This independent Seedance prompt generator helps you turn a scene into an editable production brief. Choose Text to video, Image to video, Reference to video or Video edit before writing. Each operation needs different starting information.
The generator formats text locally. It does not connect a provider account, attach files or render a clip. Copy the brief, attach the actual references in your chosen service and set its model controls separately. Below, a recorded Seedance result shows why a specific prompt still needs a complete output review.
Choose the operation before writing the prompt
| Operation | Start with | Direct next |
|---|---|---|
| Text to video | A subject and scene | The observable action and camera |
| Image to video | An approved starting composition | What moves and what stays recognizable |
| Reference to video | Assets with separate roles | The new scene, action and preservation requirements |
| Video edit | An existing clip worth keeping | One permitted change and the rest to preserve |
Higgsfield's current Seedance help page separates generation from the Edit operation and lists different family versions. Read the complete model and route in the picker. This builder's operation labels organize your brief; they do not certify that every listed model exposes every route.
Use the latest Seedance generation and Edit guide for version-specific checks. Use the Standard, Fast and Mini comparison when the decision is which earlier variant to evaluate. Keep that model decision separate from the scene writing below.
Use the free Seedance prompt generator
- Choose the operation and read its example.
- Replace the scene facts and reference roles with your own.
- Describe one observable action, one camera direction and a readable ending.
- Check what should remain unchanged.
- Copy the prompt or download text or JSON.
- Attach actual inputs and select supported output controls in your provider.
Each operation retains its own draft while you switch. Clear current operation clears only that draft; Restore example restores it. Fields and output disappear on reload. The JSON export is an editorial planning document, not a provider API request. It includes the selected operation, composed prompt and a separate checklist.
Image and reference modes require a written reference role; Edit requires a source description and a change. These checks prevent an empty brief, not a bad generation. Writing a filename or reference role here does not upload an asset, grant access to it or select a provider's reference token.
Write an action you can inspect
Compare “a cinematic creator moment” with “the courier looks at the bright sky, glances at the closed umbrella and gives a small amused smile.” The second brief gives you a visible sequence to review. Keep the action restrained when identity or an important prop needs to stay legible.
For the doorway example, the umbrella remains closed. Asking for a new outfit, opening the umbrella, a camera orbit and a background change in the same first attempt would create several independent things to inspect. Remove a change when it does not serve the intended beat.

Open the native courier reference. Use it only as a practice input if the chosen route accepts it. Read the complete image before adding motion: the source's hands, crop and prop placement can influence what the model has to infer.
Give every reference one job
A creator portrait can supply identity, a product image can supply shape and color, and a location reference can supply the environment. Describe those roles explicitly. If you want a new location, say which reference elements should not carry over into it.
Our recorded café attempt used two images: a fictional creator with a short dark bob and sage overshirt, and a coral tumbler with a cream lid. The creator image was made before the October 5 request and already shows a hand on the tumbler. That detail becomes important when the written request asks for different hand placement.

Open the creator reference and separate tumbler reference. The R2V preparation guide explains their roles. These are the first-party files recorded in our original request, not newly generated variants.
Compare a recorded prompt with its actual result
Watch this clip and view its details
The record requested six seconds at 720p in 16:9, with generated audio off. The delivered file is 1280 × 720 and approximately 6.04 seconds, with no audio stream. Its historical quote is not a current price or verified debit. No new paid render was made for this page refresh.
Use @Image1 only for the fictional adult creator's identity, short dark bob and sage overshirt. Use @Image2 for the coral tumbler and cream lid. Place her in a NEW quiet cream cafe, not the reference kitchen. One medium continuous locked-camera shot. She glances at the tumbler then looks at camera with a small amused smile. Both hands remain resting on the table. Keep one coral tumbler with its lid attached. No new people, no speech, no added text or logos.
Open the exact recorded prompt or read the original request and settings. The @Image labels belong to that recorded tool request. In your current interface, attach the files and use the reference labels it actually inserts.
The clip follows the glance-to-camera action and shows the coral tumbler in the café. It misses the hand instruction: the hands touch the tumbler rather than both remaining flat on the table. An accepted expression does not make every constraint successful.

The input image's hand on the tumbler is a competing visual cue. It might contribute to the mismatch, but one attempt cannot establish a cause. A useful follow-up would keep the scene brief fixed and compare a cleaner identity reference with a different hand pose. That revision has not been rendered here.
Revise the failure instead of adding adjectives
| What happened | A focused next test | What to hold steady |
|---|---|---|
| Hands obscure the product | Use a reference with the required hand placement or simplify the contact requirement | Identity, product and camera brief |
| The scene copies the reference location | Clarify the new setting and each reference's role | The intended action |
| The camera invents a cut | Ask for one continuous shot and one simple move | The subject and action |
| The prop changes shape | Improve the product view and reduce handling | The product identity |
| The final export has the wrong format | Check the selected output controls and downloaded file | The accepted creative brief |
These are proposed tests, not verified fixes. Save the source, prompt, selected settings and complete result from each attempt. Changing the reference, action, camera and audio together makes it harder to learn which decision helped.
Open the blank review worksheet. It leaves observations, costs and acceptance empty. Judge the full clip, including its ending and any hand contact, before recording a pass.
Keep Edit prompts narrower than generation prompts
An Edit brief starts from footage you intend to keep. “Change only the overshirt to muted teal; preserve identity, timing, camera, café and tumbler” is a different job from creating the café scene again. The workbench includes that untested color-change example as a separate draft.
Choose the supported Edit operation and attach its source in the provider. If the change would require a different performance or a new shot structure, reconsider whether a targeted edit is the right route. The release guide separates generation and Edit controls and documents where published limits differ.
Select actual controls outside the prompt
A sentence about a portrait composition does not set an aspect-ratio control. A longer story does not extend a model's supported duration. Confirm the exact model, operation, crop, resolution, audio and displayed charge where you submit the job.
Higgsfield's help page also distinguishes connected generation through Canvas, MCP, CLI and Supercomputer from web Unlimited access. Check the route's charge instead of assuming an allowance follows the prompt everywhere. For integration, read the API guide; this generator's JSON is not a signed request.
Clout publishes this guide and offers its own recurring character workflow and AI Cinema workspace. Their account access and credits are separate. For this Seedance exercise, the free generator produces a brief and the recorded clip provides one concrete result to inspect; neither is a promise of repeatable model output.



