Pencil gives you AI models from many providers for copy, images, video, voice and music. This article helps whenever you choose a model yourself, for example in Sheets, Feeds, Workflows or the editor.
The recommendations here come from each model’s described strengths. They are not benchmark results, so try two or three models on your own work before you commit to one.
Which models can you use?
Your workspace’s list is the one that counts. An admin chooses which models your workspace can use, so you may see fewer models than this article lists.
Only workspace admins see AI governance in Settings. An admin goes to Settings > AI governance > Generation models and turns on the Enabled toggle next to each model the workspace needs. If you aren’t an admin and a model is missing, ask a workspace admin to switch it on. For more help, read accessing all models.
Marketplace models work differently. They stay off until your Account Owner signs the Marketplace Model Terms, and they come without Pencil warranties or indemnities. Read standard and marketplace models before anyone switches them on.
Two more things to know:
MiniMax models. You accept a one-time acknowledgement of Chinese law before your first generation with a MiniMax model.
ElevenLabs voices. Pencil connects to ElevenLabs through its API, so Voice Design and voice cloning are not available.
Which model should I start with?
Job | Try first |
Demanding reasoning and long agentic work | Claude Fable 5 or Claude Opus 5.5 |
Nuanced, tone-sensitive writing | Claude Opus 4.7 |
High-volume copy, such as bulk work in Sheets | Gemini 3.8 Flash or GPT-5.6 Terra |
Quick, low-stakes tasks at scale | Gemini 3.1 Flash Lite or GPT-5.4 Nano |
Product fidelity and pack shots | Nano Banana 2.1 or GPT-Image-2.5 Sunburst |
Accurate text inside images | GPT Image 2 or Nano Banana Pro (Gemini 3) |
Fast concepts and social volume | GPT-Image-2.5 Flare or Nano Banana 2 Lite |
Safer copyright provenance | Adobe Firefly Image 5 |
Video with native sound | Gemini Omni Flash 1.1 |
Cinematic video | Runway Gen-4.5 |
Consistent characters, storyboards and lip-sync | Kling VIDEO 3.0 Omni |
Video at volume | Veo 3.1 Lite |
Consistent narration in many languages | ElevenLabs Multilingual V2 |
Expressive voiceover | ElevenLabs Eleven V3 |
Full-length songs | Google Lyria 3 Pro |
For every model and what it does best, please see the dedicated subsection articles:
Which model should I use for translation?
Method matters more than model choice. Translate approved copy as a separate step, keep one model for the whole project, and have a person review every translation.
The best model also depends on the language, so test on your own language pair. The text models article suggests starting points, and the translation tips cover keeping wording consistent.
How do I upscale or enhance an asset?
Pencil includes Topaz models for upscaling and restoring images and video you already have. Try a faithful pass through the model that generated the asset first. Switch to Topaz once that model reaches its own resolution limit.
Model | Type | Best for |
Standard Max | Image | Everyday all-round upscaling and restoration |
Recovery V2 | Image | Fast restoration of small, low-resolution sources |
Wonder | Image | One-step repair of compressed, soft images |
High Fidelity V2 | Image | Staying closest to the source on already-sharp images |
Redefine | Image | Prompt-driven, creative rebuilds of degraded graphics |
Starlight Precise 2.5 | Video | Taking low-resolution or AI footage to 4K, and heavy restoration |
AI Video Enhance | Video | Hands-on restoration with fully manual controls |
Proteus Natural | Video | General-purpose, everyday video upscaling |
What does each model cost?
Costs vary by model. The pricing article explains how Pencil counts credits.
Pencil adds new models often, so check which models your workspace offers before you plan work around one.

