Auto: let runime pick per task
Auto (shown as "Oto") is the first option in every model list and the default for new projects. Instead of using one model for everything, it picks the best-fitting model for each specific job — character design, a storyboard still, a moving shot, a line of dialogue, the score — because no single model is best at all of them.
Auto is also context-aware. For example, if your film features children, it steers video toward a child-safe model rather than one that blocks minors' faces; for a headscarf character or a photoreal look, it favors the model that handles that best. If you'd rather not think about any of this, Auto is the recommended setting.
Where to set models: whole-film vs. per-shot
There are two levels of control. In your project settings you set film-wide defaults for the image model, video model, voice, music, sound effects, and even the screenwriter — these apply to everything runime generates. Each list starts with Oto, so leaving them untouched keeps the whole film on Auto.
Inside the shot editor you can override the image and video model for a single shot. This is the tool for a specific problem — say one hero shot where you want a more consistent look — without changing the rest of the film. A per-shot choice always wins over the project default for that shot.
The line-up you can choose from
Every list holds only models we genuinely run — if a name is in the picker, that engine renders your shot and the take is credited to it.
- Images: Nano Banana, Nano Banana 2, Nano Banana Pro, GPT Image, Grok Imagine, and runime Görsel, our own self-hosted image model.
- Video: Veo, Seedance, Runway Gen-4.5, Kling, Grok Imagine, Luma Ray 2, and runime Video, our own self-hosted engine. The picker appends each engine's current generation (for example "Veo 3.1 Fast", the default) from the capability table, so a name never goes stale.
- Voice, music and sound effects: ElevenLabs, ElevenLabs Music and ElevenLabs SFX, alongside our self-hosted runime Ses, runime Müzik and runime SFX.
- Screenwriter: Gemini, or our self-hosted runime Senarist.
Depending on the account, a further set of engines from our TopView partner may appear in the video and image lists — Gemini Omni, Seedance 2.5 / 2.0, Kling, Wan, Veo and MiniMax Hailuo. They are hidden when they are not connected rather than falling back to something else under their name.
One safety net matters: if you pick a model that physically can't do a job (for example, a shot that needs start-and-end keyframes on a model that doesn't support them), runime won't force it — it automatically switches to a capable model so the shot still renders correctly, and says which one rendered it.
Choosing by consistency
When you open a model picker, each model carries a row of stars — a 0–5 rating for consistency, meaning how reliably that engine keeps your subjects, motion and physics coherent: no faces morphing, no objects floating or clipping through solid things, no drift across a set.
Read the colour as well as the count. An amber rating is anchored to a published benchmark for that engine at that generation. A muted rating is our own estimate — which is most of them, and the tooltip says so. Either way it is relative guidance, not a promise.
Underneath, a model also lists what it cannot do — no negative prompt, no end frame, no identity references — so you find out at the moment of choosing rather than from a wrong frame you already paid for.
Use the stars as a tie-breaker: if two models suit your style, the higher one is the safer pick for a shot with a lot of movement, a camera turn or a character reveal. If you're not sure, Auto already weighs this for you.
Quality is separate from the model
Which model renders a shot and how sharp the final file is are two different settings.
The Quality control on the home screen sets the resolution your film is genuinely rendered at: 720p by default, because it is faster and cheaper while you are still cutting, or 1080p, which asks the 1080-capable engines for a native 1080p render. Some engines cannot do 1080 and drop out of the line-up when you pick it; the control says which.
Separately, in the export dialog you choose a download tier — 720p, 1080p or 4K. Anything above the resolution you rendered at is produced by AI super-resolution: the model reconstructs detail rather than stretching pixels, but it is an enlargement of the render, not a second, higher-resolution shoot. You can also toggle a cinematic tone (film grain and colour grade) and smooth 60fps motion. These add-ons cost tokens and are priced up front; once you've produced a given tier, re-downloading that exact version is free.