---
license: apache-2.0
base_model:
- Qwen/Qwen3-VL-32B-Instruct
language:
- en
pipeline_tag: image-text-to-text
library_name: transformers
tags:
- text-generation-inference
- abliterated
- v1.0
---

![1](/static-proxy?url=https%3A%2F%2Fcdn-uploads.huggingface.co%2Fproduction%2Fuploads%2F65bb837dbfb878f46c77de4c%2Flv4kyP81dI7IX12Y9Bexe.png)

# **Qwen3-VL-32B-Instruct-abliterated**

> **Qwen3-VL-32B-Instruct-abliterated** is an *abliterated (v1.0)* variant of **Qwen3-VL-32B-Instruct**, designed for **Abliterated Reasoning and Captioning**.
> This model is optimized to generate **detailed, descriptive captions** and **reasoning outputs** across a wide range of visual and multimodal contexts—including complex, sensitive, or nuanced content—while supporting diverse aspect ratios and resolutions.

<img src="/static-proxy?url=https%3A%2F%2Fcdn-uploads.huggingface.co%2Fproduction%2Fuploads%2F65bb837dbfb878f46c77de4c%2F1xWD8FQ2qvqQzyqOzy05I.jpeg" alt="1" style="border-radius: 30px;"/>

## Key Highlights

* **Abliterated / Uncensored Captioning** – Fine-tuned to bypass conventional content filters while preserving factual, descriptive, and reasoning-rich outputs.
* **High-Fidelity Descriptions** – Generates comprehensive captions and reasoning for general, artistic, technical, abstract, or low-context images.
* **Robust Across Aspect Ratios** – Performs consistently across wide, tall, square, and irregular image dimensions.
* **Variational Detail Control** – Capable of producing outputs ranging from concise summaries to fine-grained, intricate descriptions and reasoning.
* **Foundation on Qwen3-VL-32B Architecture** – Built upon Qwen3-VL-32B-Instruct’s advanced multimodal reasoning and instruction-following capabilities.
* **Multilingual Output Capability** – Primarily optimized for English, with adaptability for multilingual prompts through prompt engineering.

## Quick Start with Transformers

```python
from transformers import Qwen3VLForConditionalGeneration, AutoProcessor
from qwen_vl_utils import process_vision_info
import torch

model = Qwen3VLForConditionalGeneration.from_pretrained(
    "prithivMLmods/Qwen3-VL-32B-Instruct-abliterated",
    torch_dtype="auto",
    device_map="auto"
)

processor = AutoProcessor.from_pretrained("prithivMLmods/Qwen3-VL-32B-Instruct-abliterated")

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
            },
            {"type": "text", "text": "Provide a detailed caption and reasoning for this image."},
        ],
    }
]

text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
image_inputs, video_inputs = process_vision_info(messages)

inputs = processor(
    text=[text],
    images=image_inputs,
    videos=video_inputs,
    padding=True,
    return_tensors="pt",
).to("cuda")

generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [out[len(inp):] for inp, out in zip(inputs.input_ids, generated_ids)]

output_text = processor.batch_decode(
    generated_ids_trimmed,
    skip_special_tokens=True,
    clean_up_tokenization_spaces=False
)
print(output_text)
```

## Intended Use

This model is suited for:

* Generating detailed, uncensored captions and reasoning for general-purpose or artistic datasets.
* Research in **content moderation**, **red-teaming**, and **generative safety evaluation**.
* Enabling descriptive captioning and reasoning for visual datasets typically excluded from mainstream models.
* **Creative applications** such as storytelling, art generation, or multimodal reasoning tasks.
* Captioning and reasoning for **non-standard aspect ratios** and **stylized visual content**.

## Limitations

* May produce explicit, sensitive, or offensive descriptions depending on the image content and prompts.
* Not recommended for production systems requiring strict content moderation.
* Output style, tone, and reasoning may vary based on input phrasing.
* Accuracy can fluctuate for unfamiliar, synthetic, or highly abstract visual content.