Seedance 2.0 Prompting Guide: Text, References, Camera Control, and Video Editing

Most AI video guides stay too abstract. They tell you to "be specific" and leave you there.

That is not enough for Seedance 2.0.

The official Volcengine Seedance 2.0 prompt guide, updated on April 2, 2026, makes something clear: Seedance 2.0 is not just a text-to-video model. It is a multi-input directing system. You can control text inside the frame, lock character consistency with images, borrow motion from reference videos, and even edit or extend existing clips.

This article turns that documentation into a practical workflow for creators using Seedance 2.0 on HeyMarmot.

What You'll Learn

• Structure prompts so the model understands subject, action, environment, and audiovisual intent
• Generate slogans, subtitles, and speech bubbles that appear at the right moment
• Use image references for products, characters, logos, scenes, and storyboard frames
• Borrow action, camera language, and effects from video references
• Edit existing clips by adding, removing, replacing, extending, or bridging footage

The Real Mental Model for Seedance 2.0

The biggest mistake people make is treating Seedance like a keyword engine. It is better to think of it as a director that responds to layered instructions.

In practice, the strongest prompts usually combine four layers:

[Subject] + [Action] + [Scene / Style] + [Advanced direction]

Here is what each layer does:

• Subject tells the model who or what must stay important
• Action defines the motion, behavior, or event
• Scene / Style sets location, lighting, mood, and visual texture
• Advanced direction adds camera moves, sound, timing, or reference instructions

That sounds simple, but it changes everything. Instead of writing:

a woman in a city

you write:

A stylish young woman walks alone through a neon-lit elevated walkway in a futuristic city at night, turning her head toward a floating holographic ad. Medium shot, slow push-in, reflective wet ground, cool cyan lighting, soft ambient city hum.

That is the level of clarity Seedance rewards.

Start With Clear Reference Assignment

Seedance 2.0 becomes dramatically more reliable once you assign references explicitly.

If you upload multiple inputs, name their roles in the prompt:

• Image 1 for the main character
• Image 2 for the product or outfit
• Image 3 for the logo
• Video 1 for movement or camera language

Do not make the model guess.

Bad:

Use the images and make a cool ad.

Better:

Use image 1 as the main character reference, image 2 as the outfit reference, and image 3 as the logo reference. Create a premium beauty ad in a bright studio. The logo stays visible in the lower-right corner near the end.

That one change removes a lot of randomness.

Text Inside Video: Slogans, Subtitles, and Speech Bubbles

One of the most useful sections in the official guide is not about motion at all. It is about text generation inside video.

1. Slogans

If you want a slogan to appear in-frame, describe:

[text content] + [timing] + [position] + [appearance method] + [style]

Example:

Hand-drawn commercial illustration style. Three friends share a bucket of fried chicken and laugh together. As the scene gradually softens out of focus, the slogan "Good Times, Frame by Frame" appears in the center in a playful handwritten style, warm yellow lettering.

This works best for end cards, brand reveals, and short ad finishes.

2. Subtitles

The official documentation is very clear here: if you want subtitles to feel right, ask for them to sync with the audio rhythm.

Example:

Generate a cinematic nature voiceover video. A calm male narrator says, "Even the darkest night gives way to dawn." The scene transitions from a star-filled sky to sunrise over mountain ridges. Show subtitles at the bottom in sync with the spoken line.

The model performs better when the spoken content and subtitle behavior are described together, not separately.

3. Speech Bubbles

Speech bubbles are useful when you want stylized dialogue instead of normal subtitle treatment.

Example:

Two students jog on a school track in the morning. The girl looks toward the boy and says, "We can do this." A speech bubble appears beside her as she speaks. Cut to the boy's close-up as he replies, "Are you sure?" Then cut back to the girl smiling brightly: "Absolutely."

This is especially effective for comic-style, anime-style, or playful social clips.

Image References: Consistency Beats Guesswork

The official guide splits image prompting into two major buckets: single-subject consistency and multi-image composition. That is exactly how you should think about it.

Single Subject, Multiple Angles

If you have several photos of the same product or character, use them to stabilize identity.

Prompt pattern:

Reference images 1, 2, and 3 for the same subject. Generate a clean studio showcase. Keep the subject's appearance consistent while the camera slowly rotates and reveals the front, side, and rear details.

This is ideal for:

• Product turntables
• Character continuity
• Outfit consistency
• Commercial close-ups

Multi-Image Scene Building

This is where Seedance 2.0 gets interesting. You can split different responsibilities across different images:

• One image for the character
• One for the wardrobe
• One for the environment
• One for the logo
• One for storyboard composition

Example:

Set the scene inside the restaurant from image 4. Use the girl from image 1 wearing the outfit from image 2. The boy from image 3 enters as a customer and approaches the counter. Keep the logo from image 5 visible in the lower-right corner throughout the video.

That is much more powerful than "make a restaurant commercial."

Storyboard and Split-Panel References

The official guide also highlights storyboard-style inputs. This matters because Seedance can use panel order and composition as a planning signal.

