Video Walkthrough
📄 Seedance 2.0 Prompting Playbook
Copy the full playbook below and upload it as a .md file into your ChatGPT Project — this gives ChatGPT everything it needs to write Seedance prompts correctly every time.
# ============================================================
# SEEDANCE 2.0 PROMPTING PLAYBOOK
# ============================================================
# THE PROMPT FORMULA
# ------------------
[SHOT STRUCTURE]
Total: [duration] / [number of shots] / [aspect ratio]
Format: [single continuous shot, multi-shot montage, POV, fight scene, transformation, animation, etc.]
[STYLE LOCK]
Cinematic lighting, photorealistic, 35mm film quality, professional color grading, sharp focus,
high-detail textures, film grain, depth of field, ARRI ALEXA aesthetic.
[REFERENCE LOCK]
Use the attached reference image/video as the identity, wardrobe, environment, and style lock.
Preserve the character's face, body proportions, outfit, color palette, lighting direction, and
camera angle unless explicitly changed.
[SCENE SETUP]
Describe the location, characters, wardrobe, emotional tone, time of day, weather, scale, and
the main visual conflict.
[CAMERA RULES]
Describe exactly what the camera does. Also describe what it must not do.
Example: no cuts, no zoom, no angle switching, no smooth stabilization, natural handheld head
movement only.
[ACTION BEATS]
Shot 1:
Shot 2:
Shot 3:
Shot 4:
Shot 5:
Shot 6:
[VFX / PHYSICS / DETAILS]
Describe practical effects, dust, water, sparks, cloth, hair, particles, explosions, motion blur,
and environmental reactions.
[SFX / AUDIO]
Optional but useful. Describe raw SFX, silence, breathing, impacts, wind, footsteps, etc.
[NEGATIVE STYLE GUARDRAILS]
No cartoon, no plastic skin, no 3D look, no extra characters, no duplicated faces, no warped
hands, no logo/text, no random camera cuts.
# ============================================================
# FORMAT 1: TRANSFORMATION SCENE
# ============================================================
Total: 15s / 6 shots / 16:9.
Montage, multi-shot action movie, do not use one camera angle or a single cut. Cinematic
lighting, photorealistic, 35mm film quality, professional color grading, sharp focus, high-detail
texture, film grain, depth of field, ARRI ALEXA aesthetic.
Use the attached reference image as the identity and style lock. Preserve the character's face,
body proportions, wardrobe, colors, and environment.
Scene: [Main character] is [doing normal casual action] in [specific location]. A
[threat/enemy/strange event] approaches from [direction]. The tone is [dark comedy / horror /
epic / surreal / action].
Shot 1: [Calm setup. Character is relaxed. Describe framing and mood.]
Shot 2: [Threat enters. Describe scale, movement, and camera reaction.]
Shot 3: [Close-up reaction. Character notices but reacts in a specific way.]
Shot 4: [Transformation begins. Describe body, energy, wardrobe, creature/mech changes.]
Shot 5: [Impact/action. Describe attack, collision, destruction, or visual payoff.]
Shot 6: [Aftermath. Character returns to normal or lands the final comedic/emotional beat.]
Camera: Handheld shake throughout, dynamic but readable, no random cuts outside the listed shots.
Realism: No 3D, no cartoon, no plastic skin, no glossy CGI, practical VFX feel, grounded realism.
# ============================================================
# FORMAT 2: POV / FIRST PERSON
# ============================================================
Total: 15s / 1 shot / 16:9.
Single continuous shot, first-person POV perspective, the camera IS the character's eyes.
Hyper-chaotic handheld motion, completely unstabilized, raw human movement, constant
micro-jitters, aggressive head swings, abrupt jerks, frequent over-rotation and harsh correction.
No cuts, no zoom, no third-person view, no cinematic cutaways.
Lens: Wide-angle with strong distortion, subtle chromatic aberration, cinematic lighting,
photorealistic, 35mm film look, heavy film grain, imperfect focus, motion blur on fast actions,
ARRI ALEXA aesthetic.
Action: The character's hands are always visible. In [location], the character
discovers/absorbs [power/object]. The energy spreads through their body [VFX: describe exact
energy pattern]. Enemies appear. Three escalating attacks follow.
Negative: No smooth camera, no stabilization, no third-person angle, no cutaways.
