Image Generation

Generate high-quality images from text prompts and edit existing images with Zen's 12B and 7B diffusion models, optimized for speed, creativity, and precision.

Zen's Image Generation family offers a comprehensive suite of diffusion models for text-to-image synthesis and image-to-image editing. Built on Mixture of Distilled Experts (MoDE) architecture, these models deliver photorealistic output with precise prompt following, specialized variants for speed and creativity, and Japanese-optimized generation.

Models

Model	Params	Resolution	Pipeline	HF	Paper
Zen3 Image	12B	1024px	text-to-image	weights	paper
Zen3 Image Fast	12B	1024px	text-to-image	weights	paper
Zen3 Image Max	12B	2048px	text-to-image	weights	paper
Zen3 Image Playground	12B	1024px	text-to-image	weights	paper
Zen3 Image Dev	12B	1024px	text-to-image	weights	paper
Zen3 Image SDXL	12B	1024px	text-to-image	weights	paper
Zen3 Image JP	12B	-	text-to-image	weights	paper
Zen3 Image SSD	12B	-	text-to-image	weights	paper
Zen Image Edit	7B	1024px	image-to-image	weights	paper

Quick Start

Text-to-Image with Transformers

from diffusers import AutoPipelineForText2Image
import torch

# Load the Zen3 Image model
model_id = "zenlm/zen3-image"
pipe = AutoPipelineForText2Image.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16
)
pipe = pipe.to("cuda")

# Generate an image
prompt = "A serene mountain landscape at sunset, photorealistic, cinematic lighting"
image = pipe(prompt).images[0]
image.save("output.png")

Image Editing with Zen Image Edit

from diffusers import AutoPipelineForInpainting
import torch
from PIL import Image

model_id = "zenlm/zen-image-edit"
pipe = AutoPipelineForInpainting.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16
)
pipe = pipe.to("cuda")

# Load your base image and mask
image = Image.open("base_image.png")
mask_image = Image.open("mask.png")  # White regions to edit

# Generate edited image
prompt = "A red sunset sky"
edited_image = pipe(
    prompt=prompt,
    image=image,
    mask_image=mask_image
).images[0]
edited_image.save("edited.png")

Zen API Access

For production deployments, use the OpenAI-compatible Zen API endpoint:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.hanzo.ai/v1",
    api_key="your-api-key"
)

# Generate image via API
response = client.images.generate(
    model="zen3-image",
    prompt="A serene mountain landscape at sunset, photorealistic",
    size="1024x1024",
    quality="hd"
)

print(response.data[0].url)

Get your API key at console.hanzo.ai — $5 free credit on signup.

Model Variants

Zen3 Image — Flagship photorealistic text-to-image model with precise prompt adherence.

Zen3 Image Fast — Optimized for low-latency inference, ideal for real-time generation workflows.

Zen3 Image Max — Extended to 2048px resolution for maximum detail and quality.

Zen3 Image Playground — Creative variant tuned for experimental and artistic outputs.

Zen3 Image Dev — Developer-focused variant for research and fine-tuning.

Zen3 Image SDXL — Compatible with SDXL tooling and ecosystem.

Zen3 Image JP — Japanese-language optimized diffusion for culturally fluent visuals.

Zen3 Image SSD — Lightning-fast SSD-based diffusion for instant generation.

Zen Image Edit — Instruction-following image editor for targeted modifications and inpainting.