dreamina-prompt-text2video

dreamina-prompt-text2video — 即梦 text-to-video prompts

Craft production-ready text-to-video prompts for 即梦 Dreamina video models (视频3.0 Pro, Doubao Seedance 2.0, Seedance 1.5, Seedance 1.0, 智能多帧).

When to use this skill

Use this skill when the user:

• Asks you to write a video generation prompt ("帮我写个视频提示词")
• Describes a scene they want as a moving image / video
• Wants to refine or optimize an existing video prompt
• Mentions keywords like: 文生视频, text2video, 视频生成, 视频提示词, 运镜, camera movement, Seedance
• Asks about how to describe motion, camera work, or temporal progression in a prompt

Do NOT use this skill for:

• Executing CLI commands to generate videos → use dreamina-cli-text2video
• Writing image-to-video prompts → use dreamina-prompt-image2video
• Writing text-to-image prompts → use dreamina-prompt-text2image

Model Version Guide

即梦 has multiple video model lines. The core writing approach differs by model:

Model	Formula	Key Difference
视频3.0 / 3.0 Pro	Subject + Action + Scene + Camera + Style + (Emotional Performance) + (Lighting)	Free-form natural language writing; supports shot cuts and creative effects
Doubao Seedance 2.0 series	Subject + Action/Motion + Spatial Background/Lighting/Style + Camera Movement/Sound Effects	Supports T2V/I2V/R2V/V2V; native audio + video joint generation; multimodal reference (image/audio/video); supports text generation (ad copy/subtitles/speech bubbles); supports video editing (element add/delete/modify/extension/track completion)
Seedance 1.5 Pro	Subject + Action + Scene + Camera + Style	Faster generation speed
Seedance 1.0 Pro / Pro Fast	Subject + Action + Scene + Camera + Style	Pro Fast: extreme speed
Image-to-Video (I2V)	Action + Camera + (Emotion/Lighting)	The image provides subject + scene; the prompt only controls motion
智能多帧	[Multi-frame images] + [duration per frame] + [camera movement prompt]	Multi-image driven single continuous shot; upload 2-10 frames; 1-6s per frame

Core Methodology

The video prompt formula extends the image formula with two critical dimensions: motion and camera.

For 视频3.0/3.0 Pro text-to-video:

主体 + 动作 + 场景 + 镜头 + 风格 + (情绪演绎) + (照明)

For 图生视频 (image-to-video):

动作 + 镜头 + (情绪/照明)

For Doubao Seedance 2.0 series text-to-video:

主体 + 动作/运动 + 空间背景/光影/风格 + 镜头调度/音效

The key difference from text-to-image prompting:

• Motion is mandatory: a video without motion is just a still image
• Camera adds storytelling: how the camera moves determines the viewer's emotional experience
• Duration shapes pacing: a 5-second clip needs different action density than a 10-second clip
• For 视频3.0 Pro specifically: free-form natural language writing, the core idea is "intuitively express the effect you want"
• For Doubao Seedance 2.0: supports multimodal reference (image/audio/video); in the prompt you can use "图片1""图片2" to refer to reference materials

How to use this skill

Step 1: Identify video scenario category

→ Load rules/video-category-table.md to match user's request to the right category and example file

Step 2: Load reference materials

Load video vocabulary — choose by what you need:

What is needed	Load file
Character action, natural motion, animal motion, mechanical movement	video-words/motion.md
Scene environment, video image quality, video style, atmosphere/mood, rhythm description	video-words/scene-style.md

Load camera movement references — choose by complexity:

What is needed	Load file
Push, pull, pan, tilt, tracking, crane up/down	camera-basic.md
Orbit, single continuous shot, Hitchcock, camera movement emotional mapping	camera-advanced.md

Load reference guides:

• references/jimeng-video-3.0-guide.md — when user mentions 视频3.0/3.0 Pro
• references/smart-multi-frame-guide.md — when user mentions 智能多帧/多帧/一镜到底

For color descriptions, cross-reference text2image's color-library/chinese-traditional.md or gugong-384-colors.md.

Step 3: Build the prompt

→ Load rules/video-core-methodology.md for component-by-component build guide + presentation format

Step 4: Apply video writing rules

→ Load rules/video-writing-rules.md for all 10 rules (motion, camera, duration, light, model-matching, differentiation, action layers)

Step 5: Validate

→ Load rules/video-validation-checklist.md and run through all checks

Step 6 (User assessment): Evaluate prompt output

When the user provides an existing prompt output and asks for evaluation ("评估""检测命中率""看看优化方向"): → Load references/evaluation-framework.md for the systematic 6-dimension assessment methodology

Gotchas

1. Action must be explicit — 即梦 cannot infer motion from static description. "一个人在街上" = static shot
2. Complex interactions fail — multi-person interactions + camera movement = distortion risk
3. Chinese camera terms work better — "镜头缓缓推近" > "dolly in slowly"
4. Seedance quality/speed tradeoff — seedance2.0 = highest quality/slowest; fast = faster/lower quality
5. Duration-to-action matching — 5s with 3-stage action = rushed. Match complexity to time
6. First-time web authorization — some models require browser auth before first use
7. Character consistency not guaranteed — same face across segments not reliable
8. Camera + complex motion = risk — prioritize one over the other
9. Model-prompt mismatch is the #1 output killer (Rule 8) — long descriptive prose → 视频3.0 Pro; structured short sentences → Seedance 2.0. Mismatching causes detail loss and poor generation. This is the single most common error in real usage.
10. Action layer count by model (Rule 10) — Seedance 2.0 can't handle 4+ action layers reliably. Cap at 3. 视频3.0 Pro can handle 4-5.
11. Multi-scheme differentiation is non-optional (Rule 9) — When providing 2+ alternatives, verify visual distinction. At least 2 dimensions must differ (运镜/动作密度/景别/节奏/场景). Similar schemes waste user's time.
12. Light must change, not just exist (Rule 5) — "洒入" is a static snapshot. Use "缓缓流动""逐渐变亮""光影移动" to encode time passage.
13. ⚠️ Never write hex color codes in video prompts — Same as image prompts: #RRGGBB values get rendered as text in the video frames, NOT as color instructions. Use Chinese color names only