# ============================================================
# FORMAT 3: FIGHT SCENE
# ============================================================
Total: 15s / [1 continuous shot or 6 shots] / 16:9.
Cinematic fight scene with clear choreography, dynamic camera, photorealistic action, practical
impact physics.
Location: [Specific environment with surfaces, obstacles, lighting, weather, props.]
Fighter A: [appearance, outfit, size, attitude, weapon/power]
Fighter B: [appearance, outfit, size, attitude, weapon/power]
Beat 1: [Opening movement]
Beat 2: [First attack and dodge/block]
Beat 3: [Environmental interaction]
Beat 4: [Reversal]
Beat 5: [Slow-motion impact or precision moment]
Beat 6: [Final hit, fall, escape, or comedic button]
Camera: Keep both characters readable in the frame during major impacts.
Style: 35mm cinematic action, realistic motion blur, dust and debris physics, dramatic lighting.
Negative: No teleporting, no extra fighters, no random weapons, no broken limbs, no extra
characters, no cartoon physics.
Step-by-Step: How to Make Yourself a Movie Star with Higgsfield + Seedance 2.0
🛠Tools You'll Need
- Higgsfield AI — for Seedance 2.0 video generation
- ChatGPT (with a Project) — for building your prompts
- CapCut (or any editor) — for chopping and assembling clips
- Suno (optional) — for generating a soundtrack
- Starlight (optional) — for upscaling your final video
Step 1: Take Your Reference Photos
Don't overthink this — your iPhone is all you need.
- @Image1 — a clean front-facing photo of yourself (good lighting, simple background)
- @Image2 — your car, location, or scene reference (e.g. a Porsche reference sheet)
- Avoid AI-generated reference images — real photos of yourself work best
Step 2: Create a ChatGPT Project and Upload the Playbook
- Open ChatGPT and create a new Project
- Upload the Seedance 2.0 Prompting Playbook
.mdfile into that project - This gives ChatGPT full context on how to prompt Seedance correctly — every time
Step 3: Add Your Images and Generate a Prompt
Inside your ChatGPT Project:
- Drop in @Image1 (your photo) and @Image2 (your scene/car reference)
- Tell ChatGPT: "Use the first image, call it @Image1, and the second image, call it @Image2. Create a [X]-second scene of [describe what you want]."
- Important: tell it "Create a Seedance prompt — do NOT generate an image"
- Read through the output and refine it until you're happy with the scene description
Step 4: Paste Into Higgsfield and Set Up Your References
- Go to Higgsfield AI
- Upload your reference images in the same order — Image 1, Image 2
- Hold the @ symbol to confirm they're linked correctly in positions 1 and 2
- Paste your ChatGPT-generated prompt directly into Higgsfield — it will automatically connect @Image1 and @Image2
Step 5: Start Generating (and Generate a Lot)
This is the part nobody talks about upfront:
- Start at 480p to save credits while you dial in your prompt
- A 15-second generation costs ~$2.50 — expect to run many of them
- Not every generation will be perfect — that's completely normal
- The pros (including Higgsfield's own team) generate thousands of outputs to find the golden ones
- Once you have prompts that consistently work, bump up to full resolution
Step 6: Cherry-Pick Your Best Clips
- Review all your generations and pull the best moments from each
- You don't need any single perfect generation — you're assembling a scene
- Look for: good facial expression, clean motion, consistent character appearance
Step 7: Edit in CapCut (or Any Editor)
- Import your selected clips into CapCut
- Chop each clip to use only the best seconds
- Line them up to create smooth, continuous motion
- This is where the magic happens — your editing skills are now your superpower
Step 8: Add Audio + Lip Sync (Bonus)
- Soundtrack: Head to Suno, generate or grab a track, and drop it into your edit
- Lip sync trick: Upload a blank video with just your audio track into Higgsfield — the lip sync output is surprisingly accurate
- This is what takes a cool AI clip to a full cinematic moment
💡 Key Mindset Shift
The pros aren't getting it perfect every time. They generate thousands of outputs, find the golden nuggets, and assemble them. Your job isn't to get one perfect generation — it's to generate enough to have great options to work with.
🔗 Links
- Higgsfield AI (Seedance 2.0) — start generating
- Suno — AI music generation
- Starlight — video upscaler
- CapCut — free video editor
