Seedance 2.5: Put Yourself Into the Viral Motion-Transfer Trend (2026)

Complete step-by-step walkthrough

Original TECH4SSD tutorial and reference media

Tools

• Seedance 2.5 in a provider with video-reference access
• Higgsfield or Luma, subject to account availability
• Optional image model for a character sheet
• Video editor for trimming, review and final assembly

Before you begin

• Account with access to Seedance 2.5 video-reference generation
• Clear face and full-body photos of yourself or a consenting subject
• Reference footage and audio you have permission to use
• Credits for a short test before a longer generation

Watch the original Instagram tutorial

Open the approved tutorial video

1. How the trend works

The idea is simple: use a video as your movement reference and a character sheet as your identity reference. The video guides the walking path, gestures, poses and camera direction; the image shows the person and outfit you want in the result. Watch the linked TECH4SSD Reel for the finished demonstration. The downloadable mansion clip is the source reference, not a screen recording or an AI-generated result. Notice the performer, the staircase and the four framed portraits: those details explain why the longer prompt below is scene-specific. Motion transfer is generative. Expect to review identity, timing and scene continuity rather than assume an exact frame-for-frame copy.

Tip: The model name alone is not enough: your account must expose video-reference generation. Controls and reference tags can differ between platforms.

• Watch the Reel demonstration
• Identify the moving performer and the four fixed portrait panels

2. Make your character sheet (@image1)

The better your identity reference, the less the face drifts. Make one image that shows the same person several times: a full-body front view, a side view, a back view and two or three face close-ups, all in the same outfit, on a plain background. You can shoot this yourself, or generate it from a few of your photos with an image model. Attach your photos and use the prompt below. Check the result carefully: the face must actually look like you before you move on. Replace the outfit placeholder before generating. Only use your own likeness or someone who has agreed to this use. Save the final sheet locally and use that same file throughout your tests.

Exact prompt

Create a clean character reference sheet of the person in the attached photos. Keep their exact face, facial structure, skin tone, hair, body proportions and expression style. Layout on a plain light grey studio background: one full-body front view, one full-body side view, one full-body back view, and three head-and-shoulders close-ups (front, three-quarter, profile). Same outfit in every view: [describe the outfit you want in the video, including shoes and accessories]. Soft even studio lighting, photorealistic, sharp focus, no text, no labels, no logos.

• Same face in every view
• Outfit, shoes and accessories are the ones you want in the final video
• No text or watermarks on the sheet

3. Pick and trim the reference video (@video1)

Download the reference example below, or use your own authorized footage. This supplied clip is 30.07 seconds, 30 fps, and widescreen. If your model accepts a maximum of 30 seconds, trim the small amount beyond that limit before uploading; never rely on automatic truncation. Start with a five-second excerpt that includes clear movement. One lead performer and fewer cuts make problems easier to diagnose. Once that short test looks good, extend to the longer shot. For the mansion example, preserve the staircase, glass railing and four portrait panels. For a different source, describe its actual scene instead of retaining the mansion-specific instructions.

Warning: Permission to download this example does not automatically clear every use of its soundtrack. Use music and speech you own or are licensed to use, respect the generation platform's rules, and credit the source when required.

• Reference saved locally
• Reference duration fits the selected platform
• Audio and likeness permissions checked

Download the mansion reference video (MP4 · 1080p · 30 seconds)

4. Upload both in the Reference tab

Open your provider's video creation screen. Select Seedance 2.5 and confirm that the chosen mode accepts both image and video references. On interfaces that expose a Reference tab, use it rather than a start-frame-only mode. Upload the character sheet as the identity image and the trimmed clip as the motion reference. In this guide, @image1 means that character sheet and @video1 means the reference clip. Match those roles to the actual tags shown by your provider; do not assume every interface assigns the same names. Check each thumbnail and reference role before spending credits. If your account cannot attach a video reference, use a supported mode or provider instead of pasting a video link into an image-only flow.

