Skip to content

feat: add lucy-vton-3.5 (realtime + batch) at 720p - #196

Merged
AdirAmsalem merged 1 commit into
mainfrom
feat/lucy-vton-3.5
Aug 10, 2026
Merged

feat: add lucy-vton-3.5 (realtime + batch) at 720p#196
AdirAmsalem merged 1 commit into
mainfrom
feat/lucy-vton-3.5

Conversation

@AdirAmsalem

@AdirAmsalem AdirAmsalem commented Aug 10, 2026

Copy link
Copy Markdown
Contributor

What

Adds lucy-vton-3.5, the newest Lucy virtual try-on model, as a canonical model on both the realtime and batch/video surfaces. lucy-vton-latest now resolves to it.

This model runs at 720p — a step up from the 1088×624 that vton-3 and vton-2 output — so the geometry the SDK reports for it (and for the lucy-vton-latest alias that now points at it) is 1280×720.

Why

Give SDK users the newest try-on quality without any migration work: callers already on lucy-vton-latest pick it up automatically, and anyone pinned to lucy-vton-3 or lucy-vton-2 is unaffected.

This re-lands a change that was previously reverted. The earlier attempt advertised the model at 1088×624, which no longer matches how it is served — that's corrected here.

Usage

import { createDecartClient, models } from "@decartai/sdk";

const client = createDecartClient({ apiKey });

// Batch / video
const result = await client.queue.submitAndPoll({
  model: models.video("lucy-vton-3.5"),
  prompt: "Wearing a red leather jacket",
  data: videoBlob,
});

// Realtime
const realtime = models.realtime("lucy-vton-3.5");
// → 1280x720 @ 30fps

Note for reviewers

The lucy-vton-latestlucy-vton-3.5 repoint needs the matching backend alias change to land as well; today the resolver still sends lucy-vton-latest to lucy-vton-3.

https://claude.ai/code/session_01QoAEWRoPpnJYkEV57yUvob


Note

Medium Risk
Changing lucy-vton-latest dimensions affects camera/stream sizing for existing alias users; backend alias must match or runtime behavior can diverge from SDK expectations until coordinated deploy.

Overview
Adds lucy-vton-3.5 as a canonical realtime and queue/video model in the SDK registry, with videoEdit2Schema inputs and dedicated generate/job paths. The model is advertised at 1280×720 (30fps realtime, 20fps batch), up from the 1088×624 geometry used for lucy-vton-3.

lucy-vton-latest client metadata is updated to match that 720p target (comments and width/height on the alias definitions), documenting server resolution to lucy-vton-3.5 once the backend alias lands. Pinned lucy-vton-2 / lucy-vton-3 callers are unchanged.

Docs (AGENTS.md), the demo index.html model picker, process input typing, and unit/e2e tests are extended accordingly.

Reviewed by Cursor Bugbot for commit 8b474bb. Bugbot is set up for automated code reviews on this repo. Configure here.

Re-adds lucy-vton-3.5 on both the realtime and batch/video surfaces after
the earlier revert, and repoints the lucy-vton-latest alias to it.

Geometry is 1280x720, not the 1088x624 the reverted change shipped: the
backend serves this model at 720p (bouncer supported_resolutions ["720p"],
inference_server/server_configs/lucy-vton-3.5.yaml canonical
supported_landscape_resolutions entry [720, 1280]). The lucy-vton-latest
rows move to 720p with it.

Claude-Session: https://claude.ai/code/session_01QoAEWRoPpnJYkEV57yUvob
@pkg-pr-new

pkg-pr-new Bot commented Aug 10, 2026

Copy link
Copy Markdown

Open in StackBlitz

npm i https://pkg.pr.new/@decartai/sdk@196

commit: 8b474bb

@AdirAmsalem
AdirAmsalem merged commit c45fb08 into main Aug 10, 2026
5 checks passed
@AdirAmsalem
AdirAmsalem deleted the feat/lucy-vton-3.5 branch August 10, 2026 08:54
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant