Cosmos3-Super-Text2Image by Ares Realm Studios: Open Model

Runs On

What it takes to serve Cosmos3-Super-Text2Image (64.6B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.

Precision	Weights	Memory needed	Cheapest setup	Per hour	Also fits
16-bit	129.2 GB	155.0 GB	1x MI300X (192 GB) Vultr	$1.85	1x MI325X $2.00 · 1x MI355X $2.59
8-bit	64.6 GB	77.5 GB	1x MI300X (192 GB) Vultr	$1.85	1x H100 $1.99 · 1x MI325X $2.00
4-bit	32.3 GB	38.8 GB	1x MI300X (192 GB) Vultr	$1.85	1x H100 $1.99 · 1x MI325X $2.00

Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Oct 16, 2026.

Cosmos3-Super-Text2Image on every accelerator the SAVRN Index prices, at every precision

Model Card

NVIDIA Cosmos™ is a world foundation model platform designed to accelerate the development of Physical AI by enabling machines to understand, simulate, and interact with the physical world across robotics, autonomous driving, and smart space environments, including industrial and factory-scale applications. Cosmos3 is a collection of Omnimodal world models capable of generating dynamic, high-quality video, image, audio, and action commands from combinations of text, image, video, and action trajectory inputs. It serves as a foundational building block for a broad range of Physical AI applications and research spanning world understanding, world generation, simulation, and embodied policy…

Excerpt from the card by Ares Realm Studios, licensed other.

Configuration

Architecture: Cosmos3ForConditionalGeneration
Context length (tokens): 262,144
Layers: 64
Hidden size: 5,120
Feed-forward size: 25,600
Attention heads: 64
Key/value heads: 8
Head dimension: 128
Vocabulary size: 151,936
RoPE base: 5,000,000
Model type: cosmos3_omni

Identity and Version

Repository: Ares-Realm-Studios/Cosmos3-Super-Text2Image
Publisher: Ares Realm Studios
Task: Text to image
Modality: Image
Library: cosmos
Parameters: 64.6B parameters
Languages: Not stated by the source
Revision: 371b70d7d049311010dedc9e843477994a0e65fb
First published: 2026-10-03
Last updated: 2026-10-03

Files and Weights

84 files, 132.5 GB in total. The weights are 30 files totalling 132.5 GB in safetensors.

