Krea 2 Me (Experimental)

CHECKPOINT

Krea 2 Me is an experimental checkpoint built upon the Krea Image / Krea 2 base architecture. Operating on a high-performance Base Turbo foundation, this model is specifically tuned to push the boundaries of high-fidelity realism and expansive, breathtaking scenery, bringing professional-grade aesthetics to fast-generation workflows.

⚠️ Disclaimer & Experimental Status

This model is currently in its Initial Experimental Phase. Our main objective with this release is testing how the dataset behaves under the new Krea 2 Turbo framework.

Note: Because this is an early experimental build, you may encounter occasional inconsistencies, artifacts, or unexpected outputs. Your feedback and posted generations are highly appreciated and will be essential in refining future iterations.

🔞 NSFW Warning: Since the underlying Krea base model natively supports adult content, this checkpoint is also capable of generating NSFW material. Please use it responsibly and at your own discretion.

💡 Key Characteristics

Base Turbo Speed: Combines the fast generation speed of a turbo model with the deep texture resolution of a full base architecture.
Scenery & Environment: Exceptionally tuned for vast landscapes, intricate interior designs, and atmospheric world-building.
Advanced Realism: Focused on authentic surface textures, complex lighting interaction, and cinematic depth of field.

🛠️ Training Specifications

Main Dataset: 3,000 High-Quality Curated Images (Focus on Realism & Scenery).
Regularization Dataset: 5,000 Balancing Samples.
Total Training Data: 8,000 Samples.
Base Foundation: Krea Image / Krea 2 (Turbo Base).

📝 Prompting Tips

To fully leverage the Krea 2 architecture, incorporate rich environmental descriptors and photographic terms:

Positive Tags: masterpiece, photorealistic, highly detailed scenery, cinematic lighting, raw photo, epic landscape, 8k resolution, intricate details.
Negative Prompts: cartoon, anime, lowres, blurry, deformed, bad anatomy, oversaturated, plastic textures.

⚙️ Recommended Settings

This model utilizes a specialized setup tailored for fast yet high-quality Turbo generation. For optimal results, please adhere to the following parameters:

Sampler: euler_a_sde (or alternative SDE samplers)
Scheduler: beta
Sampling Steps: 8 – 30 (Start at 8 for quick drafts, increase up to 30 to fully mature complex scenery detail)
CFG Scale: 1.0 (Strictly recommended for this Turbo foundation)
Clip Skip: 2

Show Me Your Creations!

Since this model is highly experimental, I would love to see how it performs in your projects. Please share your renders and leave your honest feedback in the gallery section below to help shape the next version!