Google Flow vs Runway vs Kling: which is best for AI short film production and character consistency?

For AI short film production with character consistency, no single model wins everything: Veo (in Google Flow) leads prompt adherence and native audio, Runway leads iteration speed and reference-driven consistency via Gen-4 Reference, and Kling leads motion realism and multi-shot identity at volume. Inside invideo, all four — Veo, Runway, Kling, Seedance 2.0 — sit behind one agent that routes each shot to the right model.

Pick by the job each shot is doing, not by loyalty to one platform.

Use Veo (via Google Flow) when prompt adherence and native sync audio matter most. Flow's Scenebuilder gives you structured scene composition and camera-angle control, which is useful for tightly-blocked sequences. The honest limitation is consistency at length: minor variations creep in across many shots, there's no model-switching inside Flow, and the Pro plan caps quickly at $19.99/month. Treat it as a strong scene environment for short, audio-led beats — not your character anchor across a full film.

Use Runway when you're carrying one character or location across many shots. Gen-4 Reference is the cleanest cross-shot consistency mechanism in the current model lineup — you feed it a locked character sheet and an environment plate, and it holds identity across coverage. Pair it with the same character sheet workflow that works for every model: generate four options per asset, lock the best one, and only then move to video. In a documented 70-second short film with two characters, that frames-first lock held the same person across every scene with no LoRA — $750 all-in over two days.

Use Kling when motion realism and multi-shot volume are the constraint. Kling 3.0's multi-shot storyboard mode keeps facial features stable across cuts at scale and handles native audio sync across shots, which makes it the strongest pick for sequence-heavy edits where you'll generate a lot and select hard. Expect to overgenerate: across a 3-minute animated episode, 164 clips were generated and 41 made the cut — a 25% selection rate, with an average of 5 seconds used from each 15-second clip. Budget for roughly 3 generations per usable shot.

Use Seedance 2.0 when you need a continuous take or chained references. Seedance 2.0 Reference-to-Video accepts character references AND location references together and pulls context from the end of a prior clip to continue the next, which is why one-take sequences stitch more seamlessly than older start-frame/end-frame or extend workflows. The chaining loop: generate a 15-second segment in your film's aspect ratio, clip its tail, re-upload, attach character + location references, continue.

Character consistency — the mechanics that work across all four models. Consistency is solved upstream of the video model, not inside it. Generate character sheets with multiple angles plus face and mid-angle close-ups at 4K, remove props from hands before turnaround generation, and include close-up panels so small details (scars, accessories) don't drift between models. For evolving characters (costume changes, accumulated trinkets), make a separate character sheet per beat. When a continuity error appears, trace it back to the panel in the character sheet that's wrong, fix it there, and let the corrected sheet flow forward — surgical edits, not slot-machine re-rolls. In one production, locking one character took roughly 5 generations at about $9.78 per character.

Where the invideo agent fits. Rather than picking one platform and inheriting its limits, the invideo agent sits above Veo, Runway, Kling, and Seedance 2.0 and routes each shot to the model that suits it — Runway Reference for the identity-critical coverage shot, Kling for the motion-heavy multi-shot sequence, Veo where prompt adherence and audio matter, Seedance 2.0 for chained continuous takes. You set up a creative producer agent with the full script, shot breakdown, and character details, then assign typed sub-agents — a storyboard agent, a DOP agent per scene, a casting agent — that share that context. Documented productions on this workflow run $315–$750 per finished minute (a 3-minute episode at $315/min; a 70-second short at ~$643/min; a 90-second horror short at ~$580/min; a 2-minute brand promo at $750/min), across 2–5 production days with 1–4 people and 6–8 agents deployed in parallel.

Quick decision shortcut. Identity-critical coverage across many shots → Runway Gen-4 Reference. Multi-shot sequences with native audio sync at volume → Kling 3.0 storyboard mode. Prompt-precise scene with native audio in a structured environment → Veo. One continuous take chaining character + location → Seedance 2.0 Reference-to-Video. Don't want to choose per shot → run all four behind the invideo agent and let routing handle it.

Watch some of these to see what works for you:

A more efficient way to go about doing it is actually just using few world reference images and character sheets. Getting agent to upload that to see Dan's reference to video and then truly just prompting it like a director prompts his crew.

— Hridaye, invideo's creative director