Qwen-Image-2.1-Next-Scene-LoRA

Qwen-Image-2.1-Next-Scene-LoRA

A LoRA for Qwen-Image-2.1 that turns an input frame and a director-style instruction into the next shot, keeping the scene consistent. For storyboard makers and image creators.

Made by akhaliq 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: This LoRA teaches Qwen-Image-2.1 to continue a scene. You give it an image and a director-style instruction, and it generates the next shot with camera moves, reveals or lighting changes while the characters and style stay consistent. You can chain results into a storyboard.

Who it suits: Storyboard artists, filmmakers and content creators who want to plan shot sequences from a single frame.

Industry | Who uses it | How they use it | Example
Film and video | Directors | Plan shot sequences from a starting frame | A pull-back to a wide shot
Advertising | Creative teams | Sketch a short story in stills | Four directed shots from one photo
Games | Concept artists | Explore the next beat of a scene | A reveal behind a ridge
Content creation | Storytellers | Build sequences without a video model | A chain of edited frames

You can skip it if you want one-off edits or true video. The card says this is chained image edits, not a video model.

Where to use it:

• Python with diffusers and the QwenImage21Pipeline
• ComfyUI, with a LoRA node added to a Qwen-Image-2.1 graph
• Windows, macOS or Linux, with Python for the download step

Use case | Who | What they get back
Next shot | Directors | An image of the following scene from your frame
Storyboards | Artists | A chain of frames, each fed into the next
Camera moves | Creators | Pans, pull-backs and cuts described in text

How to set it up:

• Python and pip on your computer
• The diffusers library and Qwen/Qwen-Image-2.1
• The recommended next_scene_step2500.safetensors file

1. Install the Hugging Face hub package and diffusers.
2. Download the repository with the hf command.
3. Load the base pipeline and the LoRA in Python, as shown below.

pip install -U huggingface_hub
hf download akhaliq/Qwen-Image-2.1-Next-Scene-LoRA

How to use it: Start the prompt with Next Scene, lead with the camera move, then describe the subject or lighting change. In ComfyUI, set the LoRA strength between 0.7 and 0.9 and pass your image as the reference.

from diffusers import QwenImage21Pipeline
import torch

pipe = QwenImage21Pipeline.from_pretrained("Qwen/Qwen-Image-2.1", torch_dtype=torch.bfloat16)
pipe.load_lora_weights(
    "akhaliq/Qwen-Image-2.1-Next-Scene-LoRA",
    weight_name="next_scene_step2500.safetensors",  # diffusers/ComfyUI-converted
)

image = pipe(
    prompt="Next Scene: The camera pulls back to a sweeping aerial view, "
           "revealing the fleet behind the cliffs. Soft morning light, atmospheric depth.",
    image=input_image,          # your current frame
    num_inference_steps=28,
    true_cfg_scale=1.0,
).images[0]

Next Scene: The camera tracks forward and tilts down as rain begins to streak the lens.
Next Scene: Cut to a low-angle shot; sunlight breaks through the clouds behind her.
Next Scene: The camera pans right, revealing the fleet massing behind the ridge.

Example: You feed in a frame of a coastline and write a Next Scene prompt that pulls the camera back to an aerial view. The model returns the next frame in the same world and light.

For newcomers and experts: If you are new, use the step 2500 file at a strength near 0.8 and try one prompt. Lower the strength toward 0.7 if the output copies the input too closely.

If you know the field, this ports an existing next-scene format to the Qwen-Image-2.1 base, retrained natively. The card says the recommended step 2500 checkpoint transforms the scene most strongly while holding composition. Step 1500 is softer, and step 1000 is mostly for studying training.

Your current frame
Next Scene prompt with a camera move
Qwen-Image-2.1 with the LoRA
The next frame, fed back in for a chain

Good to know:

• License: Apache 2.0
• Cost to run: your own GPU hardware, with no paid service named
• Data it reaches: the images you give it and the files you download
• Not for: video generation, since it makes still frames only
• Limits: chained frames can drift, so the card suggests a lower strength

On Hugging Face, 13 November 2026

• Task: image to image
• Library: diffusers
• Downloads in the last 30 days: 4,581
• Likes: 71
• License: apache-2.0
• Based on: Qwen/Qwen-Image-2.1
• Tags: diffusers, qwen-image-2.1, lora, next-scene, storyboard, cinematic, image-editing, scene-continuation, ai-video, image-to-image

Links

• Model: https://huggingface.co/akhaliq/Qwen-Image-2.1-Next-Scene-LoRA
• Files: https://huggingface.co/akhaliq/Qwen-Image-2.1-Next-Scene-LoRA/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download akhaliq/Qwen-Image-2.1-Next-Scene-LoRA

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