SPARK.Chroma
Collection
2 items • Updated • 3
import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("SG161222/SPARK.Chroma_preview", dtype=torch.bfloat16, device_map="cuda")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]SPARK.Chroma — Model for Realistic Image Generation
Overview
SPARK.Chroma is a fine-tuned model designed for generating realistic images. Based on the Chroma1-HD model with 8.9B parameters and Apache 2.0 license.
Warning: This model is a preview release and does not yet reflect the full vision — outputs may lack realism and fine detail, and in some cases it may be inferior to Chroma1‑HD. Future releases will use different fine‑tune settings to improve quality.
SPARK.Chroma_preview
START DATE:
October 14, 2025
LAST UPDATE:
October 30, 2025
DATASET SIZE:
2400 Images
EQUIPMENT:
1 x RTX 4090 (24GB)
TRAINING TIME:
~ 116h 40m
You can support me directly on my Boosty
How to Use
For ComfyUI:
- Place the ae.safetensors in the following path: ComfyUI_windows_portable\ComfyUI\models\vae
- Place t5xxl_fp16.safetensors in the following path: ComfyUI_windows_portable\ComfyUI\models\clip
- Place the SPARK.Chroma_preview.safetensors in the following path: ComfyUI_windows_portable\ComfyUI\models\unet or ComfyUI_windows_portable\ComfyUI\models\diffusion_models
- Load the BasicWorkflow.json into ComfyUI and run
Optimal Generation Settings
- Sampling Method and Schedule Type: dpmpp_2m simple
- Sampling Steps: 40
- CFG Scale: 3.5-4.0
- Negative Prompt: low quality, ugly, unfinished, out of focus, deformed, disfigure, blurry, smudged, restricted palette, flat colors, noisy, artifacts, fake, generated
Limitations
- This version of the model does not have a variety of styles and is intended for realism.
- There may be issues with anatomy and text.
- Downloads last month
- 233