PDF document
Complete breakdown
Kumar
A mixed reference-driven workflow that builds one stable model reference, creates selected GPT Image 2 start frames, uses Seedance for the speaking anchor shot, then cuts multiple Kling inserts into the final edit.
This breakdown is built around one main speech clip. First create the model reference image from the standard character reference sheet PDF, then generate only the stills needed for the key inserts, animate the dialogue shot in Seedance 2.0, animate the supporting editorial inserts in Kling 3.0, and assemble the final result around the main speaking timeline.
Reference Sheet PDFGPT Image 2Seedance 2.0Kling 3.0CapCut
Workflow
- 1Start by turning the base character setup into a cleaner reusable model reference with the standard reference-sheet PDF.
- 2Create only the stills needed to anchor the main speech shot and the supporting editorial inserts before any video generation begins.
- 3Use Seedance 2.0 for the speaking anchor shot, then generate the supporting editorial inserts in Kling 3.0.
- 4Build the final edit around the main Seedance speech clip, then cut the supporting Kling inserts into that base timeline.
This direct link includes the complete breakdown and every file promised with the video.
Final montage
36.9 MB • 1440×2560 • 0:19
Breakdown Step
1. Model reference image
Start by turning the base character setup into a cleaner reusable model reference with the standard reference-sheet PDF.
The workflow begins with the usual `character_reference_sheet_prompt.pdf` scaffold, which is used to create the model reference image for the man.
That model reference becomes the identity anchor for the later GPT Image 2 stills, keeping the face, hair, glasses, skin tone, and overall editorial presence stable across the breakdown.
Notes
- The PDF is treated as the reusable reference-sheet scaffold rather than a separately authored prompt block.
Breakdown Step
2. GPT Image 2 start images
Create only the stills needed to anchor the main speech shot and the supporting editorial inserts before any video generation begins.
Not every clip in the final sequence needs the same kind of setup, but the selected start images lock the exact framing, lighting, silhouette, and styling that the later video clips depend on.
The `cash` and `shoes` shots are supporting inserts rather than full identity portraits, but they still use their own start images so the later motion clips stay controlled and consistent.
Main
The office portrait start frame used for the speaking anchor shot.
Main image prompt
Creates the locked-off office portrait still in GPT Image 2.
Main image prompt
Creates the locked-off office portrait still in GPT Image 2.
Portrait 1
A high-contrast standing silhouette portrait used as one of the editorial cutaway inserts.
Portrait 1 image prompt
Creates the standing backlit silhouette still in GPT Image 2.
Portrait 1 image prompt
Creates the standing backlit silhouette still in GPT Image 2.
Portrait 2
A seated, heavily backlit silhouette portrait used as a second editorial insert.
Portrait 2 image prompt
Creates the seated sofa silhouette still in GPT Image 2.
Portrait 2 image prompt
Creates the seated sofa silhouette still in GPT Image 2.
Portrait 3
An extreme close-up backlit silhouette used for the tightest editorial insert and its orbit variation.
Portrait 3 image prompt
Creates the extreme close-up silhouette still in GPT Image 2.
Portrait 3 image prompt
Creates the extreme close-up silhouette still in GPT Image 2.
Cash
A macro insert still focused only on hands, ring, and money for one of the supporting cutaways.
Cash image prompt
Creates the macro hands-and-money still in GPT Image 2.
Cash image prompt
Creates the macro hands-and-money still in GPT Image 2.
Breakdown Step
3. Video generation
Use Seedance 2.0 for the speaking anchor shot, then generate the supporting editorial inserts in Kling 3.0.
The main office clip is the anchor because it contains the dialogue and stays as the base layer of the final timeline.
All supporting inserts were generated separately in Kling 3.0, including the `Portrait 3 Pan` variation, which reuses the `Portrait 3` start image but animates it with a different camera move.
Main
Seedance 2.0 • the locked-off speaking office clip that anchors the final edit.
