IMAGE · Krea (Open Weights)

Krea 2 Raw

Access to Krea 2 Raw is gated. Please provide the information below and agree to the license to request access.

SELF-HOSTVerified Trust BadgeComfyUI✓ Fits 24GB GPU (+19GB spare)

ENTRY COSTFree weights · $0

EXECUTIONCOMFYUI & LOCAL PIPELINE

LICENSEOpen-Weight Terms

CONTEXT / TASKText-to-Image / ComfyUI

Original Author (HF) ↗Verified Trust Certificate ← Back

The brutal truth

EDITORIAL VERDICT · SELF-HOST

Audited in Stactora CMS. Architecture verified with 12.8B parameters.

HUGGING FACE REPOSITORY · /tree/main

Files and Versions

Direct inspected tree artifacts from Hugging Face. Verified filenames, exact byte sizes, model weight shards, and configuration definitions.

Browse /tree/main↗

Discovered FilesVerified Repository

Source Repokrea/Krea-2-Raw

Files and versions are hosted on Hugging Face. You can view all branches, commits, model weights, and tokenizer artifacts directly.

Inspect krea/Krea-2-Raw /tree/main ↗Inspect /tree/main Repository ↗

✓ COMFYUI VERIFIED

Easy-to-Run ComfyUI Package with Official Workflow JSON

1. Model / Transformer: krea_2_raw_bf16.safetensors

ComfyUI/models/diffusion_models/

2. Text Encoder: text_encoder.safetensors

ComfyUI/models/text_encoders/

3. VAE (Autoencoder): ae.safetensors

ComfyUI/models/vae/

Fast 3-Step Setup with ComfyUI:

Download weights from the official Hugging Face repository.
Place each file into its corresponding ComfyUI/models/ subfolder listed in the manifest table above.
Click "⬇ Download Workflow JSON", then drag and drop the downloaded .json file into your open ComfyUI canvas.

Decision matrix

✓ RUN / USE THIS IF…

•You have a GPU with sufficient VRAM
•Your 24GB GPU comfortably fits the model weights (~5GB VRAM) with dedicated headroom for inference.
•You require 100% offline data privacy with zero third-party API logging, subscriptions, or prompt retention.
•You build visual pipelines in ComfyUI or Automatic1111 and need repeatable, node-based diffusion workflows.

✕ WALK AWAY IF…

•Your system lacks sufficient dedicated VRAM
•Your GPU has less than 5GB VRAM and you cannot tolerate slow CPU/system RAM offload paging.
•You prefer a plug-and-play web app interface and don't want to manage local model checkpoints, drivers, or runtimes.
•You need specialized frontier reasoning that exceeds this model's parameter envelope.

True cost of ownership

Upfront$0Weights free

Hardware5GB VRAMat Q2_K (2-bit Extreme)

API MarkupNoneYou own the stack

Est. Net/Mo$0–$40Electricity + storage

About

Access to Krea 2 Raw is gated. Please provide the information below and agree to the license to request access.

What it does

Download the weights from Hugging Face, pick a quantisation that matches your GPU, and run it locally with Ollama, llama.cpp, vLLM, or a compatible runtime. undefined runs on your own hardware — no per-token bill, no third-party data handling.

Key features

Use cases

01Concept art

Explore visual directions before committing to production.

02Product mockups

Place products in editorial scenes without a photo shoot.

03Marketing visuals

Generate campaign imagery with consistent art direction.

04Editorial illustration

Illustrate articles and decks with a consistent visual voice.

Quantization & Hardware Fit

✓ HARDWARE COMPATIBLESelected Profile: GeForce Rig (24 GB VRAM)

VRAM Hardware Compatibility & Memory Allocation

Test GPU:

Active Quant: Q2_K (2-bit Extreme) · Required: 5 GB+19 GB Headroom Available

Execution Speed~141 tokens/sec

Memory Bandwidth1008 GB/s

Hardware StatusRuns great

GRADE S · 99/100Runs great (~141 tok/s on 24GB · 1008 GB/s)

Quant	VRAM	Quality / Fit on 24GB	Install / Run Command (COMFYUI)
Q2_K (2-bit Extreme)● ACTIVE	5 GB	55% quality retention · ✓ Fits (19GB spare)	`comfy model download --url https://huggingface.co/krea/Krea-2-Raw --relative-path ComfyUI/models/diffusion_models/`
Q3_K_M (3-bit Compact)	7 GB	68% quality retention · ✓ Fits (17GB spare)	`comfy model download --url https://huggingface.co/krea/Krea-2-Raw --relative-path ComfyUI/models/diffusion_models/`
Q4_K_M (4-bit Sweet Spot)	8.5 GB	82% quality retention · ✓ Fits (15.5GB spare)	`comfy model download --url https://huggingface.co/krea/Krea-2-Raw --relative-path ComfyUI/models/diffusion_models/`
Q5_K_M (5-bit High Precision)	11 GB	91% quality retention · ✓ Fits (13GB spare)	`comfy model download --url https://huggingface.co/krea/Krea-2-Raw --relative-path ComfyUI/models/diffusion_models/`
Q6_K (6-bit Studio)	13 GB	96% quality retention · ✓ Fits (11GB spare)	`comfy model download --url https://huggingface.co/krea/Krea-2-Raw --relative-path ComfyUI/models/diffusion_models/`
Q8_0 (8-bit Reference)	16 GB	99% quality retention · ✓ Fits (8GB spare)	`comfy model download --url https://huggingface.co/krea/Krea-2-Raw --relative-path ComfyUI/models/diffusion_models/`
F16 (16-bit Unquantized)	29 GB	100% quality retention · ✕ Needs 29GB	`comfy model download --url https://huggingface.co/krea/Krea-2-Raw --relative-path ComfyUI/models/diffusion_models/`

LIVE INFERENCE SPEED SIMULATOR (BANDWIDTH: 1008 GB/S)

Feel how fast Krea 2 Raw (Q2_K (2-bit Extreme)) runs on your 24GB rig (~141 tok/s)

Click "▶ Simulate Live Speed" above to watch tokens stream at your GPU's real memory-bandwidth speed (141 tokens/sec ≈ 106 words/sec) before you download 5 GB of weights.

Buyer doubts, cleared

Can I actually run Krea 2 Raw on my 24GB GPU?

Yes. At Q2_K (2-bit Extreme) the model needs ~5GB, and your 24GB fits it. Expect roughly 25-45 tok/s at Q4 on a modern card. Add 1-2GB headroom for KV cache at 8K context.

What's the fine print on the license?

License: undefined. This is a restricted license — review the terms carefully before using it in a commercial product. Some restrict resale, competitive use, or require attribution.

Does the context window actually hold up in practice?

Benchmark claims for long context often degrade past 64K tokens. Test your specific retrieval workload before committing. For anything above 32K, run a needle-in-a-haystack test first.