If you already have a comic strip, storyboard, or shot board, say so directly:

Follow the panel order in image 1. Recreate the framing in sequence, then continue the scene naturally into a fast-paced fight sequence with escalating intensity.

This is one of the best ways to get more predictable pacing from a short clip.

Video References: Borrow Motion, Not Just Appearance

Image references lock the look. Video references lock the behavior.

The official guide breaks this into three categories, and that breakdown is exactly right.

1. Action Reference

Use a reference video when you want the generated subject to inherit a specific motion pattern.

Example:

Reference video 1 for the running motion. Generate a golden horse sprinting across an open grassland with the same stride rhythm and energy. At the peak of the motion, freeze the pose and transform it into a luxury horse-shaped pendant.

This is useful for:

• Dance moves
• Combat choreography
• Product handling gestures
• Animal movement

2. Camera Reference

This is one of the most underrated Seedance workflows.

Instead of saying "make it cinematic," point to a video whose camera movement already works.

Example:

Reference video 1 for the camera move. Create a first-person concept video flying toward the central tower from image 1. Keep the same diving motion and pace as the reference while emphasizing a futuristic technology-park atmosphere.

This works for dolly moves, drone-style approaches, orbit shots, and POV rushes.

3. Effects Reference

If a reference clip contains particles, wings, glow trails, or other stylized treatments, Seedance can transfer that visual behavior too.

Example:

Reference video 1 for the golden particle effect. The musician in image 1 plays a flute while matching golden particles spiral around the body in the same trajectory and density as the reference.

This is how you stop writing vague prompts like "add magic effect" and start getting repeatable results.

Video Editing: Seedance 2.0 Is More Than Generation

The official guide's editing section is probably the most practical part for production use.

Add, Remove, Replace

Think like an editor giving a surgical instruction.

Examples:

Add fried chicken and pizza snacks onto the table in video 1.

Remove the extra tools and scattered parts from the tabletop in video 1. Keep the hands, timing, and camera movement unchanged.

Replace the perfume bottle in video 1 with the cream jar from image 1. Preserve the original hand motion and camera move.

The clearer you are about what must stay unchanged, the cleaner the edit usually becomes.

Extend Forward or Backward

This is where many creators waste time by re-generating from scratch. You often do not need to.

Prompt patterns:

Generate what happens after video 1. Two late-arriving friends run into the frame and join the conversation.

Extend video 1 backward. Start with an over-the-shoulder shot of the man in white speaking before the current scene begins.

The official doc notes that the model automatically handles the overlap region, so the source clip is not simply repeated verbatim.

Bridge Multiple Clips

Seedance 2.0 can also create transition material between short clips.

Example:

Video 1: a leaf falls to the ground. A burst of golden particles rises on impact, a gust of wind sweeps across the frame, then transition into video 2.

According to the guide, this workflow supports up to three input videos with a total input length of no more than 15 seconds. That makes it perfect for short ad assembly, mood pieces, and stylized transitions.

Five Rules That Will Save You Credits

After comparing the official guide with real usage patterns, these are the rules that matter most:

1. Assign every reference explicitly. Never assume the model knows what image is for what.
2. Separate identity from motion. Use images for appearance, videos for action or camera.
3. Describe timing when text appears. This matters for slogans, subtitles, and logo reveals.
4. Tell the model what must remain unchanged in editing tasks. Especially for replacement edits.
5. Use common characters and clean phrasing for on-screen text. Fancy symbols reduce reliability.

Starter Prompt Templates

Product Ad With Logo Reveal

Use image 1 as the product reference and image 2 as the logo reference. Premium commercial lighting, dark reflective table, slow push-in toward the product. At the end, the frame softens and the logo from image 2 appears in the center in bright silver lettering.

Character Consistency With Image References

Reference images 1, 2, and 3 for the same woman. Generate a cozy cafe scene where she sits by the window eating cake. Keep facial features, hairstyle, and outfit details consistent across the shot. Warm afternoon light, natural handheld feel.

Camera Borrowed From Video Reference

Reference video 1 for the camera movement. Create a futuristic campus fly-through centered on the tower in image 1. Match the speed and dive trajectory of the reference. Blue-toned sci-fi atmosphere, subtle synth ambience.

Replace an Object in an Existing Clip

Replace the bottle in video 1 with the skincare jar from image 1. Preserve the hand gesture, timing, lighting, and camera move.

Final Thought

Seedance 2.0 gets dramatically better when you stop thinking in prompts and start thinking in production instructions.

That is what the official Volcengine guide really teaches. Not clever wording. Not prompt magic. Just clear assignment of subject, action, reference source, timing, and audiovisual intent.

If you work that way, Seedance stops feeling random and starts feeling usable.

You can try these workflows right now on HeyMarmot. Start with one clean reference setup, one camera instruction, and one simple objective. Then layer in text, effects, or editing once the base shot is stable.

That is the fastest path to videos that actually look directed.