
Mech Girl
A beginner-friendly breakdown of how two reference images became a finished sci-fi friendship montage.
This workflow starts with original character concepts, turns them into cleaner reference sheets, locks scale, explores two Seedance video variations, and finishes with a simple CapCut edit.
Workflow
- 1Start with one clean image for the girl and one clean image for the mech so every later step has a stable visual identity.
- 2Use the same PDF prompt with each original image to convert rough concept images into cleaner, more production-ready reference sheets.
- 3Combine the original girl and original mech references into one shared image that locks the size relationship before moving into video.
- 4Use the two reference sheets plus the scale image as a stable three-image input set, then generate two different 15-second variations.
- 5Pick the strongest scenes from both Seedance results, join them in sequence, and add soundtrack to set the final emotional tone.
This direct link includes the complete breakdown and every file promised with the video.
Final montage
30.5 MB • 2560×1440 • 0:14
Breakdown Step
1. Original character concepts
Start with one clean image for the girl and one clean image for the mech so every later step has a stable visual identity.
The first pass was done in GPT Image 2 to create a grounded protagonist image and a separate mech concept image.
These are the anchor references for identity, materials, silhouette, and tone across the whole project.
Girl prompt
Used in GPT Image 2 to create the girl’s original reference image.
Girl prompt
Used in GPT Image 2 to create the girl’s original reference image.
Mech prompt
Used in GPT Image 2 to create the mech’s original reference image.
Mech prompt
Used in GPT Image 2 to create the mech’s original reference image.
Breakdown Step
2. Reference sheet generation
Use the same PDF prompt with each original image to convert rough concept images into cleaner, more production-ready reference sheets.
The `character_reference_sheet_prompt.pdf` file was reused twice: once with the girl original and once with the mech original.
This keeps the process simple while producing more stable identity references for later video generation.
PDF document


Notes
- The PDF is the reusable prompt scaffold for both character reference sheets.
Breakdown Step
3. Scale reference
Combine the original girl and original mech references into one shared image that locks the size relationship before moving into video.
This step makes the cockpit size and overall scale easy to verify before animation.
It prevents the mech from drifting toward a toy scale or shrinking toward a companion-bot size in later shots.
Scale reference prompt
Used to create the shared girl-to-mech scale reference image.
Scale reference prompt
Used to create the shared girl-to-mech scale reference image.
Breakdown Step
4. Seedance video variations
Use the two reference sheets plus the scale image as a stable three-image input set, then generate two different 15-second variations.
Both prompts preserve the same girl identity, mech identity, and scale relationship.
The second pass exists to search for stronger action and ending options without restarting the whole asset pipeline.
Notes
- Video 1 supplied scenes 1–3 for the final montage.
- Video 2 supplied scenes 4–5 for the final montage.
Video prompt 1
The first Seedance 2.0 pass. Scenes 1–3 were selected from this result.
Video prompt 1
The first Seedance 2.0 pass. Scenes 1–3 were selected from this result.
Video prompt 2
The second Seedance 2.0 pass. Scenes 4–5 were selected from this result.
Video prompt 2
The second Seedance 2.0 pass. Scenes 4–5 were selected from this result.
Breakdown Step
5. Final CapCut assembly
Pick the strongest scenes from both Seedance results, join them in sequence, and add soundtrack to set the final emotional tone.
Scenes 1–3 were taken from the first generated video and scenes 4–5 were taken from the second generated video.
The edit was assembled in CapCut and paired with Randy Newman's You've Got a Friend in Me (From Toy Story) as the background song.
Notes
- The raw MP3 is intentionally not exposed as a downloadable asset in this breakdown.
More breakdowns

Japan Vlog
A vertical travel-vlog workflow that turns model-based reference stills into seven directed clips and one final CapCut edit.

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.

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.

