01
Use the official integration
Tencent's repository includes ComfyUI-AuK, an AuK Model Loader and AuK Generate/Edit nodes with a reusable workflow. That means creators do not need an unofficial node pack merely to access the newly released model.
Install AuK's documented .[comfyui] extra and place or link the included node package into the ComfyUI custom-nodes path according to Tencent's live guide.
02
Load Base, Flash and the encoder
Use the same checkpoint layout as the local setup: AuK Base or optional AuK-Flash plus Qwen2.5-Omni-3B. Confirm each configured path separately before treating a load failure as a model bug.
Base is the configurable path; Flash is Tencent's four-step distilled variant. Compare them on the same source audio rather than assuming the faster path always gives the best result.
03
Build the speech workflow
AuK chooses the job from the natural-language instruction, so the same node workflow can be adapted to TTS, content editing, enhancement or separation when the required audio input is supplied.
Prompt Enhancer is optional and requires an OpenAI-compatible LLM configuration. Leave it disabled initially when you want to distinguish raw model behavior from instruction-rewriting behavior.
04
Current ComfyUI limits and troubleshooting
Tencent's current documentation states a 30-second source-plus-target sequence limit for the ComfyUI integration. Treat that as today's integration constraint, not a permanent model-wide duration claim.
If memory is tight, load one variant at a time and avoid quoting unofficial VRAM numbers as requirements. For environment and weights, use the local guide; for supported tasks see the workflow guide; and return to the main AuK overview for license and release context.
Sources
Primary and supporting sources
Facts were rechecked against the linked sources immediately before publication. Pricing, product availability and rollout status can change.