WAN 2.2 14B Multi-Phase I2V/T2V Workflow

WAN 4-Phase V1.4 - Switch Between I2V/T2V Instantly + Use Different LoRAs Per PhaseVersion NoteV1.4 skips V1.2 and V1.3 - Those versions evolved into V1.4 during development before I posted them publicly. If you want to see V1.2 or V1.3, they're embedded in some of my previously posted videos, but V1.4 is the first "official" release after V1.1.What's Actually NewThree things that make your life easier:<ol><li>Switch between I2V and T2V with one dropdown - no more swapping workflow files</li><li>Independent positive AND negative prompt control per phase - with flexible inheritance modes</li><li>Use different LoRAs for each phase with visual toggle controls</li></ol>That's it. Simple upgrades that give you way more creative control.InstallationYou Need:Workflow File:<ul><li>WAN_4-Phase_V1_4.json</li></ul>Custom Nodes (3 files):<ol><li>smart_sequential_wan_<a target="_blank" rel="ugc" href="http://loader.py">loader.py</a></li><li>multi_phase_prompt_manager_4phase_<a target="_blank" rel="ugc" href="http://v2.py">v2.py</a></li><li>advanced_wan_settings_<a target="_blank" rel="ugc" href="http://controller.py">controller.py</a></li></ol>Steps:<ol><li>Copy the 3 .py files to: ComfyUI/custom_nodes/</li><li>Load the workflow JSON in ComfyUI</li><li>Restart ComfyUI</li><li>Configure your model paths</li><li>You're ready - now read below to understand how to use it</li></ol>Feature 1: One-Click I2V/T2V SwitchingThe Old Way (V1.1):<ul><li>Want to try I2V? Load these models manually</li><li>Want to try T2V? Load different models manually</li><li>Easy to screw up and load wrong ones</li><li>Switching = editing workflow</li></ul>The New Way (V1.4):One dropdown. Pick I2V or T2V. Done.Smart Sequential WAN Loader node:• workflow_mode: [I2V] or [T2V] ← Click and pickThat's it. The workflow loads the right models automatically.What Each Mode Does:I2V Mode:<ul><li>Converts your image to video</li><li>All 4 phases work together progressively</li><li>Phase 1 → Phase 2 → Phase 3 → Phase 4</li><li>Each phase extends the previous phase's video output</li></ul>T2V Mode:<ul><li>Phase 1 generates from text prompt ONLY</li><li>IMPORTANT: Disable Phases 2-4 (set num_active_phases to 1)</li><li>Why? T2V ignores the image input - it only reads text</li><li>Each phase would generate a completely separate video from scratch</li><li>Phase 2 wouldn't extend Phase 1 - it would ignore Phase 1 entirely</li><li>Unless you want 4 disconnected videos stitched together, only use Phase 1</li></ul>Why no T2V→I2V multi-phase workflow?<ul><li>Would need to load T2V models for Phase 1, then unload and load I2V models for Phases 2-4</li><li>That's 4 models total (2 T2V + 2 I2V) which exceeds most VRAM limits</li><li>I'm working on a smart unload/reload system for future versions</li><li>For now, T2V is single-phase only</li></ul>Bottom line: T2V is for single-phase generation only. For multi-phase video extension, use I2V mode.Why This Matters:<ul><li>I2V mode: Use all 4 phases for progressive video extension</li><li>T2V mode: Use Phase 1 only (Phases 2-4 would create separate videos)</li><li>Switch between modes to experiment</li><li>One workflow, both capabilities</li></ul>Feature 2: Independent Prompt Control Per PhaseEach phase gets its own positive AND negative prompt with flexible control modes.How It Works:Phase 1:<ul><li>Set your base positive prompt</li><li>Set your base negative prompt</li><li>These become the foundation</li></ul>Phases 2-4:<ul><li>Choose how each phase handles prompts:<ul><li>Use Base - Copy Phase 1</li><li>Use Previous Phase - Copy last phase</li><li>Unique Prompt - Completely custom</li><li>Append to Base - Add to Phase 1</li><li>Append to Previous Phase - Add to last phase</li></ul></li></ul>Both positive AND negative prompts get these controls independently.Example Workflow:Phase 1: • Positive: "woman walking, city street, confident stride" • Negative: "static, frozen, jittery movement"Phase 