Awesome Seedance 2 Prompts: Curated Open-Source Prompt Library
A curated, open-source library of Seedance 2 prompts — copy-paste recipes for product spins, image-to-video motion, character shots, and cinematic camera moves. Maintained on GitHub, paired with the Martini canvas.
Key takeaways
- awesome-seedance-2-prompt is an open-source GitHub repo of curated, production-tested prompts for ByteDance Seedance 2 (Pro, Lite, Omni) — copy any recipe and paste it straight into a Seedance 2 video node on Martini.
- The library is organized by shot intent, not by adjective: hero spins, product reveals, image-to-video motion, character close-ups, environmental establishes, and continuation seeds.
- Each prompt follows the same single-shot grammar — subject + action + camera move + lens + lighting + atmosphere — so the recipes A/B cleanly against one another and stay legible after copy-paste.
- Pair the repo with the Martini canvas: wire one reference image into multiple Seedance 2 nodes, fan the same prompt across Pro / Lite / Omni variants, and keep every take in the version tray.
- Repo lives at github.com/MartiniArt/awesome-seedance-2-prompt — open a PR to add a recipe you have tested in production, or fork it to build a private prompt pack for your team.
Why a curated Seedance 2 prompt library exists
Seedance 2 rewards prompts that read like a single line in a shot list, and punishes prompts that read like a paragraph of mood. Most public prompt collections still optimize for stylistic adjectives — "cinematic, hyper-realistic, beautiful" — which waste the model's prompt budget on words it does not act on. The awesome-seedance-2-prompt repo exists to collect prompts that have actually shipped to production, organized by what the shot needs to do, not by what it looks like in isolation.
Each entry in the repo answers a specific question a creator already has: "How do I get a label-locked hero spin?" "How do I animate a still without identity drift?" "How do I extend past the model's native length cap without a visible cut?" The recipes are short on flourish and long on camera grammar, lighting cues, and the one or two motion verbs Seedance 2 reliably executes. Copy any recipe, paste it into a Seedance 2 video node on the Martini canvas, and you are within one or two iterations of the result the recipe promises.
The library is intentionally model-specific. A prompt that performs on Kling 3 Avatar will often underperform on Seedance 2 Pro, and vice versa. Treating Seedance 2 as its own grammar — rather than a generic "video model" — is the single biggest unlock once teams move past their first week with the model.
How the recipes are organized
The repo is split into folders that map to shot intent. Hero spins, product reveals, and label-locked motion live under product/. Image-to-video recipes — push-ins, parallax, orbits, drifts — live under image-to-video/. Character close-ups (where Seedance 2 holds identity well for a few seconds before lip-sync limits kick in) live under character/. Environmental establishes, weather, and crowd shots live under environment/. Continuation seeds — prompts written specifically to be the second clip in a chain, stitched off the last frame of the first clip — live under continuation/.
Within each folder, recipes are named by the camera move first, then the lighting condition, then the variant of Seedance 2 that produced the cleanest take in QA. A filename like product/orbit-soft-rim-pro.md tells you the shot is an orbit, lit with a soft rim, and benchmarked against Seedance 2 Pro. The body of every recipe contains the prompt text, the recommended Seedance 2 variant (Pro / Lite / Omni), the reference image you should wire upstream of the node, and a notes section calling out edge cases the prompt does not handle.
Recipes that did not survive QA — prompts that produced visible identity drift, rubbery transparency, or motion that compressed multiple actions into a half-second blur — are tracked in a graveyard/ folder with a one-line note about why they failed. This is more useful than deleting them: it tells future contributors which prompt shapes Seedance 2 punishes, so they can avoid re-discovering the same dead ends.
Using the repo on the Martini canvas
The recipes are designed to drop straight into a Seedance 2 video node on Martini. The canvas pattern is image-first: generate or upload the still you want, wire it into the Seedance 2 node, paste the recipe, and iterate the prompt while the image stays pinned. The version tray on the canvas keeps every take, which means you can A/B four recipes from the repo against the same reference image without losing any of them.
For multi-shot work, duplicate the Seedance 2 node and re-wire each duplicate to the same character or product image. Paste a different recipe into each duplicate — different camera moves, different lighting cues, different micro-actions. Drop an NLE export node downstream of all of them and the canvas assembles the takes in the order you wire them. This is the workflow the repo is optimized for: many recipes, one anchor image, one canvas, one export.
When you find a recipe that consistently produces the result you want, fork the repo and save the recipe alongside the seed and the reference image hash in a private prompt pack for your team. The public repo stays curated and small; private forks let you build a brand-specific dialect of Seedance 2 prompts that ship faster than re-iterating from scratch on every campaign.
