MOSS-Audio-8B-Instruct-ComfyUI

Instruction-following audio understanding and transcription with text output.

This is a converted model package for Jack Spooner’s ComfyUI fork, derived from lakucho/MOSS-Audio-8B-Instruct. It is not a new fine-tune.

Required code

The integration modifies ComfyUI core; this is not a drop-in custom_nodes package.

• Model integration commit
• Implementation · Offline converter
• Development branch · Pinned compatible reference snapshot 9d9053c5e
• Installation, node usage and acceptance details

For an existing ComfyUI installation, use the linked core changes and their prerequisites against a compatible base. The integration commit alone is not asserted to be an independently cherry-pickable patch. The pinned reference snapshot includes the required shared loaders, attention and memory changes. Follow the fork’s installation README and dependencies when reproducing that snapshot; update the application only while its queue is idle. Upstream ComfyUI is Comfy-Org/ComfyUI; its README describes the underlying node engine and model management.

Files and installation

Place moss_audio_8b_instruct_bf16.safetensors in ComfyUI/models/text_encoders/. The tokenizer and required inference configuration are embedded; original HF shards are not required at inference.

File	Size
moss_audio_8b_instruct_bf16.safetensors	18.110 GB

Manifest/provenance files record the conversion and source revisions. SHA256SUMS and FILES.json describe this release. Filenames and model folders matter. No original source snapshot needs to be converted again.

Usage

Select moss_audio_8b_instruct_bf16.safetensors in Load CLIP, type moss_audio. Connect CLIP to Generate Text and Load Audio to its audio input. Use a specific transcription, captioning or audio question prompt and an explicit output budget. Start with thinking disabled for this Instruct checkpoint; the control changes the prompt and is not a different set of weights. Select your own audio in the example.

Example graphs are in workflows.

Verification and limits

The package has pinned conversion provenance. Native Instruct participation in the Omni model-transition/lifecycle fixture is documented. The larger full-model caption/reference acceptance suite was conducted for Thinking and must not be assumed to validate this Instruct variant. This is a converted release with bounded variant acceptance. Exact timestamps, recognition accuracy, low-VRAM operation and other backends are not guaranteed.

These statements describe the linked retained tests, not a new cross-platform acceptance suite. File size is not a minimum RAM/VRAM requirement. No inference-time downloads or external inference worker are required by these native model paths.

Conversion and provenance

Repackages the exact lakucho mirror revision, preserving 901 BF16 source tensors and embedding its tokenizer. This is an OpenMOSS model distributed via that mirror; it is not our fine-tune.

Source model revision: 8bd5b22c64dd80b59c06af4e3304e581a3e5596b. See PROVENANCE.json, the conversion manifest and NOTICE. No training data, private inputs or generated user recordings are included.

License and attribution

The pinned mirror model card declares Apache-2.0 and credits OpenMOSS. Both the actual source mirror and the original OpenMOSS authors are attributed.

See LICENSE, NOTICE, and the retained upstream model card/source notices. Model authors and publishers retain credit; this is a third-party ComfyUI repackaging and does not imply endorsement. The weights’ terms and the separately linked ComfyUI code license are distinct.

The example workflow files are copied/adapted from the linked ComfyUI fork; its GPL-3.0 license is retained in LICENSES/COMFYUI-GPL-3.0.txt. This does not relicense the model weights.