Upload one character illustration — usually an action pose with a weapon in hand — and get the character redrawn in a clean T-pose, plus every prop as its own transparent PNG. That is the input image-to-3D and auto-rigging actually need: convert an action pose directly and the weapon fuses to the hand, while the body hidden behind the arms is invented rather than observed.
1Upload a character image
One character per image, fully in frame (PNG, JPG or WebP up to 10MB) — and up to ten images in one pass, so a whole cast is planned in a single sitting. A plain background leaves the model the least to guess at.
2Read the plan — free
A vision model reports the subject, body plan, view and pose, flags what is hidden, and proposes which pieces to split out. The analysis costs no credits, so you can rerun it at a different granularity.
3Decide what counts as a piece
Four levels from don't split to modular. Untick a prop and it stays on the character instead of becoming its own asset — the T-pose is then redrawn still wearing it.
4Generate, download or send on
The body in T-pose is drawn at 2k for 3 credits, each prop at 1k for 2 — all transparent PNG. Download them individually or as a ZIP, or hand any piece straight to Image → 3D without a download round trip.
| Input | Up to 10 character images per run (PNG / JPG / WebP, max 10MB each), plus an optional note |
|---|---|
| Output | A transparent PNG per piece: the character in T-pose at 2k, each prop at 1k |
| Analysis | Free — subject, body plan, view, pose, occlusion warnings and the proposed split |
| Credit cost | 3 credits for the T-pose body, 2 per prop; you pay only for the pieces you keep ticked |
| Control | Split granularity (none / coarse / standard / modular) and a free-text instruction |
| Best for | Preparing character art for image-to-3D, auto-rigging, or a 2D skeletal rig |
Redrawing is generation, not surgery: the torso behind a crossed arm and the far side of a cloak are invented plausibly, not recovered, and a prop that is heavily hidden in the reference comes back approximated. On elaborate costumes the style drifts slightly from the original, and small details — faces, insignia, text on a banner — are the first to shift. Non-humanoid subjects split fine, but the T-pose and the standard humanoid skeleton do not apply to them. Output is a flat PNG per piece: no layered PSD, no vectors, and no guarantee that two pieces line up pixel-perfectly if you paste them back together.
Because an action pose reconstructs into one fused lump. The weapon and the hand come out as a single connected mesh, the body behind a crossing arm is invented, and skinning weights then bleed between the torso and the upper arm. A T-pose with the props separated avoids all three.
Nothing. Reading the image and proposing the split is free — credits are only spent on the images you choose to generate: 3 for the T-pose body, 2 for each prop.
Yes. Untick it and the character is redrawn still holding it; no separate asset is made for it. Everything you untick stays on the body, so the pieces always add back up to the original character.
The analysis marks it as already a rest pose and the body is not redrawn — you pay only for the props you split out, or nothing at all if there are none.
Every finished piece has a Convert to 3D button that hands the image over to Image → 3D directly — no download and re-upload, and no risk of converting a different file than the one you approved.
Yes — drop in up to ten images and every one is analysed for free, each as its own job. You then see all the plans together, tick what you want from each, and pay once for the whole batch. One image failing does not affect the others, and the ZIP comes back with a folder per character.