dreamina-prompt-image2video

dreamina-prompt-image2video — 即梦 image-to-video prompts

Craft prompts for animating reference images into videos using Dreamina's 4 video modes.

When to use this skill

Use this skill when the user:

• Provides image(s) and wants to generate a video from them ("把这张图做成视频")
• Mentions 图生视频, image2video, 图片转视频, make the image move
• Describes a specific sub-mode: 单图生视频, 首尾帧, 多帧故事, 全能参考
• Has multiple images and wants to create a narrative video
• Asks how to describe motion from a static reference image

Do NOT use for:

• Text-to-video (no reference images) → use dreamina-prompt-text2video
• CLI execution → use dreamina-cli-image2video
• Image-to-image editing (output is still an image) → use dreamina-prompt-image2image

Core Methodology

Image-to-video prompting is fundamentally incremental: the reference image already provides the visual content. The prompt only needs to describe what MOVES and CHANGES.

[从哪里开始] + [什么在动/怎么动] + [运镜方式] + [环境/光影变化] + [时长/节奏] + [风格一致]

The I2V Golden Rule

Don't describe what's already in the image. Only describe what happens next.

• ✗ "一个穿红裙子的女孩站在花园里，阳光灿烂..." — the image already shows this
• ✓ "女孩缓缓转身面向镜头，红裙随风飘动，镜头慢慢推近" — this is what happens NEXT

4 Sub-Modes Overview

#	Mode	Input	Core Prompt Focus	Example File
1	单图生视频	1 image	"Starting from this image, what movement happens next"	examples/single-image.md
2	首尾帧	2 images	"How to transition from image A to image B"	examples/first-last-frame.md
3	多帧故事	2-20 images	"The causal relationship and passage of time between frames"	examples/multi-frame.md
4	全能参考	images+video+audio	"Synthesize all reference materials to generate the target video"	examples/multimodal-ref.md

How to use this skill

Step 1: Identify the sub-mode

Ask the user or infer from context:

• Has 1 image → Sub-mode 1 (单图生视频)
• Has 2 images described as start/end → Sub-mode 2 (首尾帧)
• Has 2+ images forming a sequence/storyboard → Sub-mode 3 (多帧故事)
• Has mixed media (images + video + audio) → Sub-mode 4 (全能参考)

Step 2: Load reference materials

Load the matched example file from examples/ for sub-mode-specific patterns.

Load references/mode-library.md for detailed templates per sub-mode.

Load references/word-library.md for I2V-specific motion and coordination vocabulary.

For camera movement vocabulary, cross-reference the text2video skill's camera-library.md — it covers 7 basic + 6 advanced camera movements with emotional mappings, not duplicated here.

Step 3: Build the prompt (incremental description)

For sub-mode 1 (单图生视频):

1. Starting point (从哪里开始): Acknowledge the reference image — "从这张画面开始"
2. What moves (什么在动): The MOST important component. Be specific about WHICH elements move and HOW. "头发在微风中飘动，水面的波纹缓缓扩散" — describe only what changes
3. What stays still (什么不动): Explicitly state static elements — "建筑和远山保持完全静止"
4. Camera movement (运镜): How the camera moves through the scene
5. Duration/pacing (时长/节奏): 4-15 seconds, match complexity
6. Style consistency (风格一致): The output must match the reference image's style

For sub-mode 2 (首尾帧):

1. Describe the TRANSITION from frame A to frame B — "花瓣从含苞逐渐绽放至完全盛开"
2. Describe the pacing of the transition — "前半段缓慢积蓄，后半段加速绽放"
3. Camera can remain static or follow the transformation

For sub-mode 3 (多帧故事):

1. Describe the causal relationship between frames — "因为风吹过，所以花瓣飘落"
2. Describe the passage of time — "从清晨到正午，光影在建筑表面缓缓移动"
3. Camera work should connect frames narratively

For sub-mode 4 (全能参考):

1. Assign clear roles to each reference — "图一提供人物特征，图二提供场景风格"
2. Describe the synthesis — how elements from different references interact

Step 4: Apply I2V-specific rules

1. Incremental only: Don't repeat what the reference image already shows
2. Motion specificity: "她的头发在微风中轻轻飘动" not "有风"
3. Stillness is intentional: Explicitly state what does NOT move
4. Camera respects the reference: Camera movement is constrained by the reference composition
5. Style must match: The output must be stylistically consistent with the reference

Step 5: Validate against checklist

Run through the Validation Checklist. Every item must pass before presenting the prompt to the user.

Step 6: Present the result

Always present: sub-mode label + motion description + complete prompt + duration + camera explanation.

Writing Rules for Image-to-Video

Rule 1: Incremental description — only write what changes

The reference image IS the visual. Don't describe its static content in the prompt. Focus 100% on motion and change.

Rule 2: Explicit stillness is as important as motion

Without explicit stillness constraints, AI tends to animate everything. Write "保持建筑完全静止""只有水在流动，其他一切不动".

Rule 3: Match sub-mode to prompt structure

Each sub-mode needs a different prompt structure. Don't use a single-image prompt for a frames2video task.