• Video-reference mode available
• Character sheet assigned to identity
• Trimmed source assigned to motion
• Prompt tags match the uploaded files

Official Luma Seedance 2.5 guide

5. Paste the prompt

Use the complete scene-aware prompt below for the supplied mansion reference. It asks for one consistent identity across the moving performer and all four fixed wall portraits. These are instructions to the model, not a guarantee that every frame will satisfy them. TARGET_ACTOR in this prompt means the original lead performer; it is not an additional upload. @image1 is your character sheet and @video1 is the source clip. If your provider labels references differently, adjust those tags consistently. For another location, rewrite the CRITICAL PORTRAIT-WALL SEQUENCE to describe what is actually visible. If the clip has no portraits or screens showing the performer, use the shorter prompt in step 6.

Exact prompt

# CRITICAL SCENE-AWARE IDENTITY REPLACEMENT

INPUTS:
@video1 = motion / structural / camera reference
@image1 = MAIN CHARACTER identity reference

GLOBAL RULE:
@image1 represents the MAIN CHARACTER everywhere that character's identity appears in the source sequence.

This identity replacement applies not only to the physically moving performer, but ALSO to any visual representation, portrait, photograph, artwork, poster, framed image, screen image, reflection, or displayed likeness of that same lead character appearing inside the environment.

Do not treat those representations as unrelated background decoration.


MAIN PERFORMER:

Replace the moving lead performer completely with @image1.

Maintain @image1's exact:
face,
facial structure,
skin tone,
hair,
body proportions,
clothing,
footwear,
accessories,
and character styling

throughout the entire sequence.

@video1 controls the lead performer's:
movement,
walking path,
body poses,
gestures,
timing,
choreography,
interactions,
spatial position,
and camera relationship.


CRITICAL PORTRAIT-WALL SEQUENCE:

During the section where the lead character moves downward / through the stair or lower-level area, the camera reveals FOUR large portrait frames/panels mounted together on the wall.

These FOUR portraits are NOT generic artwork.

They are four stylized representations of the MAIN CHARACTER.

Therefore:

ALL FOUR portrait frames must display the exact identity of @image1.

Use @image1 as the identity source for every portrait.

Each portrait should clearly depict the SAME person from @image1 while preserving the source video's four-panel artwork composition.

The four portraits may use different artistic treatments, poses, crops or color palettes, but they must remain unmistakably the SAME @image1 character.

For example:

PORTRAIT 1:
@image1 identity, bold warm red/orange pop-art treatment.

PORTRAIT 2:
@image1 identity, cool cyan/blue pop-art treatment.

PORTRAIT 3:
@image1 identity, yellow/gold graphic treatment.

PORTRAIT 4:
@image1 identity, magenta/green or complementary graphic treatment.

Preserve the original:
number of frames,
frame dimensions,
wall placement,
spacing,
perspective,
camera relationship,
and overall pop-art visual concept.

Do NOT reproduce the original person's face inside any portrait.

Do NOT generate four random people.

Do NOT use background characters.

Do NOT change the portraits into abstract artwork.

Do NOT remove the portrait wall.

Do NOT duplicate the moving physical character into the room.

These are FLAT FRAMED ARTWORKS depicting @image1, not additional living people.


IDENTITY PROPAGATION RULE:

Whenever @video1 contains a recognizable depiction of TARGET_ACTOR, substitute that depiction with @image1.

This includes:

physical performer → @image1
portrait → @image1
photograph → @image1
poster → @image1
painting → @image1
screen displaying performer → @image1
framed artwork → @image1
recognizable reflection → @image1 when physically appropriate

Think of @image1 as a GLOBAL IDENTITY TOKEN for TARGET_ACTOR throughout the entire world.

If the source scene says visually:
"this is the lead character"

then the recreated scene must say:
"this is @image1."


PORTRAIT TEMPORAL CONSISTENCY:

The four portraits must already contain @image1 BEFORE they become fully visible.

Do not transform their faces after the camera reveals them.

Do not morph from the source face into @image1.

As the camera approaches, passes, tilts or changes perspective around the portrait wall, maintain the same four @image1 portraits continuously.

No face changes between frames.
No portrait flicker.
No identity drift.
No changing artwork layout.


MOTION AND CAMERA:

Preserve @video1 frame-for-frame as the authoritative source for:

walking,
gestures,
body orientation,
stair/lower-level movement,
timing,
camera tracking,
camera tilt,
camera height,
camera trajectory,
framing,
perspective,
subject scale,
and interaction timing.

Do not slow down the sequence to emphasize the portraits.

The portraits must naturally appear exactly when the camera reveals them in @video1.


