AI models
The models behind every film
Utopia makes each part of a film with Google's latest models on Vertex AI, plus an open Basque narrator on its own servers and, with an add-on, video and 3D models from other companies. Beginner mode picks the most efficient ones for you; Professional and Filmmaker let you choose each one, and the app tells you what a generation costs before you make it.
Script, assistants and the production office
Gemini 4 Argon
Google's newest and most powerful model
The first model of the Gemini 4 generation (October 2026). Writes the scenario, the characters and the storyboard with the best coherence and the most detail.
Strengths
- The best coherence in long stories with many characters
- Follows complex style and tone instructions precisely
- Richer dialogue and descriptions
Gemini 3.8 Flash
RecommendedBalance of quality, speed and cost
The latest Flash model (September 2026): fast and with very good storytelling. The recommended choice for everyday work.
Strengths
- Fast answers
- Very good script and dialogue quality
- Low credit usage
Gemini 3.5 Flash-Lite
The fastest and cheapest
A light text model (July 2026) for drafts and quick tests.
Strengths
- The fastest
- The lowest credit usage
Characters, settings and storyboard
Nano Banana Pro
The highest image quality
Gemini 3 Pro Image offers the highest quality, the best composition and legible text inside the image.
Strengths
- Maximum detail and realism
- Complex compositions with several characters
- Legible text inside the image
Nano Banana 2.1
RecommendedThe newest: great quality at a low cost
Google's latest image model (October 2026). Keeps characters and places looking the same across scenes, with low credit usage.
Strengths
- Very good quality at 2K
- Keeps the look of characters and places
- Low usage per image
Nano Banana 2 Lite
The fastest and cheapest
A light version of Nano Banana 2 (June 2026) for drafts and quick tests.
Strengths
- The fastest
- The lowest usage per image
Shots with video and sound
Veo 3.1
The best cinematic quality
Google's best video model: realistic motion and physics, great detail and dialogue with synchronized audio.
Strengths
- The best image and motion quality
- More natural synchronized voices and sound
- Up to 4K
Veo 3.1 Fast
Nearly Veo 3.1 quality, much cheaper
Makes high-quality video faster and with 30% of the credits Veo 3.1 uses at 1080p with audio.
Strengths
- Very good cinematic quality
- Faster than Veo 3.1
- Up to 4K
Gemini Omni Flash
Top-ranked realism for less
Google's newest video model (preview): in blind votes it ranks above Veo 3.1, for less than half its credits at 1080p.
Strengths
- More realistic and coherent than Veo 3.1 in blind votes
- Synchronized sound
- Up to 4K
Veo 3.1 Lite
RecommendedQuick, cheap tests
The lightest version of Veo 3.1, ideal for drafts and for trying ideas before the final version.
Strengths
- The fastest
- The lowest usage per second
- Supports synchronized audio
Music
Lyria
Original music for each film
Google's music model composes the soundtrack from the film's genre and mood.
Voices
Gemini Text-to-Speech
Narration and dialogue in more than 30 languages
Fixed voices or your own cloned voice, with the tone each line asks for, for narration, dialogue and dubbing.
Antton
A native Basque narrator
HiTZ's open voice (UPV/EHU), more natural in Basque than Gemini's voices in our blind test. It runs on Utopia's own servers in Google Cloud and narrates stories told in Basque, on the Pro and Estudio plans.
Models from other companiesAdd-on
With an add-on for the paid plans, these models join Google's and are paid with usage credits. They run on fal.ai, in the United States, and are never used if your account chooses «Google models only» or «Data in the EU».
Shots with video and sound
MiniMax H3 Max Reference
The same face from shot to shot
Starts from the storyboard image and also looks at the characters' portraits, so they keep their faces and clothes between shots. Runs on fal.ai.
Strengths
- Uses the characters' portraits as references
- Synchronized sound with dialogue
- Clips of 1 to 15 seconds
Kling 3.0 Pro
Cinematic motion with sound
Kling's top model: smooth, cinematic camera and body movement, with its own sound. Runs on fal.ai.
Strengths
- Very natural movement
- Synchronized sound
- Clips of 3 to 15 seconds
Wan 3.0 Prime
Top of the blind votes, for less
Alibaba's model, first in blind votes in October 2026, at 720p with sound. Runs on fal.ai.
Strengths
- Highly rated realism
- Synchronized sound
- Clips of 2 to 30 seconds
Seedance 2.5 Reference
Long shots with many references
ByteDance's model: clips of up to 30 seconds that keep the characters' portraits. It doesn't accept faces made from a real person's photo. Runs on fal.ai.
Strengths
- Clips of up to 30 seconds
- Uses the characters' portraits as references
- Synchronized sound
«Create in 3D»
Turns the image of a character or object into a 3D model that the Unreal Engine and Unity previs places instead of the grey stand-in.
Hunyuan3D 3
Tencent
Trellis 2
Microsoft
Tripo 2.5
Tripo AI
Your stories go to Google Cloud's Vertex AI only to generate what you ask for; under its terms, Google doesn't use them to train its models. Antton runs on Utopia's own servers, and the add-on's models on fal.ai only when you use them.