Tencent's Hunyuan Image 3 Instruct model for text-to-image generation and instruction-based image editing. Supports flexible sizing (256-1536), multiple aspect ratios, bilingual prompts, and reproducible seeds.
Added Jan 29, 2026
Approx. Price
$0.450 per image
Model Type
both
Settings
Generation controls available for this model.
Images Per Run
Up to 4
Output images
Input Images
Up to 2
Reference/edit images accepted • Route max 30 MB
Output Sizes
Custom Resolution
Number of Images
Default
1
Resolution
Default
1024*1024
Options (8)
1024*1024 (Square (1:1)), 1024*1536 (Portrait (2:3)), 1536*1024 (Landscape (3:2)), 768*1024 (Portrait (3:4)) +4 more
Seed
Default
-1
Set -1 for random or use a fixed number for reproducible results
Benchmarks
Benchmarks
Human preference benchmarks sourced from LMArena.
Image Edit
#23 / 56
Arena Score
1302.4
Votes
307,705
Confidence Interval
1299.1 - 1305.8
Published 2026-09-22 · Matched as hunyuan-image-3.0-instruct
LMArena DatasetExamples
Loading examples…
Related image models
Compare Hunyuan Image 3 Instruct with similar models from the same provider or model family.
Hunyuan Image 3
hunyuan-image-3State-of-the-art text-to-image model producing high-quality, emotionally resonant visuals with strong prompt adherence. Supports flexible sizing and reproducible seeds.
Qwen Image 2.1 Edit
wavespeed-ai/qwen-image-2.1/editEdit from up to ten reference images with natural-language instructions, strong subject preservation, flexible framing, and native output up to 2K.
Qwen Image 2.1 Edit LoRA
wavespeed-ai/qwen-image-2.1/edit-loraEdit from up to ten references while applying up to three custom LoRAs for precise identity, product, or style control at up to 2K.
Qwen Image 2.1
wavespeed-ai/qwen-image-2.1/text-to-imageCreate high-quality images with strong prompt following, multilingual text rendering, flexible framing, and native output up to 2K.
Qwen Image 2.1 LoRA
wavespeed-ai/qwen-image-2.1/text-to-image-loraGenerate Qwen Image 2.1 artwork with up to three custom character, product, or style LoRAs and native output up to 2K.
Bria Product Holding
bria/product-holdingPlace one to three referenced products naturally into a person's hands while preserving the subject and scene. Built for e-commerce, lifestyle advertising, influencer content, and product campaigns.