ENVIRONMENT:

Recreate the environment in premium photorealistic live-action quality while preserving its spatial layout.

The wall containing the four portraits must remain in the same spatial position relative to:

the stairs,
lead performer,
glass railing,
floor,
camera,
and surrounding architecture.

Do not redesign the architecture in a way that removes or hides this storytelling element.


FINAL REQUIREMENT:

The sequence must communicate that the physical MAIN CHARACTER and the person represented in all FOUR wall portraits are the SAME identity from @image1.

ONE IDENTITY.
ONE CHARACTER.
FIVE VISUAL INSTANCES:

1 physical moving performer
+
4 framed representations of that performer.

Only the physical performer moves.

The four portraits remain fixed to the wall as artwork.

PRIORITY:
1. Physical lead = @image1.
2. ALL FOUR wall portraits = @image1.
3. Preserve exact source motion.
4. Preserve exact portrait reveal timing.
5. Preserve camera trajectory.
6. Preserve four-panel composition.
7. Maintain identity consistency.
8. Premium photorealistic cinematic environment.

Tip: Keep the prompt's identity and scene roles consistent. A stronger reference usually helps more than repeatedly adding stronger adjectives.

6. Simpler prompt for one-performer scenes

For a single performer with no portraits, posters or screens showing them, use this shorter prompt instead of the mansion-specific version. It preserves the same identity-versus-motion roles without asking for background elements your clip does not contain. The exact-motion language expresses the desired result. Check the generated output yourself; it does not make the model a deterministic motion copier.

Exact prompt

INPUTS:
@video1 = motion / structural / camera reference
@image1 = MAIN CHARACTER identity reference

Replace the lead performer in @video1 completely with @image1.

Maintain @image1's exact face, facial structure, skin tone, hair, body proportions, clothing, footwear, accessories and styling throughout the entire sequence.

@video1 controls the performer's movement, choreography, gestures, timing, walking path, spatial position and camera relationship. Preserve it frame-for-frame: camera tracking, tilt, height, trajectory, framing, perspective and subject scale.

Do not morph from the source face into @image1. No identity drift, no face changes between frames, no extra duplicate characters.

Recreate the environment in premium photorealistic live-action quality while preserving its spatial layout.

PRIORITY:
1. Lead performer = @image1.
2. Preserve exact source motion and timing.
3. Preserve camera trajectory.
4. Maintain identity consistency.
5. Premium photorealistic cinematic environment.

7. Match the duration and generate

Set output duration as close as the interface allows to your trimmed reference and keep the same aspect ratio. For this widescreen example, use 16:9 if exact source proportions are unavailable. Choose 1080p when your provider and plan support it. Run the short test first. Matching duration helps evaluation but cannot guarantee identical movement timing. Compare the same moments in the source and result before generating the full sequence. Treat audio separately. If dialogue and lip-sync matter, use an authorized audio reference in a supported workflow and check the lips against the actual words. Generated audio is not guaranteed to reproduce the source soundtrack; a muted reference provides no spoken-word timing to the model. If a provider rejects the audio, use your own recording or a licensed alternative, or contact its support. Do not disguise restricted audio to get around its checks.

Warning: An AI Motion Map is optional structural guidance, not an audio clearance tool or a guarantee of lip-sync. Keep an authorized audio reference when the selected platform supports it.

• Duration within provider limit
• Aspect ratio matches the intended composition
• Short test reviewed before full generation
• Audio choice and lip-sync checked

Official Higgsfield reference-to-video parameters

8. Review, fix and post

Review the source and output side by side at the start, the staircase section, the portrait reveal and the end. Check identity, hands, foot contact, camera direction, background geometry and the four wall portraits. Listen with sound on if lip-sync is part of the result. If identity drifts, improve the face views in the character sheet. If the portrait wall is wrong, revise that scene description. If one moment breaks, test a short excerpt around that moment. Change one variable at a time so you can tell what helped. Finish in your editor: use only authorized audio, check export playback, label the generated result clearly and credit the reference where required. A labelled source-versus-AI comparison makes the transformation understandable without implying that the generated footage is a real performance.

• Identity and hands checked across the shot
• Portraits remain fixed artwork, not extra people
• Camera and motion reviewed against the source
• Lip-sync checked when relevant
• Source credited and AI result labelled