moonlight-lupin/agent-skills / creative/clips-studio
clips-studio
VerifiedUse when the user wants to "make a video/clip", "generate a video from text", "animate this image/photo", "create a marketing reel / social video / teaser", "do a 3D / parallax move", "pan/zoom/orbit a shot", or mentions fal.ai, Kling, Veo or Seedance for video. Staged fal.ai workflow — brainstorm, draft cheaply, produce the final at quality. Three modes: text-to-video, animate (still to motion), camera-move (push-in/pan/orbit). Not for still images (image-studio), slide decks, or charts. Honesty discipline: never depict real identifiable subjects via text-to-video.
| File | Kind | Size |
|---|---|---|
| README.md | markdown | 1.9 KB |
| SKILL.md | skill md | 23.5 KB |
| examples/example-run.md | markdown | 8.1 KB |
| references/fal-video-models.md | markdown | 11.9 KB |
| references/motion-honesty-guide.md | markdown | 11.4 KB |
| references/video-prompting-techniques.md | markdown | 3.4 KB |
| scripts/falvid.py | script | 34.5 KB |
| tests/__init__.py | script | 0 B |
| tests/test_falvid.py | script | 31.3 KB |
references/fal-video-models.md
View on GitHub# fal.ai VIDEO model registry — Clips Studio
Endpoint IDs the helper (`scripts/falvid.py`) uses, by mode. **fal.ai's catalogue changes often** —
if an ID errors, check the live gallery at https://fal.ai/models (filter to *Text to Video* /
*Image to Video*).
These are **text-to-video** (`generate`, a prompt → an MP4) and **image-to-video**
(`animate` / `camera`, a start image + a prompt → an MP4) endpoints. They run **async** —
`falvid.py` uses `fal_client.subscribe`, which polls the queue and blocks until the clip is ready
(often a few minutes per clip).
**Cost is an estimate, not live-exact.** Video is billed per-second (Kling, Veo, FLUX 3) or per-clip/
per-token (Seedance, Hailuo), and the exact charge depends on output **duration × fps × resolution ×
audio**, which the API does **not** return. `falvid.py` therefore computes a **verified-rate
estimate** from the rates below × the *requested* duration and flags it as an estimate (it does not
query a live pricing API for video). **The fal dashboard is the authoritative charge.** Rates
verified against fal's pricing pages on **9 Aug 2026**; *(verify)* marks anything not confirmed live
this build.
## Contents
- FLUX 3 (Black Forest Labs) — DEFAULT for draft + full
- Seedance 2.5 (ByteDance) — available option (not default)
- Seedance 2.0 (ByteDance) — available option (not default)
- Text-to-video — `generate` (a clip from a prompt, no source image)
- Image-to-video — `animate` (bring a still to life: product, people, a space, motion)
- Image-to-video — `camera` (extrapolated camera motion over a still)
- Common inputs (and how `falvid.py` maps the flags)
- Cost reality check
- Deferred / not wired here
## FLUX 3 (Black Forest Labs) — DEFAULT for draft + full
Native audio included free. Up to 20 seconds. Draft Enhance re-renders a draft at full quality
with the same seed. Five full endpoints + two draft endpoints.
| Model | Endpoint id | Role | Rate | Notes |
|---|---|---|---|---|
| FLUX 3 Text to Video | `blackforestlabs/flux-3/text-to-video` | **Full default (generate)** | $0.17/s 720p · $0.29/s 1080p | Native audio free. Up to 20s. |
| FLUX 3 Image to Video | `blackforestlabs/flux-3/image-to-video` | **Full default (animate)** | $0.17/s 720p · $0.29/s 1080p | Animate a still; native audio free. |
| FLUX 3 First-Last Frame | `blackforestlabs/flux-3/first-last-frame-to-video` | Alt | $0.17/s 720p · $0.29/s 1080p | Start + end frame → video between. |
| FLUX 3 Keyframes | `blackforestlabs/flux-3/keyframes-to-video` | Alt | $0.17/s 720p · $0.29/s 1080p | Up to 10 pinned keyframes. |
| FLUX 3 Extend Video | `blackforestlabs/flux-3/extend-video` | Alt (continuation) | $0.41/s 720p · $0.53/s 1080p | Continue a clip; ~2.5× the t2v/i2v rate. |
| FLUX 3 Text to Video Draft | `blackforestlabs/flux-3/text-to-video/draft` | **Draft default (generate)** | $0.06/s | HD only; ~1/3 of full. Native audio. |
| FLUX 3 Image to Video Draft | `blackforestlabs/flux-3/image-to-video/draft` | **Draft default (animate)** | $0.06/s | HD only; ~1/3 of full. Native audio. |
**Draft Enhance workflow:** Draft on the cheap endpoint → note the seed → `falvid.py enhance --seed
<seed>` re-renders at full quality. The seed locks the motion; the full render adds fidelity.