Weights30 files · 132.5 GB

Configuration30 files · 458.5 KB

Tokenizer8 files · 27.4 MB

Documentation6 files · 44.0 KB

Other8 files · 10.2 MB

Repository2 files · 1.7 KB

Every file

File	Type	Size	SHA-256
sound_tokenizer/diffusion_pytorch_model.safetensors	Weights	1.9 GB	9d4c61cde38a
transformer/diffusion_pytorch_model-00001-of-00027.safetensors	Weights	4.9 GB	28fe2fcd84de
transformer/diffusion_pytorch_model-00002-of-00027.safetensors	Weights	4.8 GB	9573ad1d6561
transformer/diffusion_pytorch_model-00003-of-00027.safetensors	Weights	4.9 GB	a6d1b8462e97
transformer/diffusion_pytorch_model-00004-of-00027.safetensors	Weights	4.8 GB	65e0145f7776
transformer/diffusion_pytorch_model-00005-of-00027.safetensors	Weights	4.9 GB	ca34444e30ec
transformer/diffusion_pytorch_model-00006-of-00027.safetensors	Weights	4.8 GB	57ab33ceb73c
transformer/diffusion_pytorch_model-00007-of-00027.safetensors	Weights	4.9 GB	b83bb25e88ae
transformer/diffusion_pytorch_model-00008-of-00027.safetensors	Weights	4.8 GB	a48a7b2aa7e9
transformer/diffusion_pytorch_model-00009-of-00027.safetensors	Weights	4.9 GB	63737abe6c08
transformer/diffusion_pytorch_model-00010-of-00027.safetensors	Weights	4.8 GB	42ab03b9b949
transformer/diffusion_pytorch_model-00011-of-00027.safetensors	Weights	4.9 GB	03be6e218bc8
transformer/diffusion_pytorch_model-00012-of-00027.safetensors	Weights	4.8 GB	b317d06f0adb
transformer/diffusion_pytorch_model-00013-of-00027.safetensors	Weights	4.9 GB	d6513f5245c1
transformer/diffusion_pytorch_model-00014-of-00027.safetensors	Weights	4.8 GB	bdc20ad60439
transformer/diffusion_pytorch_model-00015-of-00027.safetensors	Weights	4.9 GB	3af6c17c430c
transformer/diffusion_pytorch_model-00016-of-00027.safetensors	Weights	4.8 GB	0a65c367ac5c
transformer/diffusion_pytorch_model-00017-of-00027.safetensors	Weights	4.9 GB	0296852ac2e0
transformer/diffusion_pytorch_model-00018-of-00027.safetensors	Weights	4.8 GB	f078e14b0172
transformer/diffusion_pytorch_model-00019-of-00027.safetensors	Weights	4.9 GB	009765e7b69f
transformer/diffusion_pytorch_model-00020-of-00027.safetensors	Weights	4.8 GB	c55e2e02da62
transformer/diffusion_pytorch_model-00021-of-00027.safetensors	Weights	4.9 GB	340e4a006659
transformer/diffusion_pytorch_model-00022-of-00027.safetensors	Weights	4.8 GB	4d019a64084a
transformer/diffusion_pytorch_model-00023-of-00027.safetensors	Weights	4.9 GB	f5b974d3465a
transformer/diffusion_pytorch_model-00024-of-00027.safetensors	Weights	4.8 GB	06cf61bdb756
transformer/diffusion_pytorch_model-00025-of-00027.safetensors	Weights	4.9 GB	9f93556b82a5
transformer/diffusion_pytorch_model-00026-of-00027.safetensors	Weights	4.4 GB	a2aeef5aa284
transformer/diffusion_pytorch_model-00027-of-00027.safetensors	Weights	1.6 GB	d3cf9a5eca42
vae/diffusion_pytorch_model.safetensors	Weights	1.4 GB	230496cb59ff
vision_encoder/model.safetensors	Weights	1.2 GB	3bdba32cec6f
agentic_upsampling/__init__.py	Configuration	261 B	—
agentic_upsampling/__main__.py	Configuration	144 B	—
agentic_upsampling/clients.py	Configuration	20.7 KB	—
agentic_upsampling/constants.py	Configuration	1.2 KB	—
agentic_upsampling/data.py	Configuration	6.3 KB	—
agentic_upsampling/extract_best.py	Configuration	6.3 KB	—
agentic_upsampling/io_utils.py	Configuration	1.7 KB	—
agentic_upsampling/prompt_upsampler.py	Configuration	15.8 KB	—
agentic_upsampling/rubric.py	Configuration	9.0 KB	—
agentic_upsampling/run.py	Configuration	8.4 KB	—
agentic_upsampling/runner.py	Configuration	19.6 KB	—
assets/example_caption.json	Configuration	7.4 KB	—
chat_template.json	Configuration	5.5 KB	—
checkpoint.json	Configuration	2 B	—
config.json	Configuration	8.2 KB	—
generation_config.json	Configuration	269 B	—
model.safetensors.index.json	Configuration	179.9 KB	—
model_index.json	Configuration	502 B	—
modular_model_index.json	Configuration	1.7 KB	—
preprocessor_config.json	Configuration	390 B	—
scheduler/scheduler_config.json	Configuration	884 B	—
sound_tokenizer/config.json	Configuration	1.3 KB	—
tests/test_agentic_upsampling.py	Configuration	19.2 KB	—
text_tokenizer/added_tokens.json	Configuration	707 B	—
text_tokenizer/special_tokens_map.json	Configuration	613 B	—
transformer/config.json	Configuration	1.4 KB	—
transformer/diffusion_pytorch_model.safetensors.index.json	Configuration	138.5 KB	—
vae/config.json	Configuration	1.8 KB	—
video_preprocessor_config.json	Configuration	385 B	—
vision_encoder/config.json	Configuration	509 B	—
AGENTIC_UPSAMPLING.md	Documentation	4.8 KB	—
BIAS.md	Documentation	4.7 KB	—
EXPLAINABILITY.md	Documentation	3.2 KB	—
PRIVACY.md	Documentation	1.2 KB	—
README.md	Documentation	26.4 KB	—
SAFETY.md	Documentation	3.7 KB	—
assets/benchmark-text2image-leaderboard-all-models.jpg	Other	432.2 KB	380c72e4df9a
assets/benchmark-text2image-leaderboard.png	Other	4.1 MB	10458182a0c5
assets/benchmark-text2image.png	Other	147.3 KB	55bdd6bc6178
assets/example_image.png	Other	1.3 MB	478903c6adf0
assets/more_images.jpg	Other	4.3 MB	48efdeacbf98
assets/original_prompt.txt	Other	525 B	—
pytest.ini	Other	67 B	—
text_tokenizer/chat_template.jinja	Other	5.3 KB	—
.gitattributes	Repository	1.7 KB	—
.gitignore	Repository	15 B	—
merges.txt	Tokenizer	1.7 MB	—
text_tokenizer/merges.txt	Tokenizer	1.7 MB	—
text_tokenizer/tokenizer.json	Tokenizer	11.4 MB	aeb13307a71a
text_tokenizer/tokenizer_config.json	Tokenizer	5.4 KB	—
text_tokenizer/vocab.json	Tokenizer	2.8 MB	—
tokenizer.json	Tokenizer	7.0 MB	—
tokenizer_config.json	Tokenizer	10.9 KB	—
vocab.json	Tokenizer	2.8 MB	—

License and Download

License: other
Access: Open weights, no gate
Download size: 132.5 GB

Download from Ares Realm Studios

Released by Ares Realm Studios through its official repository on Hugging Face.

