<h1 id="new-illuminate-ai-ltx-2.3-autonomous-vision-director-v1.2">NEW — ILLUMINATE AI LTX 2.3 AUTONOMOUS VISION DIRECTOR v1.2</h1><h2 id="complete-civitai-package-workflow-custom-nodes-and-installation">COMPLETE CIVITAI PACKAGE — WORKFLOW, CUSTOM NODES AND INSTALLATION</h2>ILLUMINATE AI v1.2 is a major quality and prompting update for the LTX 2.3 Autonomous Vision Director.The workflow analyzes the supplied reference image, understands the visible starting situation, creates a physically achievable chronological scene, writes the complete LTX 2.3 production prompt and validates it against the original image before video generation begins.The final validated prompt is displayed directly inside the workflow before it is sent to LTXDirector.<hr /><h2 id="new-in-v1.2">NEW IN v1.2</h2><h3 id="improved-image-quality">Improved Image Quality</h3>The complete render and refinement path has been reworked for noticeably cleaner output.Improvements include:<ul><li>cleaner hair structure during movement</li><li>finer skin, fabric and facial detail</li><li>reduced coarse and blocky motion textures</li><li>improved moving-edge quality</li><li>stronger Stage 2 detail reconstruction</li><li>native FP16 output without the previous downscale and bicubic enlargement chain</li></ul>Current quality configuration:<ul><li>Stage 1: 8 steps</li><li>Stage 1 CFG: 1.25</li><li>Spatial latent upscale: enabled</li><li>Stage 2: 4 steps</li><li>Stage 2 sigma endpoint: 0.40</li><li>Native FP16 output: 1.00</li><li>no third sampler required</li></ul><h3 id="more-accurate-autonomous-prompting">More Accurate Autonomous Prompting</h3>The autonomous Prompt Director has been expanded and refined.It now handles:<ul><li>duration-aware scene planning</li><li>more accurate participant and anatomy tracking</li><li>improved position and action recognition</li><li>better hand, limb and contact ownership</li><li>persistent anatomy consistency during occlusion</li><li>improved multi-participant continuity</li><li>primary and secondary action validation</li><li>more stable movement planning</li><li>stronger protection against morphing, anatomy loss and role changes</li></ul>The system remains image-guided and preserves the original scene whenever a complex transition cannot be completed safely.<hr /><h2 id="four-prompt-modes">FOUR PROMPT MODES</h2><h3 id="manual-prompt">Manual Prompt</h3>Uses your own finished prompt without an Ollama request.<h3 id="standard-vision">Standard Vision</h3>Qwen analyzes the reference image and creates a compatible cinematic or character-animation prompt.<h3 id="adult-assisted">Adult Assisted</h3>Uses the reference image together with a short written idea and expands it into a complete chronological scene.<h3 id="adult-full-auto">Adult Full Auto</h3>Creates the complete scene automatically from the reference image, selected duration and visible starting situation.Adult modes require confirmed adult and consenting participants.<hr /><h2 id="included-versions">INCLUDED VERSIONS</h2><h3 id="standard-recommended">STANDARD — RECOMMENDED</h3>The main release version.Use this version first for the best balance of:<ul><li>image quality</li><li>motion quality</li><li>anatomy consistency</li><li>render time</li><li>workflow stability</li></ul><h3 id="antighost-experimental">ANTIGHOST — EXPERIMENTAL</h3>An additional experimental version with optional temporal Anti-Ghost processing.This version is included for testing and comparison. It may improve ghosting or duplicated motion in some scenes, but it will not necessarily improve every video.The Anti-Ghost processing is disabled by default and can be activated for individual tests.<hr /><h1 id="important-new-custom-nodes-are-required">IMPORTANT — NEW CUSTOM NODES ARE REQUIRED</h1>The updated <code>ComfyUI-ILLUMINATE-AI-LTX</code> node pack included with v1.2 is mandatory.The newest workflow will not work correctly with an older ILLUMINATE node pack.The installation method has not changed.<h2 id="installation">Installation</h2><ol><li>Close ComfyUI completely.</li><li>Delete the existing folder:</li></ol><pre><code></code></pre><pre><code>ComfyUI\custom_nodes\ComfyUI-ILLUMINATE-AI-LTX</code></pre>Also remove the old standalone cleanup package if it is still installed:<pre><code></code></pre><pre><code>ComfyUI\custom_nodes\ComfyUI-Illuminate-Ollama-Cleanup</code></pre><ol><li> Copy the included folder: </li></ol><pre><code></code></pre><pre><code>02_CUSTOM_NODES\ComfyUI-ILLUMINATE-AI-LTX</code></pre>into:<pre><code></code></pre><pre><code>ComfyUI\custom_nodes\</code></pre>The final path must be:<pre><code></code></pre><pre><code>ComfyUI\custom_nodes\ComfyUI-ILLUMINATE-AI-LTX\__init__.py</code></pre>Do not create an additional nested folder.<ol><li> Load the included workflow and use: </li></ol><pre><code></code></pre><pre><code>ComfyUI Manager → Install Missing Custom Nodes</code></pre><ol><li> Restart ComfyUI completely. </li><li> Load the new v1.2 workflow JSON. </li></ol>A browser refresh with <code>Ctrl + F5</code> is recommended after updating the Custom Nodes.<hr /><h2 id="ollama-and-qwen">OLLAMA AND QWEN</h2>The autonomous Vision modes use:<pre><code></code></pre><pre><code>fredrezones55/Qwen3.5-Uncensored-HauhauCS-Aggressive:9b</code></pre>Install Ollama and run:<pre><code></code></pre><pre><code>ollama pull fredrezones55/Qwen3.5-Uncensored-HauhauCS-Aggressive:9b</code></pre>Ollama must remain running whenever Standard Vision, Adult Assisted or Adult Full Auto is used.Manual Prompt mode does not require Ollama.<hr /><h2 id="required-bat-flags">REQUIRED BAT FLAGS</h2>The included system check expects:<pre><code></code></pre><pre><code>--fp8_e4m3fn-text-enc --fast-disk</code></pre>Example:<pre><code></code></pre><pre><code>.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build --fp8_e4m3fn-text-enc --fast-disk</code></pre>A ready-made BAT example is included in the package.<hr /><h2 id="first-test">FIRST TEST</h2>For the first successful generation:<ul><li> load the same starting image into the LTXDirector and the Vision Reference Image node </li><li> use the STANDARD version first </li><li> leave optional LoRAs disabled </li><li> keep Anti-Ghost disabled </li><li> generate at 24 FPS </li><li> enable RIFE only after the normal 24 FPS output works </li><li> review the final generated prompt before LTX generation begins </li></ul><hr /><h2 id="package-content">PACKAGE CONTENT</h2>The download includes:<ul><li> v1.2 STANDARD workflow </li><li> v1.2 experimental ANTIGHOST workflow </li><li> updated mandatory ILLUMINATE Custom Nodes </li><li> Windows Portable BAT examples </li><li> Ollama/Qwen installation BAT </li><li> English installation guide </li><li> German quick-start guide </li><li> model and node checklist </li><li> troubleshooting information </li><li> release notes </li></ul>Models are not included in the ZIP.Please read <code>00_READ_ME_FIRST.txt</code> and the included installation guide before the first run.

ILLUMINATE AI – LTX 2.3 I2V AUTO-PROMPT Workflow

Version ChangesNew Prompt EditionThis version introduces a complete Manual / Auto Prompt system for the 12GB Low-VRAM workflow.Auto Prompt Enhancer<ul><li>Added optional Ollama-based prompt enhancement</li><li>Uses <code>qwen2.5:3b</code> as the recommended Low-VRAM prompt model</li><li>Expands short user ideas into one coherent LTX 2.3 image-to-video prompt</li><li>Preserves the intended subject, motion, camera behavior, environment and continuity</li><li>Designed to avoid automatic scene splitting, timestamps, multiple shots or unrelated additions</li></ul>Manual / Auto Prompt Switch<ul><li>Added a dedicated lazy prompt switch</li><li>Manual mode remains available for full direct control</li><li>Auto mode sends the enhanced prompt directly to the external <code>global_prompt</code> input</li><li>Ollama is not executed while Manual mode is selected</li></ul>Prompt Preview System<ul><li>Added Auto-Enhanced Prompt Preview</li><li>Added Final Prompt Sent to LTXDirector Preview</li><li>Makes it easy to verify the final prompt before generation</li></ul>Ollama Integration<ul><li>Added integrated setup instructions</li><li>Added Ollama download link and installation guide</li><li>Added model installation commands</li><li>Added troubleshooting for connection errors, model loading and stalled prompt generation</li><li>Recommended <code>Keep Alive: 1 minute</code> for improved reliability and better Low-VRAM behavior</li></ul>LTXDirector Improvements<ul><li>External prompt routing is now fully integrated</li><li>Stored local segment prompts are intentionally cleared</li><li>Prevents duplicated instructions and unintended prompt segmentation</li><li>Image segments must still be activated manually inside LTXDirector</li></ul>Final Output Improvements<ul><li>Reworked the final video output system</li><li>Only one final video file is saved</li><li>Added a dedicated final output mode switch:<ul><li>OFF = original 24 FPS output</li><li>ON = interpolated 48 FPS RIFE output</li></ul></li><li>Added lazy image and frame-rate switching</li><li>Disabled unnecessary metadata sidecar saving</li></ul>Preview Comparison<ul><li>Added Original 24 FPS Preview</li><li>Added Interpolated 48 FPS Preview</li><li>Both previews are visible inside ComfyUI</li><li>Preview nodes do not save additional files</li><li>Makes it easier to compare interpolation quality before choosing the final output</li></ul>RIFE Interpolation<ul><li>Optional 24 → 48 FPS interpolation remains included</li><li>RIFE output is now integrated directly into the final output selector</li><li>Audio duration remains synchronized</li><li>Interpolation remains optional and does not affect the normal 24 FPS workflow</li></ul>Workflow Cleanup<ul><li>Removed the redundant legacy 48 FPS Video Combine</li><li>Removed duplicate save behavior</li><li>Cleaned unused output paths and stale links</li><li>Improved layout and grouping</li><li>Updated workflow notes and quick-start instructions</li><li>Low-VRAM workflow remains optimized for 12GB GPUs and broader accessibility</li></ul>

Files with SHA256: 1f19dacde72e90129b7dfde8dca662f834ce9356c064145a08de4171dc19e312