Choose theme
AI Visuals
Krea 2 Turbo 4-Step LoRA
Krea 2 Turbo 4-Step LoRA is a community-made adapter that reduces Krea 2 Turbo image generation from eight steps to four.
The project provides ordinary Diffusers weights, ComfyUI-ready weights, and a stock ComfyUI workflow. The shipped rank-64 adapter has separate Diffusers and ComfyUI files; earlier checkpoints are archived and no longer maintained. Use this as a first read, not a recommendation. Open the original project before trusting details like terms, limits, privacy, cost, setup, or safety.
What it is
A four-step adapter for Krea 2 Turbo
This is a LoRA trained against Krea 2 Turbo and designed to load on top of Turbo rather than the Raw checkpoint or a standalone image model. Its job is to recover detail when Turbo's ordinary eight-step schedule is shortened to four steps.
Why it stands out
Fewer steps without a custom sampler
The adapter uses Turbo's own four-point deployment schedule and works through standard LoRA loading. The project supplies files for Diffusers and ComfyUI instead of requiring custom nodes or a patched sampler.
Availability
Model files, workflow, and a hosted demo
Hugging Face hosts the released safetensor adapter, a ready ComfyUI workflow, and an independent hosted demo. Local use still requires a compatible Krea 2 Turbo base model and its supporting components.
Why it matters
What makes it useful
Local image-model iteration can be slow when every prompt needs a full sampling run. This LoRA offers Krea 2 Turbo users a documented four-step route for quicker comparisons while keeping the parent model, prompt, and ordinary workflow in place.
What to know
Where it fits
In Diffusers, load Krea 2 Turbo first, then load the adapter and generate with four inference steps. The supplied ComfyUI workflow provides the corresponding path through stock nodes.
Notable points
What stands out
The maintainer reports that the current adapter is faster than an eight-step Turbo run in a small Apple Silicon test and publishes fixed-seed image comparisons. Those are project-reported results, not an independent benchmark. Four-step Turbo without the adapter was quicker still in that test; the LoRA adds texture correction at some extra inference cost.
Before using
What to review
This is not a standalone model. Prepare Krea 2 Turbo and its text encoder and VAE before adding the LoRA.
Do not treat it as a general Krea 2 adapter. The maintainer says Krea 2 Raw is an unsupported path with mixed results; the intended setup is Turbo at four steps.
Use the settings for the interface you choose. The model card lists guidance 0.0 for Diffusers and CFG 1.0 for ComfyUI, where the two values represent the same no-guidance behavior.
Use the current rank-64 file for your interface and follow its loading instructions. Earlier checkpoints are superseded and no longer maintained.
Plan for a large local setup. The project's ComfyUI table lists Krea 2 Turbo diffusion-model variants around 13 to 26 GB, plus a separate text encoder and VAE.
The adapter declares the Krea 2 Community License and includes LICENSE.pdf and NOTICE.txt. Review its current terms and the parent-model terms for your intended use.
Reader fit
Who may find it relevant
Krea 2 Turbo users comparing four-step and eight-step local generation.
Diffusers and ComfyUI builders who already understand LoRA-based image workflows.
Creators comparing the adapter's texture correction with ordinary four-step Turbo.
Less relevant for readers seeking a small standalone model, a self-contained download or a simple consumer image editor.
Editorial note
Why LifeHubber lists it
To see what the adapter changes, compare the second and third panels in the model card's same-seed examples: four-step Turbo without the LoRA and four-step Turbo with it. Both use four steps, so this comparison isolates the adapter's effect rather than mixing it with the step-count change.
Source links
Source materials
Reader note
Before relying on this entry
LifeHubber lists entries to help readers inspect AI projects, not to endorse them or prove they are safe, suitable, accurate, maintained, or right for a specific use. We do not verify every entry in depth. Before relying on anything listed, review the original materials, terms, privacy practices, limits, and risks that matter for your situation.
What to explore next
Start with Krea 2 Turbo, the model behind this adapter.
The LoRA changes Krea 2 Turbo's sampling schedule, but it still depends specifically on Turbo, its large components, and its access and license terms.
More in AI Models
Keep browsing this category
Explore more AI model resources.
Gemma 4
google/gemma-4
A Google DeepMind Gemma 4 model family collection with public checkpoints including Gemma 4 12B, a dense multimodal model Google describes around local agentic workflows, native audio input, and encoder-free vision/audio handling.
MiniMax H3 Integrations
MiniMax-AI/awesome-minimax-h3-integration
A community-maintained MiniMax H3 integration index that maps checkpoints, hardware and VRAM starting points, runtimes, ComfyUI nodes, prompting tools, acceleration routes, and deployment options.
OLMoEarth
allenai/olmoearth
An Ai2 remote-sensing foundation model family for satellite imagery and planetary-scale mapping, with v1.2 Nano, Tiny, Small, and Base models, model weights, training code, and a technical report.
For project maintainers
Listed here? You can use the badge.
If you maintain a project with a current LifeHubber listing, you may add the optional “Listed on LifeHubber AI Resources” badge to its README, docs, or website. No introduction or permission request is needed.
