Why the extra step is worth it

Pure text-to-video gives you one shot at composition, lighting, character, colour and motion all at once. You almost always lose one of them. Splitting the problem — nail the frame first, then animate — is how you get shots that look intentional.

Step 1 — the hero frame

Use Flux 2 Pro, Seedream 5 or Nano Banana 2. Prompt for the exact first frame of your shot — including where the subject is looking, how the light falls, the aspect ratio. Regenerate until this single frame is right. This is cheap.

Step 2 — light retouch if needed

If one detail is wrong, edit it (Flux 2 Pro Edit, Nano Banana 2 Edit). Don't roll the whole image again for a small fix.

Step 3 — animate

Feed the finalised frame to Kling 2.5, Veo 3.1 or Seedance in image-to-video mode. Your prompt now only describes motion — 'slow push-in, subject blinks at 2s, dust settles' — not composition. Higher hit rate, lower cost.

Step 4 — audio (optional)

If you need dialogue or foley, add it in the same generator when supported (Sora 2 doesn't do image-to-video with audio yet — use Veo 3.1 for audio + i2v, or layer separately).