The Current State of Video-Generation AI
OpenAI discontinued Sora: the web and mobile apps closed on 2026-04-26, and the Sora 2 API shuts down on 2026-09-24, with no successor announced. Plan around Runway, Kling, or Google Veo instead. Video-generation AI reached practical level in 2025-2026. The standard is to make 5-10-second shots and stitch them. For long form, use Runway's/Pika's frame-extension and Veo's long generation.
Selective Use of Main Models
| Model | Strength | Use |
|---|---|---|
| Runway Gen-4 / Turbo | Character consistency, editing integration | Film production, ads |
| Google Veo | Simultaneous audio, long form, Workspace linkage | Corporate video, education |
| Kling 2.1 / Master | Cinematic motion, low price | Social, personal creation |
| Pika 2.2 | Unique effects, for TikTok | Social viral |
Production Workflow
- Storyboard: decide shot composition on paper or a tool (Storyboard.ai etc.)
- Keyframe generation: make each shot's "hero frame" with Midjourney or Flux
- Shot generation: image-to-video to move each shot for 5-10 sec
- Add audio: native with Google Veo; otherwise generate separately with ElevenLabs / Suno. Runway also added Audio Nodes to Workflows — drop in Text to Speech, SFX, dubbing, or voice isolation as nodes in the same workflow (Source: https://x.com/runwayml/status/1991548978295455780).
- Edit: stitch/color-correct in DaVinci Resolve, CapCut, Premiere
- Finish: BGM, captions, output-format adjustment
Tips to Keep Consistency
- Reference images: reproduce the same person with Runway References or Kling's reference-based generation
- Character fixing: record the seed number, reuse on regeneration
- Unify color palette: same color temperature/lighting in the prompt
- Motion continuity: use the previous shot's last frame as the next shot's start frame
Prompt Design
Unlike stills, you need vocabulary expressing motion and the time axis.
- Motion type: pan, zoom, dolly, tracking, push-in
- Speed feel: slow motion, time-lapse, normal speed
- Camera move: handheld, smooth gimbal, drone aerial
- Character action: walking, turning, gesturing
Commercial Use and Terms
- Google Veo: outputs always carry a SynthID invisible watermark
- Runway: commercial OK from Standard up, no watermark
- Kling: commercial use on higher plans
- Common to all: generating others' characters/celebrities is prohibited
Cost Guide
| Use | Monthly guide |
|---|---|
| Individual creator (short-form centric) | $10-$30 |
| Small marketing | $50-$150 |
| Corporate PR/ad production | $200-$1,000 |
| Film/drama assistance | $1,000-$10,000 |
Failure Patterns
- Making one shot too long (quality drops over 10 sec)
- Not minding character consistency, breaking later
- Planning around a discontinued service such as Sora instead of models still in operation
- Delivering to client without confirming commercial license
Summary
Video-generation AI is practical with the workflow storyboard → keyframe → shot → edit. Use Runway, Veo, and Kling selectively by purpose, and guarantee consistency with reference images and seed management—the modern production style.
