Local workflow

The workflow you already tuned, ready to use

Other tools make you adapt to the workflow. YanDong AI makes the workflow adapt to your project — no JSON editing, no plugins, no node graphs to understand.

Supported models and workflows

VideoMiniMax H3 multi-referenceImage, text and audio references enter together with 30s direct output. Six workflow variants are adapted, including dual-reference and multi-reference export packages.

VideoH3 image-to-video acceleratedBF16 high-quality accelerated pipeline plus a 10 Eros mixed Turbo variant, suited to batch production.

ImageQwen Image 2.1Enters through the ComfyUI image channel, with node names and model file paths matched automatically.

AudioFish Speech S2 / Breeze TTS 2Voice cloning and line-by-line dubbing; generated reference audio can be fed back into local video workflows.

AnyAny custom workflowUI-exported JSON is read directly. If a node is missing, the software names it.

What happens when you connect one

Detect nodes

input positions for image, text, output, audio and duration are found by node type, not by node name

Convert format

UI-exported JSON is converted to API format automatically

Compute frames

converted to each model's legal frame count so durations never mismatch

Inject assets

reference images, prompts, voices, driver audio, duration and seed are placed automatically

Collect output

finished results land in project history, identical to cloud output

The awkward parts it handles

Model paths spelled differentlyeach vendor formats paths differently; only the final filename segment is compared

Cloud-only nodeswhen a workflow references a cloud node, the equivalent local node is connected instead

Deliberately bypassed linksthe single viable connection is followed rather than failing outright

Stray note nodesskipped automatically, so an extra text box never breaks a run

Disconnects and restartsthe job is found back in the queue by its unique id; if not found the software says so rather than silently re-running and wasting GPU hours

What running locally means

No per-second billing

local rendering consumes no credits — 100 clips a day costs the same as 1

No queueing

cloud peaks mean waiting; local runs when you want, and keeps running offline

Nothing leaves the machine

script, characters and voices stay on your computer

Local deployment, no cloud content review

Generate with your own GPU and local models through ComfyUI workflows, without this software’s cloud content review. Restrictions depend on the models and nodes used and whether external services are called.