seedance-2-0 Checked

| Surface | Env | OpenMontage tool | Status | Notes |

|---|---|---|---|---|

| **fal.ai** (primary) | FAL_KEY | seedance_video | ✅ wrapped | Model IDs below. Supports T2V, I2V, reference-to-video; standard and fast variants. Default in OpenMontage. |

| **Replicate** | REPLICATE_API_TOKEN | seedance_replicate | ✅ wrapped | bytedance/seedance-2.0 + bytedance/seedance-2.0-fast. Standard Replicate prediction API. |

| **Runway** | RUNWAY_API_KEY | runway_video (model: seedance_2.0) | ✅ wrapped | Third-party Seedance 2.0 model inside Runway. **Unlimited/Enterprise plans, non-US only**. Selected via model param. |

| **Higgsfield** | HIGGSFIELD_API_KEY + _SECRET | higgsfield_video (model: seedance_2.0) | ✅ wrapped | Seedance 2.0 is the default model on this tool. Emphasis on character identity + long-form chaining. |

| **HeyGen** | HEYGEN_API_KEY | heygen_video (1.x only) + TODO | ⚠️ 1.x only | The seedance_pro / seedance_lite workflow provider strings on HeyGen map to Seedance 1.x. 2.0 access flows through Video Agent / Avatar Shots endpoints — a separate seedance_heygen tool is on the roadmap. |

| **BytePlus ModelArk / Volcengine** | BytePlus token | not wrapped | 🔜 roadmap | Direct from ByteDance. Pro ~$0.15 / 5 s, Lite ~$0.010/s. Token-based. |

| **Freepik** | Freepik token | not wrapped | 🔜 roadmap | POST /v1/ai/image-to-video/seedance-pro-1080p for 1080p I2V |

| **Pollo / PiAPI / Atlas Cloud / AIMLAPI** | various | not wrapped | 🔜 roadmap | Aggregators resell fal.ai or ByteDance endpoints |

### fal.ai model IDs (used by seedance_video)

bytedance/seedance-2.0/text-to-video
bytedance/seedance-2.0/image-to-video
bytedance/seedance-2.0/reference-to-video        # 9 img + 3 vid + 3 audio
bytedance/seedance-2.0/fast/text-to-video
bytedance/seedance-2.0/fast/image-to-video
bytedance/seedance-2.0/fast/reference-to-video

Pricing (fal.ai, 720p): standard $0.3034 / s (T2V), $0.3024 / s (I2V). Fast $0.2419 / s across endpoints.

The fast variant trades some camera/motion fidelity for latency and cost — do **not** route slow-mo, multi-shot, or dolly-heavy prompts to fast on the first try.

Calling Seedance 2.0 inside OpenMontage

Always go through video_selector with preferred_provider="seedance" (or let the scoring engine pick it):

from tools.tool_registry import registry
registry.ensure_discovered()
selector = registry.get("video_selector")
result = selector.execute({
    "prompt": PROMPT,
    "preferred_provider": "seedance",
    "operation": "text_to_video",       # or image_to_video / reference_to_video
    "aspect_ratio": "21:9",             # 21:9 / 16:9 / 9:16 / 4:3 / 1:1 / 3:4
    "duration": "10",                   # auto / 4..15
    "resolution": "720p",               # 480p / 720p
    "output_path": "projects/<proj>/assets/video/clip_01.mp4",
})

Direct call to the provider tool (only when you must bypass the selector):

seedance = registry.get("seedance_video")
seedance.execute({
    "prompt": PROMPT,
    "model_variant": "standard",   # "standard" or "fast"
    "operation": "text_to_video",
    "aspect_ratio": "21:9",
    "duration": "10",
    "resolution": "720p",
    "generate_audio": True,
    "seed": 12345,                 # optional, for reproducible variations
    "output_path": "...",
})

Prompt structure — The Higgsfield Methodology (canonical as of 2026)

**CRITICAL: Open every prompt with a shot-structure declaration.** Seedance rewards prompts that declare format upfront before any creative description. This is the single biggest quality lever.

### Opener templates (copy one verbatim, then extend)

**For action/combat/multi-shot (highest-performing format):**

Montage, multi-shot Hollywood action, don't use one camera angle or single cut, cinematic lighting, photorealistic, 35mm film quality, ARRI ALEXA aesthetic, heavy film grain, sharp but imperfect focus, motion blur on fast actions, halation on highlights, soft highlight rolloff, wide-angle lens with strong distortion, subtle chromatic aberration near frame edges, no 3D, no cartoon, no VFX aesthetic.

**For single-POV continuous shots (orbs, walkthrough):**

Single continuous shot, first-person POV perspective, the camera IS [his/her] eyes, hyper-chaotic handheld motion, completely unstabilized, violent raw human movement, constant micro-jitters, aggressive head swings, abrupt jerks, frequent over-rotation, no smoothness at all, no cuts, no zoom, 35mm film, photorealistic.

**For locked-POV reaction scenes:**

One continuous shot, POV [setting] perspective, no cuts, no zoom, natural head movement, photorealistic, 35mm film grain.

### Body structure (after the opener)

1. **Environment/location** — sensory detail (wet asphalt, sodium lamps, neon bleed, rain particulates, volumetric haze)

2. **Character block** — with reference tags and identity-lock language (see Reference-to-video below)

3. **Enemy/secondary character block** — same detail level

4. **Beat-by-beat choreography** with TEMPORAL MARKERS: 0–3s: … 3–6s: … 6–10s: …

5. **VFX inline in brackets:** [VFX: branching white-blue electric arcs pulsing along forearms, sparks jumping between fingers]

6. **Slow-motion markers:** write RAMPS TO SLOW MOTION before the impact beat, SNAPS BACK TO REAL TIME on resume

7. **Sound design block:** either no music, only raw SFX or explicit SFX sequence. Music language stays textural.

### Combat vocabulary (proven to hit)

• snaps forward, lunges, sprints, weaves, chambers, drives, pivots, redirects, ducks, slips
• explodes outward, devastating, raw force, kinetic, overload, compresses, erupts, fractures, ripples
• Avoid soft verbs: attacks, hits, fights — these read generic and Seedance underdelivers on them

### Camera behavior — state what it IS and ISN'T doing

Seedance misfires when camera intent is ambiguous. Always explicitly negate what you don't want:

• no cuts (for continuou