SuperAPI AI Directory

seedance-pipeline

General & other

View repository

Updated 10-16 07:17

About this skill

seedance-pipeline

CLI-driven pipelines, ComfyUI integration, and post-processing for Seedance 2.0.


CLI-First Pipeline

The seedance CLI replaces direct API calls. All generation, polling, and downloading are handled through the CLI binary. See [skill:seedance-cli] for full command reference.

Single-Shot Pipeline (Generate → Wait → Download)

seedance generate "Glass perfume bottle on white marble. Slow 360° orbit. Soft studio light." \
  --model standard --duration 8 --ratio 16:9 --resolution 1080p \
  --wait --output product_shot.mp4

Fire-and-Forget Pipeline

# Submit without waiting
TASK_ID=$(seedance generate "Sunset mountain lake, drone pullback" -q)

# Check status later
seedance status $TASK_ID

# Download when done
seedance download $TASK_ID -o sunset_lake.mp4

Batch Pipeline (Multiple Variants)

for seed in 1000 1001 1002 1003 1004; do
  seedance generate "Urban street at night, neon reflections, handheld tracking" \
    --seed $seed --duration 8 -q
done

# Review all tasks
seedance list -n 5 --json

I2V Pipeline (Image-to-Video)

seedance generate "Character walks forward, wind blows coat" \
  --first-frame hero_pose.png \
  --duration 8 --wait --output hero_walk.mp4

Multi-Reference Pipeline (R2V)

seedance generate "Character A throws a punch, Character B blocks with crossed arms" \
  --image charA.png --image charB.png \
  --video fight_reference.mp4 \
  --duration 10 --wait --output fight_scene.mp4

Audio-Driven Pipeline

seedance generate "Character speaks to camera, medium close-up, locked" \
  --image character.png --audio dialogue.mp3 \
  --audio-gen --duration 8 --wait --output dialogue_clip.mp4

CI / Automation Pipeline

# Strict mode: exit 2 on timeout (for CI failure detection)
seedance generate "Product ad" \
  --wait --strict --timeout 600 --json \
  --output artifact.mp4

# Parse JSON output
seedance status $TASK_ID --json | jq '.status'
seedance list --status failed --json | jq '.[].task_id'

Platform Access

Surface Endpoint / App Notes
CLI seedance binary Full pipeline control
Web jimeng.jianying.com (Dreamina) 4–15 s, up to 1080p
Mobile CapCut / Jianying · Xiaoyunque 5–10 s
Consumer Doubao app Standard web limits

File Budget ("Rule of 12")

Type Max count Max size each Format
Image 9 30 MB JPG · PNG · WEBP
Video 3 combined ≤ 15 s total MP4 · MOV
Audio 3 total ≤ 15 s MP3
Total files 12 — —

The CLI validates the Rule of 12 before sending. Local files are automatically encoded to base64.


ComfyUI Node Workflow

[Load Image / Load Video] → [Seedance2 Sampler]
      ↓                           ↓
[CLIP Text Encode]          [Prompt Compiler]
      └────────────────────────→ ↓
                         [Video Output Node]
                                 ↓
                      [Frame Interpolation]
                                 ↓
                         [Upscale Node]
                                 ↓
                       [Color Grade Node]
                                 ↓
                        [Export / Mux Audio]

Key node parameters: duration, aspect_ratio, resolution, seed, motion_strength.


Post-Processing Chain

1 · Upscale

  • Tool: Topaz Video AI · Real-ESRGAN · ffmpeg scale=iw*2:ih*2
  • Target: 720p → 1080p (standard) · 1080p → 2K (premium)

2 · Frame Interpolation

  • Tool: RIFE v4.x · DAIN
  • Standard: 24 fps → 60 fps (smooth motion)
  • Fight / fast action: 24 fps → 120 fps

3 · Color Grade

  • Tools: DaVinci Resolve · FFmpeg LUT
  • Workflow: normalize exposure → apply LUT → mask-lift shadows → finalize.
  • LUT slots: Rec.709 (web) · Log-C (archive).

4 · Audio Mux

  • Merge generated stereo audio with video: ffmpeg -i video.mp4 -i audio.mp3 -c:v copy -c:a aac -shortest out.mp4

5 · Metadata Clean

  • Strip generation metadata before distribution: exiftool -all= output.mp4
  • Rename: {project}_{shot}_{take}_{date}.mp4

6 · Composite (optional)

  • Layer generated clips in After Effects / DaVinci Fusion.
  • Match color temperature across cuts before export.

End-to-End Shell Pipeline Example

#!/bin/bash
set -euo pipefail

PROMPT="Glass tower, 60 floors. Drone approach from south-east. Golden hour."
OUTPUT_RAW="raw_tower.mp4"
OUTPUT_FINAL="tower_final.mp4"

# Step 1: Generate with CLI
seedance generate "$PROMPT" \
  --model standard --duration 10 --ratio 16:9 --resolution 1080p \
  --wait --strict --timeout 600 --output "$OUTPUT_RAW"

# Step 2: Upscale
ffmpeg -i "$OUTPUT_RAW" -vf "scale=iw*2:ih*2" -c:v libx264 -crf 18 upscaled.mp4

# Step 3: Frame interpolation (RIFE)
# rife -i upscaled.mp4 -o interpolated.mp4 -n 60

# Step 4: Color grade
# ffmpeg -i interpolated.mp4 -vf lut3d=grade.cube -c:v libx264 graded.mp4

# Step 5: Metadata clean
exiftool -all= -overwrite_original "$OUTPUT_FINAL"

echo "Done: $OUTPUT_FINAL"

Output Specs

Use case Resolution FPS Container Audio
Web / social 1080p 30 MP4 H.264 AAC 192k stereo
Film festival 2K 24 MOV ProRes PCM 48kHz
Archive 2K 24 MKV H.265 FLAC stereo

Routing

For CLI command reference → [skill:seedance-cli] For prompt issues → [skill:seedance-prompt] For camera/storyboard → [skill:seedance-camera] For QA / errors → [skill:seedance-troubleshoot]