Nolan runs several frontier models behind one chat, and its agent picks automatically. But when you're driving in Manual mode, this is the map we use.
Talking humans → Seedance 2
The only model with native lip-synced dialogue — write the line in quotes and the person says it. Also does multi-shot sequences in a single clip and takes up to 9 image + 3 video + 3 audio references. UGC ads live here.
Stylised, cinematic, consistent → Kling 3.0
The strongest look for story-driven work: up to 4K, native audio, multi-shot, and Elements — named characters/objects (2–4 images each) that stay consistent across the shot. Prompt with a camera verb early: dolly, orbit, crane, handheld.
The utility belt → Wan 2.7
Four modes in one model: text-to-video, image-to-video (first/last frame or continue a clip), reference-to-video with up to 5 refs plus a voice clip for lip-sync, and Video Edit — change an existing clip by instruction. Keep prompts brief and concrete, and use the negative prompt.
Animate a still → Kling Motion Control
One subject image + one driving video (3–30s): the image performs the video's motion. Your prompt only describes fidelity — the driving video is the performance.
Stills → GPT Image 2 and Nano Banana Pro
GPT Image 2 for product shots, graphics, and reliable in-image text. Nano Banana Pro when the same face or brand look must hold across a whole set — feed it multiple references of the subject.
Cost cheat sheet
At 1 credit = $0.01: a 2K still runs ~20 credits; a 10-second 720p clip lands around 200–420 credits depending on the model; 1080p roughly doubles it. The composer shows the estimate before you spend.