Chroma1-HD

Chroma1-HD

Turns a text prompt into an image with an 8.9B open base model built on FLUX.1-schnell, made as a starting point for artists and developers who want to finetune their own image models.

Made by lodestones 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: Chroma1-HD is an 8.9B text-to-image model from lodestones, built on FLUX.1-schnell and released under Apache 2.0. It is a base model, meant as a neutral starting point that you can finetune into a specialized image generator. A separate Chroma1-Flash model is the fast version.

Who it suits: It suits developers, researchers and artists who want an open image model they can use as it is or train further on their own pictures.

Title: Who uses this model
Industry | Who uses it | How they use it | Example
Software | Developers | Generate images inside their own apps | Call the pipeline from a Python script
Research | Machine learning researchers | Use it as a base to finetune | Train a specialized model on a custom set of images
Creative work | Artists and designers | Make pictures from written prompts | Describe a portrait and a lighting style, then save the result

If you want the faster version, you can skip it and look at Chroma1-Flash instead.

Where to use it:

• Python scripts that use the diffusers library
• ComfyUI, which the card lists as another way to run it
• Windows, macOS or Linux with Python

Title: Jobs it does
Use case | Who | What they get back
Text to image | Anyone with a prompt | A saved image file made from the description
Custom model base | Researchers and artists | A starting checkpoint for finetuning
Quantized runs | Developers short on GPU time | Faster generation with the gemlite sample

How to set it up: You need Python and a GPU, since the card's samples are written for one.

• Python with pip
• About 17.8 GB of disk space for BF16 weights, or about 4.4 GB for a 4-bit copy by Hugging Face's estimate
• The four packages in the install line

1. Install the requirements with pip.
2. Save the sample script below as a Python file.
3. Run it, and the image is written to chroma.png.

pip install transformers diffusers sentencepiece accelerate

How to use it: Change the prompt in the script and run it again. The negative prompt lists what you want to avoid, and the seed makes a run repeatable. The card also has a second sample that uses gemlite for quantized, faster runs.

import torch
from diffusers import ChromaPipeline

pipe = ChromaPipeline.from_pretrained("lodestones/Chroma1-HD", torch_dtype=torch.bfloat16)
pipe.enable_model_cpu_offload()

prompt = [
    "A high-fashion close-up portrait of a blonde woman in clear sunglasses. The image uses a bold teal and red color split for dramatic lighting. The background is a simple teal-green. The photo is sharp and well-composed, and is designed for viewing with anaglyph 3D glasses for optimal effect. It looks professionally done."
]
negative_prompt =  ["low quality, ugly, unfinished, out of focus, deformed, disfigure, blurry, smudged, restricted palette, flat colors"]

image = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    generator=torch.Generator("cpu").manual_seed(433),
    num_inference_steps=40,
    guidance_scale=3.0,
    num_images_per_prompt=1,
).images[0]
image.save("chroma.png")

Title: Main parts of the sample
Tool | What it does
ChromaPipeline | Loads the model from the repository
enable_model_cpu_offload | Keeps parts of the model in CPU memory to save GPU memory
negative_prompt | Lists what the picture should avoid
image.save | Writes the result to a file

Example: You describe a close-up fashion portrait of a woman in see-through sunglasses under split teal and red light. The script runs 40 steps and writes the picture to chroma.png.

For newcomers and experts: Think of it as an image generator you describe pictures to in words. A first small test is the sample script as it is. Then change a few words in the prompt and compare the pictures.

Chroma1-HD is a base model on the FLUX.1 architecture, so its value is as a clean checkpoint for finetuning. The sample uses 40 inference steps and a guidance scale of 3.0. For speed, the card shows gemlite quantization with compiled forward passes. Quantizing the transformer layers trades a little precision for faster, lighter runs.

Text prompt and negative prompt
ChromaPipeline on diffusers
40 inference steps
Image saved as chroma.png

Good to know: A separate Flash version exists with guidance baked in for faster generation. Other people have published 11 quantized and 18 fine-tuned versions of Chroma1-HD.

Title: Estimated weight size
BF16: 17.8 GB
4-bit: 4.4 GB

• License: Apache 2.0.
• Cost to run: none named, apart from a GPU.
• Data it reaches: the files come from Hugging Face, and the model then runs on your own computer.
• Not for: fast generation out of the box, since the card points to Chroma1-Flash for that.

On Hugging Face, 13 November 2026

• Task: text to image
• Library: diffusers
• Parameters: 8.9B
• Weights: about 17.8 GB in BF16; about 4.4 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 83,687
• Likes: 457
• License: apache-2.0
• Quantized versions: 11
• Fine-tunes: 18
• Tags: diffusers, safetensors, text-to-image

Links

• Model: https://huggingface.co/lodestones/Chroma1-HD
• Files: https://huggingface.co/lodestones/Chroma1-HD/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download lodestones/Chroma1-HD

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