Main video prompt
Uses the office portrait still for the spoken Seedance 2.0 anchor shot.
Main video prompt
Uses the office portrait still for the spoken Seedance 2.0 anchor shot.
Portrait 1
Kling 3.0 • restrained chin-touch silhouette insert.
Portrait 1 video prompt
Animates the standing silhouette still in Kling 3.0.
Portrait 1 video prompt
Animates the standing silhouette still in Kling 3.0.
Portrait 2
Kling 3.0 • slow push-in sofa silhouette insert.
Portrait 2 video prompt
Animates the seated sofa silhouette still in Kling 3.0.
Portrait 2 video prompt
Animates the seated sofa silhouette still in Kling 3.0.
Portrait 3
Kling 3.0 • locked-off extreme close-up silhouette insert.
Portrait 3 video prompt
Animates the close-up silhouette still in Kling 3.0.
Portrait 3 video prompt
Animates the close-up silhouette still in Kling 3.0.
Portrait 3 Pan
Kling 3.0 • orbit variation created from the same Portrait 3 start image.
Portrait 3 Pan video prompt
Creates the orbital variation from the Portrait 3 start image in Kling 3.0.
Portrait 3 Pan video prompt
Creates the orbital variation from the Portrait 3 start image in Kling 3.0.
Shoes
Kling 3.0 • minimal footwear-detail pan insert.
Shoes video prompt
Animates the shoe-detail still in Kling 3.0.
Shoes video prompt
Animates the shoe-detail still in Kling 3.0.
Breakdown Step
4. Final CapCut assembly
Build the final edit around the main Seedance speech clip, then cut the supporting Kling inserts into that base timeline.
The main office speech video is the anchor clip. Everything else is edited into it rather than replacing it.
The final sequence cuts in `Portrait 1`, `Portrait 2`, `Portrait 3`, `Portrait 3 Pan`, `Cash`, and `Shoes` as supporting inserts around the spoken message, keeping the workflow simple and editorial rather than overbuilt.
More breakdowns
Mech Girl
A beginner-friendly breakdown of how two reference images became a finished sci-fi friendship montage.
Japan Vlog
A vertical travel-vlog workflow that turns model-based reference stills into seven directed clips and one final CapCut edit.
Time Freeze
A frozen-time editorial workflow that starts from GPT Image 2 stills and turns them into slow-orbit Kling O1 clips before a simple CapCut assembly.
Golf Products
A golf product-style commercial breakdown that starts from GPT Image 2 stills, builds a real swing clip in Kling O1, then salvages weaker closeups in CapCut instead of wasting regeneration credits.
Bangkok Luxury Ad
A deliberately simple luxury fashion workflow: build the model, create the reference sheet, define the Bangkok scene, generate the whole ad in one Seedance 2.0 prompt, then finish with music.
Dune Editorial
A stark desert fashion breakdown that starts with five blue-hour stills in Nano Banana Pro and turns each one into a matching motion shot in Kling 3.0.
Ferrari Ad
A simple luxury automotive workflow: generate five Ferrari-and-horse stills in GPT Image 2, animate each one in Kling 3.0, then sequence the clips in CapCut with a single soundtrack.
Lambo x Inception
A surreal automotive breakdown that uses one impossible folded-city reference scene to create four Lamborghini start images, then animates them in Seedance 2.0.
Tiger Luxury Ad
A luxury animal-fashion breakdown that starts with one tiger portrait reference, expands it into three matching stills, and turns the set into a finished motion campaign.
Typography Layouts
A pure poster-study breakdown focused on fashion-led image prompts where layout, typography, styling, and art direction are all solved inside a single generated still.
Large AI Influencers
A pure GPT Image 2 breakdown focused on giant-scale fashion-editorial stills that make impossible influencer scenes feel photographic.
You vs You
A minimal reference-driven video breakdown that uses one younger-self image and one older-self image to build a single motivational identity-contrast scene.