Short and unambiguous. This is the name you'll find him under in the library, and how you call him in prompts.
2What does he look like?
Describe only what always the same is: armor, colors, helmet, insignia, build. No pose, no place, no mood — that comes later into each image individually.
3Reference images
From these images the model learns what your character looks like. Ideal are 4–8 images of the same character from different directions — front, back, left, right — same clothing, calm pose, good light. After uploading, choose the direction under each image: that way the model knows which side it sees.
4Create
The character is saved in your library. After that you can use him in any image — in free mode simply pick him under "Who is in the picture?".
1What is the location's name?
2What does he look like?
Write it like a stage direction: place, ground material, architecture, light, time of day, weather, mood. The more concrete, the closer Marble hits your idea.
3One key image?optional
Optional — an image Marble orients itself by.
✕ remove
4Build world
Draft = quick preview in 30 seconds to check the description; the upgrade afterwards costs only the 3D part. If the description is right, the direct path is cheaper: panorama and 3D model in one step.
If the draft is enough for you, you're done. The full 3D world (20 tokens) builds a walkable model from the same panorama — giving you extra views from other viewpoints. Once built, the generator uses it automatically.
Which image to edit?optional
＋
Choose the image you want to change.
✕ other image
What should happen?
Who is in the picture?Choose character
Choose your character — he is used with all references.
What happens?Ideas
Improve prompt— optional
More images?optional
Location, object or vehicle from your library — or an image from your device. Nano is told the role of every image and uses it exactly that way; your prompt says where in the location the scene takes place.
No film yet. Create one — then it continues in three steps: Screenplay (your idea, ORIGIN drafts), Prepare (world and figures of the next scene), Shoot (scene by scene, like in the Video tab).
⟲ Undoundo the last change
Screenplay
✦ ScreenplayHave a draft offeredrough: scenes, endings, sound — you change it until it fits
ORIGIN · Rules · Light · Look
rebuild · one change
⟲ Discard screenplay
‹ Screenplay
Prepare
‹ Prepare
‹ Screenplay
Shoot
⟲ Start overReset scene — prompt and state gone, undo possible
Prompt — building blocks apply here too
✦ Improve prompt— beats, sound & a clean endingtargeted change
↓ Take the image prompt from the Image tab
Characters & Parts + Select
loading…
Werbespot
Motion
A real video delivers the Motion — your character performs it. Ideal: one main person, well lit, similar body height.
Style
Ton
Music
The video with the motion
No video chosen yet · video or GIF, max ~45 MB.
Several people? Drag a frame around the target person with your finger.✕ Remove marking
Prompt — only for a DIFFERENT change
Technical details
Figure+ Select
loading…
The identity comes from the character's FRONT image.
Library
Characters & Worlds
loading…
Enhancer
Videos appear here.
Jobs
Generations
No generations yet — they appear here as soon as the generator is connected.
Keys
API keys
Safely saved. Your key is written directly into the protected .env written on the server — never in the database, never back to the browser. Only whether a key is stored is shown.
From the description + first image, Marble builds a 360° panorama. Four views (front/right/back/left) are created from it automatically as references. When generating: attach the location — Nano gets the views, your prompt says where in the location the scene takes place. Draft 3 tokens, full 3D world 20 tokens (only for a walkable camera).
Identity pipeline
From your images a clean reference video is created, and from that an angle deck with 8 views. The deck is the single truth for images and videos — the best one wins automatically.
1
Images (bible)
—
Fixed details
What must always be exactly the same on this character: lettering, codes, insignia, emblems. Mark: Pick an image, drag a frame around the detail, name it — ORIGIN describes it precisely, the crop becomes a detail reference and goes along with every image and video.
+ enter yourself
1b
8 views — angle set from the existing images
front, back, left, right and the four diagonals. Without a reference video — the existing images are the template; usable ones are counted, only missing angles are generated.
2
Reference video — 360° turnaround from the front image
—
3
Angle deck — 8 views from the video
—
4
3D model — geometry and angles (top view, low angle)
—
Library tool: check the character's shape and angles (top view, low angle). Does NOT go into generation — static 3D renders would shape the models' look. The proper path (animated grey model as motion reference) comes once rig and animation exist.
More — quick variant, video identity, image checklist
Image checklist for the optimum: 10–25 images (never over 30) · ≥1024px · different poses/angles/backgrounds, ALWAYS the same armor · NO second character · at least 1 image from the BACK · 2–3 CLOSE-UPS of the emblems (back Y filling the frame, helmet Y frontal) · nothing covered · hands not constantly in front of the helmet.
Video identity (Wan) — only for solo videos of this character: —
ORIGIN STUDIO