Memory Requirements

Precision	Weights in memory
As published	132.5 GB
16-bit	129.2 GB
8-bit	64.6 GB
4-bit	32.3 GB

Weights only, from the published parameter count; the key-value cache and runtime add to this.

Questions About Cosmos3-Super-Text2Image

How much GPU memory does Cosmos3-Super-Text2Image need?

About 155 GB at 16-bit and 38.8 GB at 4-bit: the weights (64.6B parameters) plus a working margin. A long context needs more.

What is the cheapest GPU to run Cosmos3-Super-Text2Image on?

At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.

What license is Cosmos3-Super-Text2Image released under?

other, as its publisher declares it. Read the license text before commercial use.

What is Cosmos3-Super-Text2Image's context length?

262,144 tokens, from the maximum position embeddings in its published configuration.

Similar Models

Model · Text to image

Qwen-Image

Qwen

We are thrilled to release Qwen-Image, an image generation foundation model in the Qwen series that achieves significant advances in complex text rendering and precise image editing. Experiments show strong general capabilities in both image generation and editing, with exceptional performance in text rendering, especially for Chinese. - 2025.08.04: We released the Technical Report of Qwen-Image! - 2025.08.04: We released Qwen-Image weights! Check at huggingface and Modelscope! - 2025.08.04: We released Qwen-Image! Check our blog for more details! Install the latest version of diffusers The following contains a code snippet illustrating how to use the model to generate images based on text…

Open weights apache-2.0 20.4B parameters diffusers

View model

Model · Text to image

Krea-2-Raw

KREA

This is the Krea 2 Raw checkpoint, its not recommended for inference use. Its a good base for finetuning or post-training for your own needs and domains. For example, one use-case is to train LoRAs on midtrain and directly use them on Krea 2 Turbo. See our collection of in-house trained LoRAs trained on Raw and meant to be used with Turbo: Krea-2 LoRA Collection 1. Setup the official Krea 2 codebase 2. Download raw.safetensors in this repo 3. export OSSRAW= Install diffusers from source (for Krea2Pipeline): Install SGLang from source (https://github.com/sgl-project/sglang) See the full SGLang Krea 2 Cookbook here This model card covers the Krea 2 model family, including the following…

Access requested at publisher other 12.8B parameters diffusers

View model

Model · Text to image

damn

C

Pipeline Class: CraftlyImagePipeline; Transformer Diffusion Backbone: CraftlyImageTransformer2DModel (28 Layers, 48 Attention Heads, 64-channel Latent Input); Text Conditioning Backbone: CraftlyVisionModel (36 Layers, 2560 Hidden Dim); Latent Autoencoder: 16-channel Spatial VAE; Denoising Trajectory: Flow-Matching Euler Discrete Scheduler (Optimized 8-step Distillation).

Open weights other 12.8B parameters diffusers

View model

Model · Text to image

FLUX.1-dev

Black Forest Labs

FLUX.1 [dev] is a 12 billion parameter rectified flow transformer capable of generating images from text descriptions. For more information, please read our blog post. 1. Cutting-edge output quality, second only to our state-of-the-art model FLUX.1 [pro]. 2. Competitive prompt following, matching the performance of closed source alternatives. 3. Trained using guidance distillation, making FLUX.1 [dev] more efficient. 4. Open weights to drive new scientific research, and empower artists to develop innovative workflows. 5. Generated outputs can be used for personal, scientific, and commercial purposes as described in the [FLUX.1 [dev] Non-Commercial…

Access requested at publisher other 11.9B parameters diffusers

View model

Model · Text to image

NSFW_flux_dev

Eric McNair

NSFW Weights for this model are available in Safetensors format. Download them in the Files & versions tab.

Open weights apache-2.0 11.9B parameters diffusers

View model

Model · Text to image

FLUX.1-schnell

Black Forest Labs

FLUX.1 [schnell] is a 12 billion parameter rectified flow transformer capable of generating images from text descriptions. For more information, please read our blog post. 1. Cutting-edge output quality and competitive prompt following, matching the performance of closed source alternatives. 2. Trained using latent adversarial diffusion distillation, FLUX.1 [schnell] can generate high-quality images in only 1 to 4 steps. 3. Released under the apache-2.0 licence, the model can be used for personal, scientific, and commercial purposes. We provide a reference implementation of FLUX.1 [schnell], as well as sampling code, in a dedicated github repository. Developers and creatives looking to…

Access requested at publisher apache-2.0 11.9B parameters diffusers

View model