← ClaudeAtlas

seedance-prompt-forgelisted

Write, audit, and repair prompts for Seedance 2.5 video generation. Use whenever the user mentions Seedance, Dreamina, or a Seedance/Dreamina video prompt — and also when they want to fix a video generation that came out wrong, even if they never name the model. Covers text-to-video; reference-driven generation with images, videos, or audio; 30-second multi-stage videos; video editing; forward or backward extension; first/last-frame and multi-keyframe generation; storyboard grids; coarse or fine blockout re-rendering; one-click video; and seamless transitions. Symptoms it fixes: swapped subjects, identity or clothing drift, backgrounds or people leaking from references, all references appearing at once, missing or rushed events, broken continuity across a cut, dialogue in the wrong language. Trigger phrases: "Seedance prompt", "Dreamina prompt", "optimize my video prompt", "why did my video come out wrong", "fix this generation".
silentbuilds/seedance-prompt-forge · ★ 3 · AI & Automation · score 77
Install: claude install-skill silentbuilds/seedance-prompt-forge
# Seedance Prompt Forge Turn an idea, draft prompt, reference set, or failed generation into a copy-ready Seedance 2.5 prompt. Prompts are flexible natural language built from optional components; every reference material gets an explicit role. Source: the official Dreamina Seedance 2.5 prompt guide. Do not apply these templates to other Seedance versions without checking that version's own guidance. Platform behaviours described here (locked parameters, multimodal reference mode) are documented for Dreamina; verify before assuming an API surface behaves identically. ## Pick a route | The user gives you | Route | |---|---| | An idea, brief, or reference set | **A — Author** | | A generation that came out wrong (with or without the prompt) | **B — Diagnose** | | An existing prompt, no output yet | **C — Audit** | Routes B and C both end by re-entering Route A step 3. ## Core structure Components, in priority order: **Subject + Action/Event + Scene/Environment + Visual Style + Camera Movement/Cut + Audio** - Lead with what you want generated: the subject and its primary action or event. - Add only the components that matter; omit the rest. - Keep generation parameters (duration, aspect ratio, resolution, audio toggle) out of the prompt. They are set on the generation page or through the API. ```text <Subject> performs <primary action or event> in <scene and environment>. The visuals feature <visual style>. Use <shot size, camera angle, camera movement, or cuts>. Aud