---
license: apache-2.0
language:
  - en
base_model:
  - Qwen/Qwen3-VL-4B-Instruct
  - prithivMLmods/Qwen3-VL-4B-Instruct-Unredacted-MAX
pipeline_tag: image-text-to-text
library_name: ggml
tags:
  - gguf
  - qwen3-vl
  - vision-language
  - comfyui
  - llama.cpp
  - uncensored
  - abliterated
---

# qwen3-vl-4b-unleashed

GGUF packaging of **Qwen3-VL-4B** (vision-language) for local ComfyUI / llama.cpp use — especially as a prompt-writer VLM for MiniMax H3 workflows.

## Files

| File | Role | Notes |
|---|---|---|
| `Qwen3-VL-4B-Instruct-Unredacted-MAX.BF16.gguf` | Language model (BF16) | ~8 GB |
| `Qwen3-VL-4B-Instruct-Unredacted-MAX.mmproj-bf16.gguf` | Vision projector (mmproj) | Required for image understanding |

## ComfyUI (H3 Prompt Writer)

Place both files under:

```text
ComfyUI/models/llm_gguf/
```

In **H3 Prompt LLM Loader (GGUF)**:

- `model_name` → the `.BF16.gguf`
- `mmproj_name` → the `.mmproj-bf16.gguf`

## Attribution / lineage

- Base model: [Qwen/Qwen3-VL-4B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct) (Apache-2.0)
- Unredacted / abliterated finetune: [prithivMLmods/Qwen3-VL-4B-Instruct-Unredacted-MAX](https://huggingface.co/prithivMLmods/Qwen3-VL-4B-Instruct-Unredacted-MAX)
- GGUF source pack: [prithivMLmods/Qwen3-VL-4B-Instruct-Unredacted-MAX-GGUF](https://huggingface.co/prithivMLmods/Qwen3-VL-4B-Instruct-Unredacted-MAX-GGUF)

This repo redistributes those GGUF weights under **Apache-2.0** for convenience. All credit to the original authors; this is a rehost with a shorter community name.

## License

Apache License 2.0 — same as the upstream Qwen3-VL and Unredacted-MAX releases.