<h1 id="promptwaffle-wf-krea2-lora-strength-testing-ladder-x6">Promptwaffle WF - Krea2 LoRA Strength Testing Ladder X6</h1>A purpose-built ComfyUI workflow for quickly testing a Krea2 LoRA at six different strengths using the exact same prompt, seed, latent, and sampling settings.The workflow generates six separate images at:<code>0.25 · 0.50 · 0.75 · 1.00 · 1.25 · 1.50</code>This makes it easy to see where a LoRA starts working, where it looks best, and where it begins overpowering the base model.<h2 id="choose-your-krea2-checkpoint">Choose Your Krea2 Checkpoint</h2><h3 id="uncensored-version">Uncensored Version</h3>For unrestricted local generation, use:<a target="_new" rel="ugc" href="https://civarchive.com/models/2777907/my-krea2-unlocked-int8-convrot-diffusion-model?modelVersionId=3128223">My Krea2 Unlocked INT8 ConvRot</a>The workflow is configured for:<code>myKrea2UnlockedInt8_v10.safetensors</code>Place it in:<code>ComfyUI/models/diffusion_models/</code><h3 id="official-krea2-version">Official Krea2 Version</h3>Those who do not need the unlocked model can use the official Krea2 Turbo INT8 checkpoint:<a target="_new" rel="ugc" href="https://huggingface.co/Comfy-Org/Krea-2/resolve/main/diffusion_models/krea2_turbo_int8_convrot.safetensors">Official Krea2 Turbo INT8 ConvRot</a>Select the downloaded model in the Krea-2 Diffusion Model node.<h2 id="what-the-workflow-does">What the Workflow Does</h2>One positive prompt and one shared seed are sent through six independent branches. Every branch uses identical noise and sampling settings, with only the LoRA strength changing.Each result is automatically decoded and saved into its own strength-labeled folder:<ul><li><code>str_0.25</code></li><li><code>str_0.50</code></li><li><code>str_0.75</code></li><li><code>str_1.00</code></li><li><code>str_1.25</code></li><li><code>str_1.50</code></li></ul>This gives you a clean, repeatable comparison without manually changing the LoRA strength and generating each image separately.<h2 id="recommended-settings">Recommended Settings</h2><ul><li>Resolution: 1024 × 1024</li><li>Steps: 8</li><li>CFG: 1.0</li><li>Sampler: Euler</li><li>Scheduler: Simple</li><li>Denoise: 1.0</li></ul>Krea2 Turbo is distilled, so increasing CFG above 1.0 will generally reduce quality rather than improve it.The workflow uses Conditioning Zero Out instead of a traditional negative prompt.<h2 id="required-components">Required Components</h2><h3 id="text-encoder">Text Encoder</h3><a target="_new" rel="ugc" href="https://huggingface.co/Comfy-Org/Krea-2/resolve/main/text_encoders/qwen3vl_4b_fp8_scaled.safetensors">qwen3vl_4b_fp8_scaled.safetensors</a>Place in:<code>ComfyUI/models/text_encoders/</code>The CLIP Loader type must be set to:<code>krea2</code><h3 id="vae">VAE</h3><a target="_new" rel="ugc" href="https://huggingface.co/Comfy-Org/Krea-2/resolve/main/vae/qwen_image_vae.safetensors">qwen_image_vae.safetensors</a>Place in:<code>ComfyUI/models/vae/</code><h2 id="required-custom-node">Required Custom Node</h2><h3 id="comfyui-lora-manager">ComfyUI-Lora-Manager</h3><a target="_blank" rel="ugc" href="https://github.com/willmiao/ComfyUI-Lora-Manager">https://github.com/willmiao/ComfyUI-Lora-Manager</a>This provides the six Lora Loader (LoraManager) nodes used by the strength ladder.Everything else in the workflow uses standard ComfyUI nodes.<h2 id="important-before-running">Important Before Running</h2>Load the same LoRA into all six LoRA Loader nodes. Each node should keep its assigned strength:<ul><li>0.25</li><li>0.50</li><li>0.75</li><li>1.00</li><li>1.25</li><li>1.50</li></ul>Do not leave the LoRA fields empty, or all six branches will generate the same base-model image.The strengths must also remain different. When two branches have identical inputs, ComfyUI may reuse the cached result instead of running a separate comparison.Built by Promptwaffle for fast LoRA testing, strength selection, epoch comparisons, and repeatable Krea2 evaluations.

Files with SHA256: 59dae870a6588e80a704272cafdd0c8dd7fd3f475b6f642a632cb45727a6d4a8