2: • Positive Mode: Append to Base • Positive Custom: ", turns head, looks at camera" • Negative Mode: Use Base → Result Positive: "woman walking, city street, confident stride, turns head, looks at camera"Phase 3: • Positive Mode: Append to Previous Phase • Positive Custom: ", raises hand, waves" • Negative Mode: Append to Base • Negative Custom: ", robotic motion" → Result Negative: "static, frozen, jittery movement, robotic motion"Phase 4: • Positive Mode: Unique Prompt • Positive Custom: "smooth motion, cinematic camera movement" • Negative Mode: Use Base → Result Positive: "smooth motion, cinematic camera movement"Why This Matters:Action Control:<ul><li>Direct what happens in each phase of the video</li><li>Add new actions as the video progresses</li><li>Change the scene completely in later phases</li></ul>Motion Quality:<ul><li>Keep consistent motion negatives across all phases (Use Base)</li><li>Add phase-specific motion restrictions (Append modes)</li><li>Completely different motion direction per phase (Unique)</li></ul>No More Manual Copying:<ul><li>Don't manually copy Phase 1 prompt to Phase 2</li><li>Just set the mode and add what's new</li><li>The system handles inheritance automatically</li></ul>Feature 3: Different LoRAs Per PhaseEach of the 4 phases has its own LoRA loader where you can select different LoRAs to influence the video generation.What You Can Do:Each phase gets separate LoRA control (for I2V multi-phase workflows):Phase 1: • Load 3-4 LoRAs that influence initial video generation • Toggle trigger words on/offPhase 2: • Load 3-4 DIFFERENT LoRAs to change how the video extends • Toggle their triggersPhase 3: • Load 3-4 DIFFERENT LoRAs for continued generation • Toggle their triggersPhase 4: • Load 3-4 DIFFERENT LoRAs for final extension • Toggle their triggersResult: Use 12-16 different LoRAs across your video, each influencing different phases.Why Different LoRAs Per Phase?In WAN video generation, LoRAs influence:<ul><li>Motion and movement styles</li><li>Visual effects and aesthetics</li><li>Character appearances and actions</li><li>Camera movement and framing</li></ul>By changing LoRAs per phase, you can:<ul><li>Start with one motion style, transition to another</li><li>Add different effects as the video progresses</li><li>Experiment with progressive visual evolution</li><li>Use specific LoRAs for specific phases</li></ul>Feature 4: Smart Trigger Word ManagementFixed from V1.1: Trigger words no longer accumulate across phases.The V1.1 Problem:In V1.1, triggers would pollute between phases:Phase 1: "walking" trigger Phase 2: "turning" trigger Result: Phase 2 got BOTH "walking" AND "turning" ❌How V1.4 Works:Each phase only gets its own triggers:Phase 1: Gets "walking" ONLY Phase 2: Gets "turning" ONLY Phase 3: Gets "waving" ONLY Phase 4: Gets "smooth_motion" ONLYClean separation - no cross-contamination.Other Improvements:Trigger Positioning:<ul><li>V1.1: Triggers added at end of prompt (often lost)</li><li>V1.4: Triggers at beginning for maximum effect</li></ul>Automatic Deduplication: If you accidentally add the same trigger twice, it's removed automatically.Visual Toggle Controls:Each LoRA shows its trigger words with on/off toggles:<ul><li>See what triggers each LoRA needs</li><li>Click to enable/disable</li><li>No memorizing trigger words</li><li>No typing them manually</li></ul>Carryover Triggers:Each phase has Carryover TriggerWord Toggle nodes that let you pull in trigger words from previous phases:<ul><li>Phase-specific triggers: Use the regular TriggerWord Toggle (only affects that phase)</li><li>Carryover triggers: Use the Carryover TriggerWord Toggle (pulls triggers from earlier phases into current phase)</li></ul>Example:<ul><li>Phase 1: Has "character_name" trigger</li><li>Phase 2: Add "character_name" to Carryover Toggle → Phase 2 now uses Phase 1's "character_name" trigger</li><li>Phase 2: Add "turning" to regular Toggle → Phase 2 also gets its own "turning" trigger</li></ul>This lets you reuse important triggers from earlier phases without them automatically accumulating.Feature 5: Advanced Settings ControllerOne node controls all your video settings.Workflow defaults:<ul><li>base_resolution: 896×1152 (7:9 Portrait)</li><li>upscale_factor: 1.35</li><li>length: 80 frames</li><li>generation_fps: 16</li><li>final_output_fps: 32</li><li>num_active_phases: 2</li><li>batch_size: 1</li></ul>Available options:<ul><li>Resolution: 20+ presets from 576×1024 to 1920×1080 (portrait, landscape, square)</li><li>Upscale: 0.05x to 4.0x in 0.05 increments</li><li>FPS: Both generation and final output adjustable</li></ul>Change once, affects entire workflow.Quick Setup GuideStep 1: Enable/Disable PhasesAdvanced WAN Settings Controller: • num_active_phases: CRITICAL SETTING<ul><li>For I2V: Set to 1, 2, 3, or 4 (controls which phases execute)</li><li>For T2V: MUST be 1 (T2V only works for Phase 1)</li></ul>⚠️ IMPORTANT: This controls the entire workflow execution. Wrong setting breaks everything.You MUST bypass phases you're not using:<ul><li>Using 1 phase? Bypass phases 2, 3, and 4</li><li>Using 2 phases? Bypass phases 3 and 4</li><li>Using 3 phases? Bypass phase 4</li><li>Using 4 phases? Nothing to bypass</li></ul>To bypass phases:<ul><li>Use the "Fast Groups Bypasser" node in the workflow</li><li>Toggle the phases you want to bypass</li></ul>Step 2: Upload Image (I2V Mode Only)If using I2V mode:<ul><li>Load your image in the "Load Image" node</li><li>This becomes the starting frame for Phase 1</li></ul>If using T2V mode:<ul><li>Skip this step - T2V generates from text only</li></ul>Step 3: Write Your PromptsMulti-Phase Prompt Manager: • Phase 1: Base prompt (positive and negative) • Phase 2-4: Choose mode (Use Base, Append, Unique, etc.) + add custom textExample modes:<ul><li>"Append to Base" - adds to Phase 1 prompt</li><li>"Use Base" - copies Phase 1 exactly</li><li>"Unique Prompt" - completely different</li></ul>Step 4: Pick Your ModelsSmart Sequential WAN Loader: • workflow_mode: I2V or T2V • i2v_high_model: Your I2V high model • i2v_low_model: Your I2V low model • t2v_high_model: Your T2V high model (for T2V mode) • t2v_low_model: Your T2V low model (for T2V mode)Step 5: Configure SettingsAdvanced WAN Settings Controller: • base_resolution: (choose from dropdown - default is 896×1152) • upscale_factor: 0.05-4.0 (default is 1.35) • length: (default is 80 frames) • generation_fps: (default is 16) • final_output_fps: (default is 32) • batch_size: (default is 1)Step 6: Load Your LoRAsFor each phase you're using:<ol><li>Open the LoRA Loader node</li><li>Select 3-4 LoRAs</li><li>Open TriggerWord Toggle nodes</li><li>Click to enable/disable triggers</li></ol>Step 7: GenerateQueue and run.What's Different from V1.1Added:<ul><li>I2V/T2V mode switching dropdown</li><li>Per-phase LoRA loaders (separate for each phase)</li><li>Per-phase prompt control with inheritance modes</li><li>Visual trigger word toggles</li><li>Fixed trigger word accumulation bug</li><li>Independent negative prompt control per phase</li><li>num_active_phases control</li></ul>Removed:<ul><li>Manual UNET loaders</li></ul>Same:<ul><li>4-phase architecture</li><li>Video generation quality</li><li>VRAM requirements</li></ul>Download FilesIncluded:<ul><li>WAN_4-Phase_V1_4.json</li><li>smart_sequential_wan_<a target="_blank" rel="ugc" href="http://loader.py">loader.py</a></li><li>multi_phase_prompt_manager_4phase_<a target="_blank" rel="ugc" href="http://v2.py">v2.py</a></li><li>advanced_wan_settings_<a target="_blank" rel="ugc" href="http://controller.py">controller.py</a></li></ul>Installation:<ol><li>Put .py files in ComfyUI/custom_nodes/</li><li>Load .json workflow</li><li>Restart ComfyUI</li><li>Configure and generate</li></ol>Version: 1.4 Author: DarthRidonkulous Type: ComfyUI WAN 2.1/2.2 14B Multi-Phase License: Free to use and modify

Files with SHA256: 5dc6b51cb5a57578a941d5f08258934dfc3e3d6626a2c89ddaf7f7e0f9ae1463