |
| GPT Image 2 | You want an OpenAI still with a quality setting of low, medium, or high. |
| GPT Image 2.5 Flare | You want quality up to Max, a background of auto, opaque, or transparent, and up to 16 references. |
| GPT Image 2.5 Sunburst | You want the same 2.5 controls on the Sunburst endpoint. |
| Grok Imagine Edit | You want an edit pass. |
| Grok Imagine Quality Edit | You want the higher-quality edit pass. |
| FLUX | You want a still with no reference images. |
| Flux 3 Image | You want a Flux still or edit at 1K, 2K, or 4K. |
| Krea 3 | You want a still or an edit with up to 8 references. Quality and Fast are variants of it. |
| SeedDream, ByteDance SeedDream 4.5, 5 Lite, 5 Pro | You want a ByteDance still. Each version has its own resolution options. |
| Recraft V4.1 and the Pro, Utility, and Vector rows | You want a Recraft still. The rows named Vector give you vector outputs. |
Video models
| Model | Use it when |
|---|---|
| Seedance 2.0 | You want reference video from images, video, and audio, plus a first and last frame. Clips run 4 to 15 seconds. Fast, Mini, and Enterprise are versions of it. |
| Seedance 2.5 | You want longer reference video. Clips run 4 to 30 seconds, and it takes up to 10 reference audio clips. |
| Kling 3.0 | You want Kling V3 Pro, with motion control. Kling V2.6 Pro, V2.5 Pro, and O3 Pro have their own rows. O3 Pro edits a clip you already have. |
| Veo 3.1 | You want a 4, 6, or 8 second clip at 16:9 or 9:16, with audio. Veo 3.1 Lite is the lighter row. On full Veo, 1080p and 4K fix the clip at 8 seconds. |
| Gemini Omni Flash | You want reference video, image to video, or a video edit. |
| Grok Imagine Video 1.5 | You want a fast clip with audio. Grok Imagine Video is the earlier version. |
| Flux 3 | You want Flux video. Flux 3 Max is the higher-quality version. |
| MiniMax H3 | You want MiniMax video. H3 Max is the higher row. MiniMax Hailuo 2.3 [Standard] and Hailuo 2.3 Fast [Pro] are the older rows in the same menu. |
| Wan 2.7, Wan 3.0, Wan 3.0 Prime | You want Wan video. |
| Happy Horse 1.0 | You want reference video, like Seedance. |
| Chromatic Lipsync | You want a face to speak a voice track. |
| Kling V2 Avatar Pro | You want an avatar or a talking performance. |
Sound
Sound has no menu of its own. You get it with the video, from a clip you already have, or through lipsync.
Veo 3.1, Seedance, Kling V3 Pro, MiniMax H3, Wan 3.0, Happy Horse, Flux 3, and Grok Imagine Video 1.5 make dialogue, ambience, and music in the same take as the picture. Some of them show a sound switch in the generate bar. On Kling V3 Pro, sound is off until you turn it on. On the rest, sound is always on.
To use sound you already have, upload an audio file, or open Audio on a video and choose Extract audio. Seedance 2.0 takes up to three reference audio clips, and Seedance 2.5 takes up to ten.
For a talking shot, use Chromatic Lipsync or Kling V2 Avatar Pro. Give either one a face and a voice track, and the person speaks the lines.
How to pick
Start from the job you need done, then choose the model.
- For a new still from a sentence, use any image model that works without references. Nano Banana, GPT Image, Midjourney, Flux, Krea, SeedDream, and Recraft all do.
- To change a still you already have, use an image model that can edit, or Edit & Merge, which connects the source for you.
- For a clip from a sentence, use a video model. The duration and aspect lists in the generate bar show what each model allows.
- To animate a still, choose Video on the image, then pick an image-to-video model.
- To keep the same character or product in a new clip, use a model that takes references, like Seedance, Happy Horse 1.0, Veo 3.1 or Gemini Omni Flash.
- To drive a character with a performance you already shot, use Motion Control.
- To change a clip you already have, use Edit Video.
- To make an image bigger or cut out its background, use Image tools instead of a larger image model.