## Seedance 2.5 (ByteDance) — available option (not default)
Native 30-second generation, up to 50 multimodal references, native audio. Token-based pricing makes
it **~9× more expensive** than Seedance 1.5 for a 5s clip. Use for long takes or reference-to-video,
not for short camera moves.
| Model | Endpoint id | Rate | Notes |
|---|---|---|---|
| Seedance 2.5 Image to Video | `fal-ai/bytedance/seedance-2.5/image-to-video` | ~$0.473/s (720p w/audio) | 30s max; `end_image_url`; `duration: "auto"`; `aspect_ratio: "auto"`. |
| Seedance 2.5 Text to Video | `fal-ai/bytedance/seedance-2.5/text-to-video` | ~$0.473/s (720p w/audio) | 30s max. |
| Seedance 2.5 Reference to Video | `fal-ai/bytedance/seedance-2.5/reference-to-video` | ~$0.473/s (720p w/audio) | Up to 50 refs (30 images + 10 videos + 10 audio). |
**Cost reality:** tokens = (height × width × duration × 24) / 1024, at $0.0214/1k tokens.
5s/720p/16:9 = 108k tokens ≈ $2.31. 30s/720p = 648k tokens ≈ $13.87.
## Seedance 2.0 (ByteDance) — available option (not default)
Native 15-second generation, 1080p output, native audio free. Token-based pricing. Better physics
and camera control than 1.5, but **no `camera_fixed` parameter** (use prompt language for static
shots). Standard + Fast tiers.
| Model | Endpoint id | Rate | Notes |
|---|---|---|---|
| Seedance 2.0 Image to Video | `fal-ai/bytedance/seedance-2.0/image-to-video` | $0.30/s (720p) · $0.68/s (1080p) | 15s max; `end_image_url`; multi-shot; audio free. |
| Seedance 2.0 Fast Image to Video | `fal-ai/bytedance/seedance-2.0/fast/image-to-video` | $0.24/s (720p only) | Lower latency; no 1080p. |
| Seedance 2.0 Text to Video | `fal-ai/bytedance/seedance-2.0/text-to-video` | $0.30/s (720p) · $0.68/s (1080p) | 15s max; multi-shot; audio free. |
| Seedance 2.0 Reference to Video | `fal-ai/bytedance/seedance-2.0/reference-to-video` | $0.30/s (720p) | Up to 9 images + 3 videos + 3 audio refs. |
**Cost reality:** tokens = (height × width × duration × 24) / 1024, at $0.014/1k tokens.
5s/720p/16:9 = 108k tokens ≈ $1.51. Use Fast tier for 720p to save ~20%.
## Text-to-video — `generate` (a clip from a prompt, no source image)
Rates here are **`(verify)`** — t2v IDs/prices move and weren't all confirmed live this build; the
helper's estimate is deliberately conservative, and the fal dashboard is authoritative. **Never use
text-to-video to depict a real, identifiable subject** (it invents everything) — `animate` a real
photo instead.