Recipes worth bookmarking on day one
Slow orbit hero spin (label-locked) — the signature ecommerce hero shot. The repo's recipe wires a brand-approved product still upstream and prompts only for camera and light: "Slow 360-degree orbit around the referenced product, 8 seconds, even rotation, soft front-key plus rim light, shallow depth of field, label remains crisp and centered, subtle floor reflection, locked focal length, no zoom, no cuts." Pair with Seedance 2 Pro for finished work, Lite for variant sweeps.
Slow push-in (subtle drift) — the default cinematic motion for any still. "Slow camera push-in toward subject in referenced image, 5 seconds, gentle parallax in foreground vs background, locked subject, subtle breathing, no rotation, no zoom-out, ends 1.2x closer than start." Works on portraits, products, and environments. Default to Seedance 2 Pro; switch to Kling 3 only if the image has heavy depth-of-field already baked in.
Character identity-locked turn — the recipe that justifies running Seedance 2 Omni instead of Pro. Wire the same character reference image into the Omni node and prompt: "Subject begins still facing camera, slow head turn 30 degrees to camera left, micro smile at the end, locked focal length, soft window light from frame right, 4 seconds." Repeat across multiple Omni nodes with different head turns and you have a sequence that holds identity without fine-tuning.
How to contribute back
Open a pull request against awesome-seedance-2-prompt with a recipe you have tested in production. The repo expects four things in the PR body: the prompt text, the Seedance 2 variant you ran it on (Pro / Lite / Omni), a sample output (any video clip is fine — Twitter post, internal upload, embedded GIF), and a short note on the edge cases the prompt does not handle. PRs without a sample output are not merged, because the repo's value is exactly that every recipe has actually shipped.
Recipes that fail QA are not rejected — they are moved to graveyard/ with a note. This keeps the repo honest about which prompt shapes Seedance 2 actually rewards. If you discover a recipe that worked on Seedance 1 but no longer works on Seedance 2 (this happens often with multi-action prompts), file a graveyard PR and link the new replacement.
The repo is paired with the Martini canvas but not gated to it. You can copy any recipe into ByteDance's direct API, into BytePlus, or into any third-party tool that exposes Seedance 2. The recipes are written in plain English and do not depend on Martini-specific node syntax — only the multi-node fan-out workflow assumes a canvas.
Workflow example
A typical workflow using awesome-seedance-2-prompt on Martini for a campaign sprint: pick three recipes from product/ that match the brief (one orbit, one push-in, one rim-light reveal), generate the hero still in Nano Banana 2 or GPT Image 2, lock the seed, and wire the still into three Seedance 2 Pro nodes side by side on the canvas. Paste one recipe into each node verbatim. Render two takes per node, pick the strongest variant from each, and wire the chosen takes into an NLE export node downstream. The whole loop — from open the canvas to finished sequence — takes under an hour for a creator who has the repo bookmarked.
Recommended models
Recommended features
Related models and tools
Related how-to guides
Related reading
Seedance 2 Handbook: Variants, Best Workflows, and How to Use It on Martini
Hands-on guide to Seedance 2 — variants, strengths, and the production workflows it fits on Martini's canvas.
How to Turn an Image Into Video With AI
End-to-end image-to-video workflow on Martini — model choice, motion control, and chaining shots.
GPT Image 2 Guide: Workflows, Strengths, and Where It Fits on Martini
How GPT Image 2 fits product, text, and reference image workflows on Martini's multi-model canvas.
Open-source resources
Frequently asked questions
- Where do I find the awesome-seedance-2-prompt repo?
- The repo is hosted at github.com/MartiniArt/awesome-seedance-2-prompt. Star it to follow updates, fork it to build a private team prompt pack, or open a pull request to contribute a recipe you have tested in production.
- Do the prompts only work on Martini?
- No. The recipes are written in plain English and run on any Seedance 2 surface — ByteDance direct API, BytePlus, or any third-party tool. The Martini-specific value is the multi-node fan-out workflow on the canvas, where one reference image feeds many Seedance 2 nodes in parallel.
- How is this different from the Seedance 2 Handbook on the blog?
- The handbook explains how Seedance 2 thinks — variants, strengths, prompt grammar, when to swap to Kling or Veo. The awesome-seedance-2-prompt repo is the executable version: every recipe is copy-paste ready, organized by shot intent, and paired with the Seedance 2 variant that produced the best take in QA.
- Can I contribute a prompt I built for Seedance 1?
- Probably, but test it on Seedance 2 first. Multi-action prompts that worked on Seedance 1 often compress on Seedance 2 — file those as graveyard entries instead. Recipes that survive QA on Seedance 2 are merged with a sample output attached.
- Why is the repo paired with Martini specifically?
- Martini's canvas lets you wire one reference image into many Seedance 2 nodes in parallel and keep every take in a version tray. That is the workflow the recipes assume. You can still use the recipes on a single-node tool — you will just iterate sequentially instead of fanning out.
Ready to try it on the canvas?
Open Martini and fan your prompt across every frontier model in one workflow.