Write director-grade prompts for ByteDance Seedance 2.5 (audio–video joint generation). Part of https://github.com/xianyu110/awesome-seedance-2.5 (gallery + skills).
When to use
- User asks for Seedance 2.5 / Seedance2.5 / Dreamina Seedance /
dreamina-seedance-2-5-*prompts - Need a shot timeline, character lock, DV look, or one-take story
- Converting a vague idea into a paste-ready Seedance brief
Model facts (keep accurate)
- Official: seedance2_5
- Common API / product names: Seedance 2.5, Dreamina Seedance 2.5
- Strength: ~4–30s single-pass A/V, multimodal refs, one-take storytelling
- Not Seedance 2.0 / Mini / Fast unless the user explicitly asks for those
Prompt skeleton (always)
MODEL: Seedance 2.5
DURATION: <Ns, usually 10–30>
ASPECT: <9:16 | 16:9 | 1:1>
STYLE LOCK: <photoreal / DV home-video / cinematic CG / brand film / …>
AUDIO: <dialogue? BGM? ambience-only? no subs?>
[CHARACTER CARDS]
@imageN = <name> — face, hair, body, wardrobe, props. Keep identity locked.
[ENVIRONMENT]
Place, time of day, weather, light.
[CAMERA]
Lens feel, move (handheld / locked / one continuous track), cut rhythm.
[ACTION TIMELINE]
00:00–00:03 …
00:03–00:08 …
…
[NEGATIVE / NOT]
No watermarks, no fake UI, no beauty filter (unless requested), no random text overlays.
Camera grammar cheat-sheet
| Look | Cue words |
|---|---|
| DV / home video | early-2000s DV, handheld shake, AF hunting, soft digital noise, clumsy zoom, no BGM |
| Cinematic action | anamorphic feel, motivated camera, impact frames, practical light, grit |
| Brand film | night plates, controlled motion, logo-safe negative space, premium grade |
| Fantasy duel | dual character cards, scale contrast, staged beats, score layers |
| POV explore | first-person, natural head sway, underwater rays / bubbles / ambience |
Audio layers
Prefer layered ambience over wall-to-wall music unless the user wants an MV. Explicitly say: dialogue / no dialogue, subs / no subs, SFX hits only on contact.
Identity lock (I2V)
If the user has a reference image:
- Name it
@图片1/@image1 - Restate face, hair, outfit, accessories
- Forbid wardrobe changes mid-clip unless scripted
- Prefer "image holds identity; video model owns physics"
Delivery
- Ask only missing critical inputs (duration, aspect, refs, language of dialogue)
- Output one paste-ready prompt block
- Offer a short variant (15s) and a long variant (30s+) when useful
- Link related case demos from the gallery when teaching a pattern
Related skills in this repo
seedance-25-cinematic-action,seedance-25-dv-day-in-life,seedance-25-boxing-handheldseedance-25-mech-backpack,seedance-25-cat-knight-duel,seedance-25-ocean-kingdom-povseedance-25-character-lock-i2v,seedance-25-fight-previz-pipelineseedance-25-brand-commercial,seedance-25-music-mv-longform,seedance-25-gallery-index
Not for
Seedance 2.0-only workflows, guaranteeing platform recommendation, or claiming private/unpublished prompts as public.