| Model | Endpoint id | Role | Rate | Notes |
|---|---|---|---|---|
| Wan 2.2 | `fal-ai/wan/v2.2-a14b/text-to-video` | **Draft (default for `generate`)** | ~$0.10 / s *(verify)* | Cheap t2v draft. If the ID errors, try `fal-ai/wan-pro` or fall back to a Kling/LTX t2v. |
| Veo 3.1 | `fal-ai/veo3.1` | **Quality final** | $0.20/s no-audio · $0.40/s audio · ×2 4K | Veo 3.1 text-to-video — best realism + native audio. (The base id is the t2v endpoint; `…/image-to-video` is the i2v one.) |
| Kling 2.5 Turbo Pro | `fal-ai/kling-video/v2.5-turbo/pro/text-to-video` | Alt | ~$0.224/s no-audio · ~$0.28/s audio *(verify)* | Note: Kling **t2v prices higher than its i2v** ($0.07/s). 1080p. |
| Kling 3.0 Pro | `fal-ai/kling-video/v3/pro/text-to-video` | Alt (cinematic) | ~$0.224–0.336/s *(verify)* | Up to 15s, native audio. |
| LTX-2 | `fal-ai/ltx-2-19b/text-to-video` *(verify)* | Alt (very cheap) | ~$0.0018 / MP | Per-megapixel; cheap, lower fidelity. `falvid.py` estimates ~249 MP for 5s/1080p@24fps → ~$0.45. |
| Wan-Pro | `fal-ai/wan-pro` *(verify)* | Alt (fallback) | ~$0.10 / s *(verify)* | Fallback if the default Wan 2.2 ID errors. |
| Seedance 1.5 Pro | `fal-ai/bytedance/seedance/v1.5/pro/text-to-video` | Alt | ~$0.26 / clip *(verify)* | Per-clip; 4–12s. *(verify)* — confirm endpoint exists on the live gallery. |
| Hailuo 2.3 Pro | `fal-ai/minimax/hailuo-2.3/pro/text-to-video` | Alt | ~$0.49 / clip *(verify)* | Per-clip; ~5s. *(verify)* — confirm endpoint exists on the live gallery. |
## Image-to-video — `animate` (bring a still to life: product, people, a space, motion)
| Model | Endpoint id | Role | Rate | Notes |
|---|---|---|---|---|
| Kling 2.5 Turbo Pro | `fal-ai/kling-video/v2.5-turbo/pro/image-to-video` | **Draft (default for `animate`)** | **$0.07 / s** | Cheap workhorse for iterating on motion. Up to 10s, 1080p. `image_url`; `duration` is a **string** enum. |
| Veo 3.1 | `fal-ai/veo3.1/image-to-video` | **Quality final (default)** | **$0.20/s** no-audio · **$0.40/s** audio · **×2 at 4K** | Google Veo 3.1 — best human motion + realism + native audio. ~8s. `image_url`, `aspect_ratio` (default `auto`). A 5s/1080p clip ≈ $1 (no audio) / $2 (audio). |
| Veo 3.1 Fast | `fal-ai/veo3.1/fast/image-to-video` | Quality (cheaper) | ~$0.10/s no-audio · ~$0.20/s audio *(verify)* | Faster/cheaper Veo tier for budget runs. |
| Kling 3.0 Pro | `fal-ai/kling-video/v3/pro/image-to-video` | Quality (leaner) | ~$0.11–0.20 / s | Cinematic, fluid motion, native audio; up to 15s, 1080p. Uses `start_image_url`. `falvid.py` uses the midpoint ($0.15/s) for estimates — the dashboard is authoritative. |
## Image-to-video — `camera` (extrapolated camera motion over a still)
| Model | Endpoint id | Role | Rate | Notes |
|---|---|---|---|---|
| Seedance 1.5 Pro | `fal-ai/bytedance/seedance/v1.5/pro/image-to-video` | **Default for `camera`** | **~$0.26 / clip** (720p/5s w/audio; scales with tokens) | ByteDance — strongest **camera granularity** (pan, zoom, dolly, orbit) and a true **`camera_fixed`** lock (used by `--static`). `image_url` + optional `end_image_url`; `resolution` 480p/720p (helper defaults 720p); `duration` 4–12 int; `generate_audio`. |
| Kling v3 4K | `fal-ai/kling-video/v3/4k/image-to-video` | Premium native 4K | **$0.42 / s** | Native 4K, no upscaling; `duration` "3"–"15"; uses **`start_image_url`**; camera via prompt language. The dearest option — reach for it only when 4K is genuinely needed. |
| Hailuo 2.3 Pro | `fal-ai/minimax/hailuo-2.3/pro/image-to-video` | Alt | ~$0.49 / clip | MiniMax; 1080p, ~5s; camera via prompt. |
## Common inputs (and how `falvid.py` maps the flags)
- **Start image** — `image_url` for most models; **`start_image_url`** for the Kling v3 family. The
helper picks the right key automatically; override with `--arg`.
- **`prompt`** — describes the action *and* the camera move. Name moves in cinematographic terms
("slow dolly push-in", "pan left", "shallow orbit"); say "static shot, locked-off camera" to hold
still on models without a `camera_fixed` parameter.
