Private AI

Nano GPT logo
NanoGPT
Nano GPT logo
NanoGPT
Nano GPT logo
NanoGPT
Search
 
 
 
 
All conversations

Create

Chat
Studio

Explore

Models
Pricing
AI Apps
Blog

Library

Gallery
Workspace
Usage & billing
API
Suggestions and bugs
Use cases

|

Terms

|

Privacy

|

Refunds
Usage & billing
Teams
Referrals
Presets
Settings
FAQPress & mediaBrand assetsBenchmarksNano RPCEarn crypto
  1. Models
  2. Image
  3. NVIDIA

Cosmos 3 Super

Try Model

Cosmos 3 Super generates physically grounded, photorealistic images from text prompts with strong scene understanding.

Added Jun 2, 2026

Details

Approx. Price
$0.040 per image
Model type
text-to-image

Settings

Generation controls available for this model.

Images Per Run

Up to 1

Output images

Input Images

N/A

No reference/edit image input

Output Sizes

Square
Portrait
Landscape

Image Size

Default

square_hd

Number of Images

Default

1

Resolution

Default

square_hd

Options (6)

square_hd (Square (1:1)), square (Small square (1:1)), landscape_4_3 (Landscape (4:3)), landscape_16_9 (Landscape (16:9)) +2 more

Benchmarks

Human preference benchmarks sourced from LMArena.

Text to Image

#36 / 92

Arena Score

1149.7

Votes

17,450

Confidence Interval

1143.7 - 1155.7

Published 2026-11-09 · Matched as Cosmos3-Super-Text2Image

LMArena Dataset

Examples

Loading examples…

Related image models

Compare Cosmos 3 Super with similar models from the same provider or model family.

Topaz Super Focus

topaz-super-focus

Topaz Super Focus recovers detail from blurred images while keeping the source resolution. Choose Super Focus V3 for selective refinement or V2 for severe blur recovery. Process one PNG, JPEG, or WebP image and return JPEG or PNG.

MiniMax H3 Image

minimax-h3-image

MiniMax H3 Image generates images from text and edits a single reference image with text instructions. Create scenes, characters, product concepts and illustrations at 1K or 2K, choose from fifteen aspect ratios, and save one PNG, JPEG or WebP image per request. Reference images must be PNG, JPEG or WebP, with each side between 128 and 8192 pixels and an aspect ratio no greater than 4:1.

Seedream 5.5

bytedance/seedream-v5.5

Seedream 5.5 by ByteDance generates images from text and edits images with up to 10 ordered references. Choose from fifteen aspect ratios and 1K, 1.5K, 2K or 3K output. Save as PNG or JPEG, with transparent backgrounds available in PNG.

Seedream 5.5 Layer Decomposition

bytedance/seedream-v5.5/layer-decomposition

Separate one image into a base image followed by transparent PNG layers. Add an optional prompt to guide the split, choose 1K, 1.5K or 2K, and set a limit of 2 to 100 images including the base. The charge uses the selected limit even when fewer images are returned. The base can be PNG or JPEG.

GPT Image 2.6 Flare

gpt-image-2.6-flare

GPT Image 2.6 Flare generates images from text and edits images with up to 16 references. Designed for fast iteration. Choose low, medium, or high complexity independently of five quality levels, with 1K, 2K, or 4K output and 15 aspect ratios. Custom pixel sizes are also supported through the API. Supports PNG, JPEG, WebP, transparent PNG/WebP backgrounds, and transparent PNG markup annotations for regional edits through the API. Also generates animated GIFs from a prompt or from reference images. Prices are estimated before generation and settled from actual token usage.

GPT Image 2.6 Sunburst

gpt-image-2.6-sunburst

GPT Image 2.6 Sunburst generates images from text and edits images with up to 16 references. Designed for detailed compositions. Choose low, medium, or high complexity independently of five quality levels, with 1K, 2K, or 4K output and 15 aspect ratios. Custom pixel sizes are also supported through the API. Supports PNG, JPEG, WebP, transparent PNG/WebP backgrounds, and transparent PNG markup annotations for regional edits through the API. Also generates animated GIFs from a prompt or from reference images. Prices are estimated before generation and settled from actual token usage.