ai-video-shots
Skill2 starsChanged 45 days agoMITView on GitHub
Choose and use paid AI video generators (Google Veo / Gemini Omni, Kling, Runway, Seedance, Hailuo, Luma) to add a few realistic shots to a code-made video — which tool for what, prices per second as of 2026-10-01, cost estimate before spending, image-to-video from our own still, and how to blend clips into a HyperFrames video. Use when Alexey mentions Veo, Kling, Sora, Runway, Seedance, AI video, AI-generated footage or realistic shots. For detailed Seedance prompting, see the community seedance-2.0 skill (github.com/Emily2040/seedance-2.0).
In this repository2 other files it gives its agents.
- video-deliverySkill
- video-studioSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 18
Every generation costs money.
- Distinctive: line 45
Keep clips 2–4 s. Conform them with FFmpeg: 30 fps, 1920×1080 (or crop for square), BT.709.
- Distinctive: line 55
No real people's likeness, no real brands.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AI video generators (for a few realistic shots)
Code (HyperFrames, see video-studio) is still the backbone: it renders exact words, numbers, UI and
brand colours perfectly. AI generators are for 2–4 short realistic shots code can't draw (a tired
recruiter at a desk, hands on a keyboard, an office, a video call). Never for text, numbers, logos
or UI.
Full research with sources, tagged OFFICIAL / 2ND / UNCONFIRMED, checked 2026-10-01:
references/tools-research.md. Prices change monthly: re-check the
official pricing page before quoting a cost.
Money rule (always)
Every generation costs money. Before any call: say the tool, clip count, seconds, resolution and the total in dollars (×3–5 for re-rolls), then wait for the owner's yes. Free tiers usually add a watermark or forbid commercial use, so they're for look tests only. Use a generator's API only with a key the owner has set up; never sign up for anything on their behalf.
Which one (as of 2026-10-01)
| Need | Pick | Price (per second of video) |
| Default, scriptable via API, image-to-video | Google Veo 3.1 via the Gemini API | Lite $0.05 (720p) / $0.08 (1080p, 8 s clips only); Fast $0.12 (1080p) |
| Most realistic people, strong image-to-video | Kling 3.0 Pro (e.g. via fal) | ~$0.11–0.17 |
| Newest Google model, edits by conversation, up to 4K | Gemini Omni Flash 1.1 | ~$0.10 (720p), ~$0.15 (1080p) |
| Cheap look tests | Veo 3.1 Lite, Hailuo H3 (~$0.08, 768p) | |
| Strong motion, multi-shot | Seedance (ByteDance) | ~$0.23 at 720p (2nd-hand figure) |
| Gone or not usable: OpenAI Sora (app closed April 2026; API removed 2026-09-24). No "Veo 4" | ||
| exists. Open models (Wan 2.2, LTX) need an NVIDIA GPU; without one, free hosted GPUs are | ||
| too slow to be practical. |
Example estimate: 4 clips × 4 s at 1080p ≈ $2–3 on Kling 3.0 Pro or Veo 3.1 Lite (Veo Lite bills 8 s per 1080p clip). With re-rolls, budget $8–15. Start with the same 2 prompts on two tools (~$1–2), pick the winner, then do the rest.
Workflow that keeps the brand
- Design the shot's first frame as a still in our style (
hyperframes snapshotof a scene, or a frame from the video) and generate image-to-video from it. Use last-frame conditioning where the tool offers it, so the clip ends where the next code scene begins. - Prompt the feeling and the camera, not every detail (the community seedance-2.0 skill explains this well): subject, action, camera move, light, mood, "no text, no logos".
- Keep clips 2–4 s. Conform them with FFmpeg: 30 fps, 1920×1080 (or crop for square), BT.709.
- Grade towards the brand: lift shadows towards your background colour, tint highlights with your accent colour, the same film grain as the code scenes (4–6 %) on top, so AI and code shots read as one film.
- Bring each clip into HyperFrames as a
<video>layer (see themedia-useandhyperframesskills in the project's.claude/skills), muted; sound stays our music and SFX. - Download results straight away: Veo files are deleted from Google's server after 2 days.
Traps
- Text inside AI clips comes out garbled: never ask for signs, screens with words, or logos.
- Hands and fingers on keyboards still fail often; frame them loosely or out of focus.
- No real people's likeness, no real brands. Check the tool's rules on realistic people.
- Many 2026 blog posts are out of date (they still say "use Sora 2"). Trust the official pricing and model pages, with a date.
- Watermarked or non-commercial free-tier clips must never end up in a posted video.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.