- **`duration`** — seconds. Kling wants a **string** ("5"); Seedance/Veo take an **int**. The helper
casts per model. Keep it short — cost scales with every second.
- **`resolution`** — model-dependent (Seedance 480p/720p; Veo 720p/1080p/4K; Kling v3 native 4K).
Only sent when you pass `--resolution`, except Seedance defaults to `720p`.
- **`aspect_ratio`** — for `animate`/`camera`, only sent when you pass `--aspect`; the models do
**not** reliably inherit the still's framing (Seedance defaults to 16:9 and centre-crops a
portrait/square still), so **pass `--aspect` to match the source**. For `generate` there's no
source, so the helper sends the target aspect (default 16:9).
- **`generate_audio`** — **default ON for FLUX 3** (native audio included free; pass `--no-audio`
to disable). **Default OFF for all other models** (Veo, Kling, Seedance) — opt in with `--audio`.
Turning it on raises Veo's per-second rate.
- **`camera_fixed`** — Seedance only; set by `--static`. Other models get a "static shot" instruction
in the prompt instead.
- **`seed`** — `--seed` for repeatability. **`negative_prompt`** — via `--arg negative_prompt='…'`
(Kling defaults to `"blur, distort, and low quality"`).
- **Output** — every model returns a `video` object with a `url` (MP4); the helper downloads it to
`--out-dir` as `[name].mp4`.
## Cost reality check
A full **brainstorm → draft → final** run is dominated by the **final clip**:
- Drafts on **Kling 2.5 Turbo Pro** are ~$0.35 for 5s — iterate freely.
- A **Veo 3.1** people-final is ~$1 (5s/1080p, no audio) to ~$2 (with audio); **4K roughly doubles**
it. A **Seedance** camera-final is ~$0.26.
- Keep finals **short** (4–6s) and audio **off** unless asked. Quote the estimate before the final
render — see the SKILL.md "Costs & balance" section.
## Deferred / not wired here
- **End-frame / first-last-frame** transitions, **multi-shot** prompts, **lip-sync / dialogue**,
**LoRA-trained styles**, and **dedicated 3D / Gaussian-splat tour** generators — possible later
via `--arg`/new models, but not defaults. For a faithful walkthrough, a real shoot or a proper 3D
tour is the gold standard (see `motion-honesty-guide.md`).Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| canvas-designanthropics/skillsCreate beautiful visual art in .png and .pdf documents using design philosophy. You should use this skill when the user asks to create a poster, piece of art, design, or other static piece. Create original visual designs, never copying existing artists' work to avoid copyright violations. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| docxanthropics/skillsUse this skill whenever the user wants to create, read, edit, or manipulate Word documents (.docx files) or Word templates (.dotx files). Triggers include: any mention of 'Word doc', 'word document', '.docx', '.dotx', or requests to produce professional documents with formatting like tables of contents, headings, page numbers, or letterheads. Also use when extracting or reorganizing content from .docx or .dotx files, inserting or replacing images in documents, performing find-and-replace in Word files, working with tracked changes or comments, or converting content into a polished Word document. If the user asks for a 'report', 'memo', 'letter', 'template', or similar deliverable as a Word or .docx file, use this skill. Do NOT use for PDFs, spreadsheets, Google Docs, or general coding tasks unrelated to document generation. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| pdfanthropics/skillsUse this skill whenever the user wants to do anything with PDF files. This includes reading or extracting text/tables from PDFs, combining or merging multiple PDFs into one, splitting PDFs apart, rotating pages, adding watermarks, creating new PDFs, filling PDF forms, encrypting/decrypting PDFs, extracting images, and OCR on scanned PDFs to make them searchable. If the user mentions a .pdf file or asks to produce one, use this skill. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| pptxanthropics/skillsUse this skill any time a .pptx or .potx file is involved in any way — as input, output, or both. This includes: creating slide decks, pitch decks, or presentations; reading, parsing, or extracting text from any .pptx or .potx file (even if the extracted content will be used elsewhere, like in an email or summary); editing, modifying, or updating existing presentations; combining or splitting slide files; working with templates (.potx), layouts, speaker notes, or comments. Trigger whenever the user mentions "deck," "slides," "presentation," or references a .pptx or .potx filename, regardless of what they plan to do with the content afterward. If a .pptx or .potx file needs to be opened, created, or touched, use this skill. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |