<h1 id="xelion-prompt-and-lora-scheduling-example-workflow">XeLion – Prompt &amp; LoRA Scheduling Example Workflow</h1>This is a supporting/example ComfyUI workflow for <a target="_blank" rel="ugc" href="https://github.com/ajayrajsharma79/ComfyUI-XeLion">ComfyUI-XeLion</a>, demonstrating how prompt conditioning and LoRA influence can be scheduled across different stages of the sampling process.The workflow is built around two custom nodes:<ul><li>Multi Prompt Schedule — progressively changes the positive prompt during sampling.</li><li>Multi LoRA Schedule — activates different LoRAs during specific sampling percentage ranges.</li></ul><h2 id="what-this-workflow-demonstrates">What this workflow demonstrates</h2>Instead of applying the same prompt and LoRA configuration for the entire generation, you can divide the sampling process into stages.<h3 id="prompt-scheduling">Prompt scheduling</h3>The example uses three progressive prompt stages:0–30% — Base / CompositionEstablishes the primary image structure:<ul><li>Character identity</li><li>Overall appearance</li><li>Body proportions</li><li>Clothing</li><li>Pose</li><li>Environment</li><li>Basic lighting</li></ul>30–60% — RefinementThe second prompt builds on the established image and adds/refines:<ul><li>Facial expression</li><li>Hair movement</li><li>Clothing detail</li><li>Body posture</li><li>Atmospheric lighting</li><li>Composition and depth</li><li>Environmental details</li></ul>60–100% — Detail / FinishThe final prompt concentrates on:<ul><li>Facial details</li><li>Hair strands</li><li>Skin texture</li><li>Fabric texture</li><li>Fine clothing details</li><li>Hands and fingers</li><li>Lighting and shadows</li><li>Photorealism</li><li>High-frequency detail</li></ul>The important idea is progressive refinement rather than simply dividing one prompt into three unrelated pieces.<h2 id="lora-scheduling">LoRA scheduling</h2>The workflow also demonstrates using different LoRAs during different portions of sampling.The example includes:<ul><li>Dynamic Poses — 0–30%</li><li>Hourglass — 30–60%</li><li>EyesXL — 60–100%</li><li>Add Detail XL — 70–100%</li></ul>This allows different LoRAs to influence different aspects of the generation at the point where they are most useful.For example, a pose LoRA can be active during the early structural phase, while detail-oriented LoRAs can be introduced later in the sampling process.<h2 id="workflow">Workflow</h2>The example uses:<ul><li>AnimagineXL Realistic v0.8</li><li>25 sampling steps</li><li>Euler sampler</li><li>Beta scheduler</li><li>CFG 3</li><li>SDXL resolution of 832 × 1216</li><li>XeLion <code>MultiPromptSchedule</code></li><li>XeLion <code>MultiLoraSchedule</code></li></ul>The workflow is intended primarily as a demonstration of scheduling, so you can replace the checkpoint, prompts, LoRAs, sampler settings, and other components with your own setup.<h2 id="comfyui-xelion">ComfyUI-XeLion</h2>The custom nodes are available here:<a target="_blank" rel="ugc" href="https://github.com/ajayrajsharma79/ComfyUI-XeLion/">https://github.com/ajayrajsharma79/ComfyUI-XeLion/</a>Installation:<ol><li>Clone or copy the repository into: <code>ComfyUI/custom_nodes/ComfyUI-XeLion</code></li><li>Restart ComfyUI.</li><li>The nodes will be available under the scheduling/hooks section.</li></ol><h3 id="multi-prompt-schedule">Multi Prompt Schedule</h3>The Multi Prompt Scheduler accepts a CLIP input and produces a combined conditioning output.Each prompt has an end point between 0.000 and 1.000.For example:<pre><code>Prompt 1 → 0.30
Prompt 2 → 0.60
Prompt 3 → 1.00</code></pre>This results in:<pre><code>0–30% → Prompt 1
30–60% → Prompt 2
60–100% → Prompt 3</code></pre><h3 id="multi-lora-schedule">Multi LoRA Schedule</h3>Each LoRA can independently be assigned:<ul><li>Enable/disable</li><li>Model strength</li><li>Schedule start %</li><li>Schedule end %</li></ul>For example:<pre><code>Dynamic Pose LoRA → 0–30%
Body/Shape LoRA → 30–60%
Eye LoRA → 60–100%
Detail LoRA → 70–100%</code></pre>The LoRA remains at its configured strength during its scheduled interval and is inactive outside that interval.<h2 id="why-schedule-prompts-and-loras">Why schedule prompts and LoRAs?</h2>The goal is to give different conditioning concepts influence at different points in the denoising process.A useful way to think about it is:Early sampling → structureMiddle sampling → refinementLate sampling → detailThis can be particularly useful when you want to separate structural concepts such as pose and composition from later-stage concepts such as facial detail, material texture, lighting, or fine rendering characteristics.Experiment with different timing ranges and strengths for your model and LoRA combination. The optimal schedule will vary depending on the checkpoint, sampler, scheduler, number of steps, and LoRA.<hr />Custom nodes: <a target="_blank" rel="ugc" href="https://github.com/ajayrajsharma79/ComfyUI-XeLion/">https://github.com/ajayrajsharma79/ComfyUI-XeLion/</a>Workflow: Included with this post.Feedback, bug reports, and suggestions are welcome.

Files with SHA256: 6bf600a2e4665a1b077caf0a7676e1fe3d41a7118faf4fbc0059bfd56d00132a