ReelVeo or Sora vs videos made in code
Video models generate footage from a prompt. A coding agent writes the film as code and your computer renders it. They suit different videos, and the market for the first group changed this autumn. Facts checked on 2026-11-09.
How the two work
A video model such as Google's Veo creates pixels from a text prompt, usually billed by the second of video. A coding agent such as Claude Code writes HTML and JavaScript that draws every scene, and a renderer such as HyperFrames records it to an MP4 on your computer. ReelStyles gives the agent a finished look to follow: palette, characters, scene templates and a checklist it grades its own stills against.
What changed in the video model market
- Sora. OpenAI's deprecations page lists the Videos API and the Sora 2 models (
sora-2,sora-2-proand their dated versions) as removed from the API on 2026-09-24. The notice went out on 2026-03-24, and the page names no replacement. You cannot build on the Sora API today. - Veo. Google's video guide says the Veo 3.1 preview models in the Gemini API were shut down on 2026-10-22. It points to Gemini Omni Flash, or to the generally available Veo models on Google's enterprise agent platform.
- Price of the replacement. Google's pricing page lists Gemini Omni 1.2 Flash video output at $12.00 per million tokens, which it puts at about $0.07 per second of 720p video, and about $0.104 per second at 1080p with upsampling.
Model names and prices move fast. Read the source pages below before you plan around any of this.
Side by side
| Video model (Veo, Gemini Omni) | Video made in code (ReelStyles) | |
|---|---|---|
| What it makes | Generated footage, including realistic scenes and people | Drawn and animated films in one set look |
| Billing | Per second of video; Google lists about $0.07 per second at 720p for Gemini Omni 1.2 Flash | No per second fee. You pay once for the kits and use your agent on your own plan |
| Same character every scene | Depends on the model and your prompts | Built into the kit, because the agent draws the same cast from the same code |
| Edit one detail later | Generate again | Change a line of code and render again |
| Runs where | On the provider's servers | On your computer |
Where video models are better
- Realistic footage. A model can show a real-looking street, a face or a camera move. A coding agent draws cartoons, paper, paint, diagrams and toy-like 3D, and it cannot make a photograph of a place.
- Speed to a first clip. A prompt gives you a clip in a short time, with no scene planning.
- Native sound. Gemini Omni Flash makes video with native audio, according to Google's guide. With a kit you add voice and music through your agent.
Where code-made video fits
- Faceless channels that need the same look in every video.
- Explainers, stories and Shorts in a drawn style. 46 of our 126 styles are tested in 9:16.
- Teams that want a cost they can predict, since the film is not billed per second.
Our measured cost for one finished Short, next to the listed model prices, is in Claude Code video without Veo or Sora.
Sources, checked 2026-11-09
| Claim | Source |
|---|---|
| Videos API and Sora 2 models removed on 2026-09-24, announced 2026-03-24, no replacement listed | OpenAI API deprecations |
| Veo 3.1 preview models shut down on 2026-10-22; Gemini Omni Flash and generally available Veo named as the way forward; native audio; 3 to 15 second clips | Gemini API video guide |
| Gemini Omni 1.2 Flash video output $12.00 per million tokens, about $0.07 per second at 720p and about $0.104 at 1080p | Gemini API pricing |
| HyperFrames licence Apache-2.0 | github.com/heygen-com/hyperframes |
We make ReelStyles, so we are not neutral. If we got a fact wrong, tell us and we fix it.
Questions
Is Sora still available?
Can code-made video look realistic?
Do I pay per second of video with ReelStyles?
Also see: Claude Code video styles · Claude Motion vs ReelStyles · faceless Shorts styles.