Rule 4: Camera movement respects the reference

Unlike text-to-video where camera can go anywhere, I2V camera movement is bound by the reference image's composition. Camera descriptions should work WITH the existing composition.

Rule 5: Duration scales with complexity

Sub-mode	Motion Complexity	Suggested Duration
单图	1-2 moving elements	5-8s
首尾帧	1 complete transformation	8-12s
多帧	Multi-stage narrative	10-15s
全能参考	Complex multi-source synthesis	8-12s

Validation Checklist

• Prompt describes only what MOVES/CHANGES (not static content from the reference image)
• Specific elements are named as moving + specific motion verbs used
• Elements that should stay still are explicitly listed
• Camera movement is described (not just named — explain the movement logic)
• Duration matches motion complexity
• Style consistency with reference image is addressed
• Correct sub-mode is identified and communicated

Gotchas

1. Don't describe the reference image — the #1 I2V mistake. "A girl in a red dress standing in a garden" is useless if the reference already shows this. Write what HAPPENS: "The girl slowly turns..."
2. Everything moves unless you stop it — AI's default is to animate everything. Use "everything else remains completely still" to lock down static elements
3. Camera terms need explanation, not just naming — 只写"推镜头"不够，需要写"镜头缓缓向前推进，逐渐聚焦人物的面部表情"
4. Resolution varies by sub-mode — Seedance 2.0 = 720p, legacy 3.5pro frames2video = 1080p
5. Multiframe limit — 2-20 images for multi-frame story mode. More images = better consistency but slower generation
6. Multimodal limits — max 9 images + 3 videos + 3 audio clips
7. Style drift possible — output video may not perfectly match reference image style; use "保持参考图的风格和色调完全一致" to minimize drift

Available Resources

Resource	Description	When to Load
references/mode-library.md	4 sub-mode templates with prompt structures	When building any I2V prompt
references/word-library.md	I2V-specific motion, stillness, coordination vocabulary	When building any I2V prompt
examples/single-image.md	Single-image-to-video examples	User has 1 reference image
examples/first-last-frame.md	First-and-last-frame examples	User has 2 images as start/end
examples/multi-frame.md	Multi-frame story examples	User has 2+ sequential images
examples/multimodal-ref.md	Omni-reference examples	User has mixed media references

When to use

Load this skill when the user needs to perform a generation or write operation after clarifying input, budget, and delivery constraints . First extract the goal, input, constraints, delivery format, and acceptance criteria from the request; the description summary is: "Provides comprehensive guidance for crafting image-to-video (图生视频) prompts for 即梦 (Dreamina/Jimeng) Seedance 2.0. Image-to-video prompts are incremental — the reference image provides the visual content, the prompt only describes what moves and changes. This skill covers 4 sub-modes: single image→video (单图生视频), first-last frame transition (首尾帧), multi-frame storyboard narrative (多帧故事), and multi-modal reference compositing (全能参考). Use when the user has reference image(s) and wants to animate them into a video; mentions "图生视频", "image2video", "图片转视频", "让这张图动起来", "首尾帧", "故事板", "多帧故事", "frames2video"; or provides images and asks to generate a video from them. This skill covers 4 sub-mode templates with 18 annotated examples, a mode decision tree, and I2V-specific incremental description vocabulary. Always use this skill when the user wants to animate reference images into video using 即梦.".

Rules

• Read before write: first confirm the current state and real capabilities, then perform actions that will change external state.
• Least privilege: only use 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 phase state; 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: Perform pre-checks

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

Step 3: Form a plan

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

Step 4: Execute actions

Execute once approved and record request identifiers; for ambiguous results, query first instead of resubmitting; retain correlatable 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, and no adjacent task was misrouted to this skill.
• Inputs, target objects, versions, and output locations are all clear, and no guessed values were used in place of required values.
• All write, paid, publish, or irreversible actions are within the scope of user authorization.
• The results have been verified with independent checks; merely 'the command succeeded' or 'the file exists' does not count as complete acceptance.
• The output includes actual evidence, failed/skipped items, remaining risks, and actionable next steps.

Gotchas

1. Treating plans as results: Documentation or prompts are not the same as actual execution; the actual execution level must be indicated.
2. Error retries: A timeout or lost response may already have produced remote state; query first before deciding whether to retry.
3. Implicit scope expansion: Batch, full-volume, publish, overwrite, and paid 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

Paid, publish, overwrite, upload, or external write operations must use the explicit authorization of the current task; do not automatically expand the number of times or the budget. If the request requires a different skill, do not copy its body text; hand off by skill name and preserve the current task context.

Progressive disclosure

• When you need to determine inputs/outputs, state, and authorization points, read references/workflow-contract.md。
• When you need a pre-delivery self-check, read references/validation-checklist.md。
• When encountering timeouts, partial success, or recovery scenarios, read references/error-recovery.md。
• For first runs, refusal of privilege escalation, and failure recovery, refer respectively to examples/happy-path.md、examples/boundary-refusal.md、examples/failure-recovery.md。