QMA Camera Angles

QMA Camera Angles

Re-renders a picture from a new camera angle, chosen with a short prompt, as an edit LoRA for Qwen Image 2.1. It suits artists and developers who edit images in ComfyUI or diffusers.

Made by gribbliest on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: QMA Camera Angles changes the viewpoint of a picture, as a LoRA for editing images with Qwen Image 2.1. You supply a picture and say where the camera should be. The subject, the scene and the look stay put.

The camera can circle the subject, rise or drop, or move in close. It always needs a source image, so it edits pictures and does not make them from nothing.

Who it suits: Artists, designers and developers who want to see one picture from other viewpoints suit it best. Anyone building an image editor can add camera control with it.

Industry | Who uses it | How they use it | Example
Design | Designers | Show one subject from several viewpoints | Orbit around a product picture
Creative work | Artists | Check a scene from another camera position | Raise the camera and compare the new view
Software | Developers | Build camera control into an image editing workflow | Pair it with the ComfyUI node

Anyone who wants to make a picture without a source image can skip it.

Where to use it: You run it with Qwen Image 2.1 in a tool that loads LoRA files.

• ComfyUI, which has a camera node by CoreyCorza for building the prompts
• Python with the diffusers library, the library the record lists
• Windows, macOS or Linux with Python, to download the files

Use case | Who | What they get back
Orbiting a subject | Artists | The same subject seen from another side
Changing height | Designers | The scene from a lower or higher camera
Pulling in or out | Developers | A close-up or a wide shot of the same scene

How to set it up: You need the Qwen Image 2.1 base model, a tool to run it, and a source image. Load the LoRA at strength 1.0 and connect your image as the reference image.

• The Qwen Image 2.1 base model
• ComfyUI or a diffusers script that can apply a LoRA
• A source image to edit

1. Install Python.
2. Install the hub tools with pip.
3. Download the LoRA files with the hf command.
4. Load Qwen Image 2.1 with the LoRA at strength 1.0 in your tool.
5. Send your source picture to the model as its reference.

pip install -U huggingface_hub
hf download gribbliest/qwen-image-2.1-qma-camera-angles

The card gives these generation settings.

Setting | Value
LoRA strength | 1.0
Steps | 30
CFG | 3
Sampler | euler
Scheduler | simple
Output size | Around 1 MP for the best detail, or 768 for quick previews

How to use it: Every prompt has the same shape. You write the tag, then one azimuth, one elevation and one distance, in that order.

<qma> {azimuth} {elevation} {distance}

Azimuth picks the side the camera moves to. Choose one of these.

front view
front-right quarter view
right side view
back-right quarter view
back view
back-left quarter view
left side view
front-left quarter view

Elevation sets how high the camera sits.

low-angle shot
eye-level shot
elevated shot
high-angle shot

Distance sets how far away the camera is.

close-up
medium shot
wide shot

These are three complete prompts.

<qma> right side view eye-level shot medium shot
<qma> back view elevated shot wide shot
<qma> front-left quarter view low-angle shot close-up

Directions are set from where the photo was taken, not from the subject's point of view. A right side view swings the camera 90 degrees to its own right, and a back view takes it behind the subject.

Example: You give the model a photo and use the first of the three example prompts above. It returns the same scene with the camera swung to its own right at eye level.

For newcomers and experts: Think of this as a virtual camera operator for a single picture. Your first small try is to load a photo, write one of the three example prompts, and look at the new view.

Under the hood, it is a LoRA of rank 64, trained with AI-Toolkit for 3,000 steps at both 512 and 1024 resolution. Training used photos and Blender renders. Each pair shows one subject from two camera positions, so together the pairs cover all 96 ways to combine angle, height and distance.

Your source image
A camera prompt: angle, height, distance
Qwen Image 2.1 with the LoRA at 1.0
The same scene from the new camera

Good to know: The model has to guess what the source image does not show, so results are convincing but not exact.

• Reliable: quarter views, higher camera positions and distance changes work best
• Big orbits: in a wide outdoor scene, the camera may not reach a full 90 degrees, and another seed or the next view over usually helps
• Hidden parts: the back of a character, for example, is a guess
• Extra edits: a prompt with only camera words is cleanest, and other edit instructions can follow it
• Based on: Qwen Image 2.1
• License: apache-2.0
• Cost to run: none named
• Data it reaches: the images you give it, and the files downloaded from Hugging Face

On Hugging Face, 13 November 2026

• Task: image to image
• Library: diffusers
• Downloads in the last 30 days: 13,875
• Likes: 55
• License: apache-2.0
• Based on: Qwen/Qwen-Image-2.1
• Tags: diffusers, lora, qwen-image, image-editing, camera-control, camera-angle, multi-view, image-to-image

Links

• Model: https://huggingface.co/gribbliest/qwen-image-2.1-qma-camera-angles
• Files: https://huggingface.co/gribbliest/qwen-image-2.1-qma-camera-angles/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download gribbliest/qwen-image-2.1-qma-camera-angles

Then follow the usage in its model card. Nothing in this kit runs by itself.