xkcd-artist

xkcd-artist

Turns a text prompt into an XKCD-style comic by planning the scene, poses and speech bubbles, with SVG or PNG output. For people who want quick comic panels from an idea.

Made by gnumanth 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: A tiny text to image setup that draws XKCD-style comic panels from a sentence. It pairs a small planner with a diffusion LoRA. The planner decides the scene layout, props, poses and speech bubbles, and the LoRA adds the look. It has about 1.6M parameters in its main weights, and it exports the result as SVG or PNG.

Who it suits: People who want quick comic panels from an idea and who can run a short Python snippet or a browser demo.

Industry | Who uses it | How they use it | Example
Education | Teachers | Illustrate a technical idea with a comic panel | A panel that explains a concept from one sentence
Publishing | Bloggers and writers | Add a quick comic to a post | A comic exported as a PNG for an article
Software | Web developers | Run a tiny model in a page through ONNX | Generate panels in the browser
Research | ML students | Study a small planner plus diffusion LoRA design | Compare the PyTorch and ONNX runtimes

You can skip it if you want realistic or painterly images, since it targets comic panels.

Where to use it: It works from Python with a PyTorch or ONNX backend, and the card also lists a browser runtime and a live studio page.

• Python scripts, with the PyTorch or the ONNX backend
• A browser, through the Int8 ONNX graph with WASM SIMD
• The project's live interactive studio, which the card links

Use case | Who | What they get back
Comic from a prompt | Writers | A panel as SVG markup or a PNG file
In-browser generation | Web developers | The same planner running as a small ONNX graph
Model study | Students | Small files and a self-contained PyTorch definition to read

How to set it up: Get the project code from GitHub, then the model files from Hugging Face. The card's snippet imports the xkcd_artist package from that project.

• Python
• The xkcd-artist project code from GitHub
• Room for a few megabytes of model files

1. Get the xkcd-artist code from GitHub.
2. Install the Hugging Face tools.
3. Download the model files to your computer.
4. Run the card's snippet and open the files it writes.

pip install -U huggingface_hub
hf download gnumanth/xkcd-artist

How to use it: Call xkcd_artist with your prompt and pick a backend, then export the result. The card's snippet is below.

from xkcd_artist import xkcd_artist

comic = xkcd_artist(
    "Machine learning pile of linear algebra stir the pile",
    backend="onnx",  # 'onnx' | 'pytorch'
)

svg_markup = comic.to_svg()
png_bytes = comic.to_png("machine_learning_xkcd.png")

Tool | What it does
xkcd_artist | Turns a prompt into a comic object, using the onnx or pytorch backend
to_svg | Returns the comic as SVG markup
to_png | Saves the comic as a PNG file
model.safetensors | The planner's FP32 weights, 1,586,603 parameters
lora_weights.safetensors | The planner's internal LoRA adapter, 93,568 parameters
onnx/model_quantized.onnx | An Int8 ONNX graph that is 3.58x smaller
sd_xkcd_lora/adapter_model.safetensors | A LoRA for the segmind/tiny-sd UNet, trained on panels with speech bubbles masked

Example: Pass the sentence about machine learning from the card's snippet and keep the onnx backend. The call returns a comic you can turn into SVG markup, or save as machine_learning_xkcd.png.

For newcomers and experts: A text to image model draws a picture from a sentence. This one is tiny and aims only at comic panels. A first small try is to run the card's snippet as it is and open the PNG it saves.

For experts, a 5-head planner predicts the scene, poses and speech-bubble layout, and a rank-16 LoRA with 1,726,464 trainable parameters adapts the segmind/tiny-sd UNet. Training used 891 ExplainXKCD stage directions and 226 panels with the bubbles masked. The card reports 100.0% layout accuracy on its held-out set, so check results on your own prompts.

Write a one-line prompt
Planner lays out scene, poses and bubbles
Diffusion LoRA adds the look
Export SVG or PNG

Good to know: The card lists these file sizes.

Title: File size of the main artifacts
Planner FP32 weights: 6.16 MB
Int8 ONNX graph: 1.72 MB
Diffusion LoRA adapter: 6.60 MB

• Benchmark: the card reports 100.0% layout and stage prop accuracy for both runtimes on its held-out data.
• Latency: median 0.629 ms for PyTorch and 0.621 ms for the ONNX build.
• Training data: real XKCD panels, so think about rights before you publish the output.
• License: MIT.
• Cost to run: none named, no API key, account or paid service.
• Data it reaches: your prompt and the files you save, plus the model files downloaded from Hugging Face.
• Not for: the card names no limits.

On Hugging Face, 13 November 2026

• Task: text to image
• Library: transformers
• Parameters: 1.6M
• Weights: about 6 MB in F32; about 1 MB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 42
• Likes: 3
• License: mit
• Languages: en
• Tags: transformers, onnx, safetensors, xkcd_artist, xkcd, comics, diffusion-lora, stick-figure, scene-graph, pytorch, text-to-image

Links

• Model: https://huggingface.co/gnumanth/xkcd-artist
• Files: https://huggingface.co/gnumanth/xkcd-artist/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download gnumanth/xkcd-artist

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