Available Resources

Resource	Description	When to Load
rules/video-category-table.md	12 video scenario categories	Step 1
rules/video-core-methodology.md	Component-by-component build guide	Step 3
rules/video-writing-rules.md	10 video-specific writing rules	Step 4
rules/video-validation-checklist.md	Basic + Advanced + Multi-Option Validation Checklist	Step 5
video-words/motion.md	Motion vocabulary library	When writing motion descriptions
video-words/scene-style.md	Scene/style/vocabulary library	When describing scenes
camera-basic.md	7 basic camera movements	Basic camera work needed
camera-advanced.md	7 compound moves + emotion mapping	Advanced camera work
references/jimeng-video-3.0-guide.md	视频3.0/3.0 Pro: 8 dimensions	User mentions 视频3.0
references/smart-multi-frame-guide.md	智能多帧: multiple images in one continuous shot	User mentions 智能多帧/多帧
references/evaluation-framework.md	6-dimension evaluation framework: vocabulary hit rate/rule compliance/formula completeness/model matching/distinctiveness/validation	User asks to assess/evaluate skill output

When to use（When to use）

Load this skill when the user needs execute generation or write operations after clarifying inputs, budget, and delivery constraints . First extract the goal, inputs, constraints, delivery format, and acceptance criteria from the request; the description summary is: Use when the user wants to write, refine, or optimize a video-generation prompt for 即梦 (Dreamina/Jimeng) video models (视频3.0 Pro, Doubao Seedance 2.0, Seedance 1.5/1.0, 智能多帧); mentions 文生视频, text2video, 视频提示词, 运镜, camera movement, 生成一段...的视频, 帮我写个视频; describes a scene they want as a video; or supplies a rough scene description. Covers 12 scenario categories with 35 annotated examples, adds explicit motion description and camera direction beyond image prompts, and includes a motion/camera word library plus 7 basic and 6 advanced compound camera techniques with emotional effect mappings..

Rules

• Read before write: first confirm the current state and actual capabilities, then perform actions that change external state.
• Least privilege: use only the files, tools, accounts, and network scope required to complete the current step.
• Evidence first: when run results, resource IDs, versions, hashes, or test outputs are missing, explicitly mark them as NOT_VERIFIED。
• Idempotency first: retain request identifiers and stage status; when results are unclear, query first rather than blindly retrying.
• Privacy and security: logs, examples, receipts, and error messages must not contain tokens, cookies, keys, or personally sensitive data.

Workflow

Step 1: Clarify Intent

Confirm whether this skill matches the goal; if it is only an adjacent need, hand it off to a more precise skill.

Step 2: Run Prechecks

Validate inputs, model/tool capabilities, output paths, budget caps, and approval status; if any critical condition is unknown, stop at the read-only stage.

Step 3: Form a Plan

List the tools to be called, the objects that will change, the success criteria, and the safe exit method after failure.

Step 4: Execute Actions

Execute once as approved and record the request identifier; query ambiguous results first instead of resubmitting; retain traceable status or receipts for every external call.

Step 5: Verify Delivery

Verify artifact existence, format, hash/identifier, cost status, and quality gates, and state facts, inferences, and unverified items separately.

Validation checklist

• The skill trigger conditions match the user's intent, with no adjacent task misrouted to this skill.
• Inputs, target objects, versions, and output locations are all clear, and no guessed values are used in place of required values.
• All writes, payments, publishes, or irreversible actions are within the scope of user authorization.
• Results have been verified with independent checks; only "command succeeded" or "file exists" does not count as complete acceptance.
• The output includes actual evidence, failed/skipped items, remaining risks, and executable next steps.

Gotchas

1. Treating a plan as a result: A document or prompt does not equal actual execution; the actual level of execution must be indicated.
2. Retrying on error: A timeout or lost response may already have produced remote state; query first before deciding whether to retry.
3. Implicit scope expansion: Batch, full-scale, publish, overwrite, and payment operations are not a natural extension of ordinary reads and writes.
4. Version drift: When referencing external resources, record the version, tag, or commit; do not treat a mutable branch as release evidence.
5. Stale evidence: Caches, old screenshots, and historical tests cannot prove the current environment; refresh key evidence before delivery.

Non-applicability and boundaries

Payment, publishing, overwriting, uploading, or external writes must use the explicit authorization of the current task; do not automatically expand the number of times or the budget. If the request requires another skill, do not copy its body text; hand off by skill name and preserve the current task context.

Progressive disclosure

• When input/output, state, and authorization points need to be determined, read references/workflow-contract.md。
• When a pre-delivery self-check is needed, read references/validation-checklist.md。
• When encountering timeouts, partial success, or recovery scenarios, read references/error-recovery.md。
• For the first run, refusal to exceed authorization, and failure recovery, refer respectively to examples/happy-path.md、examples/boundary-refusal.md、examples/failure-recovery.md。