<h1 id="igen-simple-workflow-guide">iGEN Simple — Workflow Guide</h1>78 nodes · 5 groups · 36 unique node types · 2-pass generation pipeline<ul><li>Built around ComfyUI Smart Model Loader</li><li>Uses ComfyUI Eclipse for compact routing, prompt, upscale, and save utilities</li><li>Includes optional Detail Daemon sampling enhancement</li></ul><hr /><h2 id="what-is-this">What Is This?</h2>iGEN Simple is a compact text-to-image workflow for ComfyUI. It keeps the useful core of a larger generation setup—template-based model loading, wildcard prompts, custom sampling, a low-denoise upscale pass, comparison, metadata, and organized saving—without turning the canvas into a maze.The included configuration is tuned for a Krea2-style pipeline. It loads the <code>Krea2_GPT</code> Smart Model Loader template, generates a base image, enlarges it by 1.5×, performs a short refinement pass, and saves the result with generation data.This is intentionally a focused workflow. It does not include iGEN ONE's image loaders, ControlNet paths, face detailers, identity tools, watermarks, or long post-processing chain.<hr /><h2 id="how-it-works">How It Works</h2>The canvas reads from left to right through five groups:<ol><li>Model Loader — loads the model stack and publishes its components and settings.</li><li>Prompt — builds the positive prompt from wildcard choices and supplies a negative prompt.</li><li>Sampler — creates the base image with custom sampling and optional Detail Daemon processing.</li><li>Upscale — resizes the selected image and performs a short, low-denoise refinement pass.</li><li>Save Image — compares the base and refined images, records generation data, and saves the chosen final result.</li></ol>The workflow uses Set/Get routing to publish named values such as <code>main_model</code>, <code>prompt_positive</code>, and <code>init_images</code>. <code>GetAllActive</code> collects whichever compatible values are active, while switches choose the LoRA-adjusted or original model path. This keeps most long-distance connections off the canvas.The historical <code>[Eclipse]</code> suffix on Smart Model Loader node IDs is preserved for workflow compatibility. Those loader, conditioning, pipe, and sampler nodes are supplied by the standalone ComfyUI Smart Model Loader package.<hr /><h2 id="1.-model-loader">1. Model Loader</h2>The Smart Model Loader is the main control center. The bundled workflow opens with the <code>Krea2_GPT</code> template and the following saved setup:<ul><li>Primary model: <code>krea2GPTGrandPUSSYTruth.safetensors</code></li><li>CLIP: <code>qwen3vl_4b_fp8_scaled.safetensors</code></li><li>CLIP type: <code>krea2</code></li><li>VAE: <code>qwen_image_vae.safetensors</code></li><li>Resolution: <code>960 × 1440</code></li><li>Batch size: <code>1</code></li><li>Attention mode: <code>flash-attention2</code></li><li>Model precision: <code>bfloat16</code></li></ul>The loader sends one <code>PIPE</code> into IO Checkpoint Loader, which exposes the model, CLIP, VAE, latent, dimensions, names, and sampler values for the rest of the workflow.<h3 id="verified-model-acquisition">Verified model acquisition</h3>The primary model carries a saved CivitAI AIR reference. If it is missing, select the template and use Download from CivitAI in Smart Model Loader. The download is explicit—it does not begin merely because the workflow or template was loaded—and the resulting file is verified before use.If the file is present but its hash does not match the saved identity, the loader marks it as a mismatch and exposes the same download action so you can replace it with the verified CivitAI file.The external CLIP and VAE must also exist in the model folders shown by Smart Model Loader. If you substitute another template, review its model type, CLIP type, VAE, resolution, and sampler recommendations together.<h3 id="optional-loras">Optional LoRAs</h3>The Lora Stack provides ten slots and is empty by default. Add LoRAs there when needed. The workflow automatically prefers the LoRA-adjusted model and CLIP when the stack is active, then records both loader and stack LoRA names in the saved generation data.Start with the workflow unchanged and confirm one successful image before adding LoRAs or replacing individual model components.<hr /><h2 id="2.-prompt">2. Prompt</h2>The positive prompt is generated by Wildcard Processor. Its included example uses brace choices such as <code>{slavic|asian|european|african}</code> and <code>{straight|messy|wavy}</code>, so each population can produce a different resolved prompt.You can:<ul><li>replace the example with an ordinary prompt;</li><li>use <code>{choice one|choice two|choice three}</code> for inline random choices;</li><li>use installed wildcard files through the processor's wildcard selector;</li><li>change the wildcard seed to reproduce or vary the populated text.</li></ul>The Negative Prompt field defaults to:<code>worst_quality, low_quality, lowres, blurry, motion lines</code>The sampler's Fast Mode Toggle controls whether the negative encoder is active. In the saved workflow, negative prompting is switched off and zero conditioning is used instead. Turn Negative Prompt on in the toggle when the selected model benefits from explicit negative conditioning.<hr /><h2 id="3.-sampler">3. Sampler</h2>Smart Sampler Settings supplies the main sampling controls. The saved configuration uses:<ul><li>Sampler: Euler</li><li>Scheduler: Simple</li><li>Steps: 30</li><li>CFG: 5.0</li><li>Guidance: 3.5</li><li>Denoise: 1.0</li><li>Seed: <code>988134358229864</code> in the saved workflow; change it for a new variation</li></ul>The base pass uses ComfyUI's advanced custom-sampling nodes, then decodes the latent and opens Image Selector. When generating a batch, select the image or images you want to continue into the upscale stage.<h3 id="fast-mode-toggle">Fast Mode Toggle</h3>The toggle panel manages three optional sampler features:<ul><li>Detail Daemon — enabled in the saved workflow; wraps the selected sampler and adjusts its sigma schedule to emphasize detail.</li><li>Flux Guidance — disabled in the saved Krea2 configuration; enable it for a compatible Flux-family setup.</li><li>Negative Prompt — disabled by default; enable it to use the text from the negative prompt field.</li></ul>After changing templates or model families, review all three. Settings that are useful for one architecture may be inappropriate for another.<hr /><h2 id="4.-upscale-and-refinement">4. Upscale and Refinement</h2>The selected base image passes through two operations:<ol><li>Image Upscale With Model v2 enlarges the image by 1.5× with Lanczos resampling. No AI upscale model is selected by default. Smart sharpening is enabled with a restrained amount of <code>2</code>.</li><li>Eclipse KSampler (Pipe) performs a short refinement pass using 3 steps, Euler, the Simple scheduler, and 0.15 denoise.</li></ol>At the default <code>960 × 1440</code> base resolution, the 1.5× resize produces approximately <code>1440 × 2160</code> before the refinement pass.To use a learned upscaler, choose a model in Image Upscale With Model v2. The node runs that model at its native scale and then resizes to the requested multiplier. For limited VRAM, enable tiled decode in the refinement sampler and choose a suitable tile size.Low denoise preserves the original composition. Raise it gradually if you want the second pass to redraw more detail; large changes can alter faces, text, and small objects.<hr /><h2 id="5.-compare-and-save">5. Compare and Save</h2>The Image Comparer receives both versions:<ul><li>A: refined/upscaled image</li><li>B: original base image</li></ul>Use the comparison view to check whether the second pass improved the result. The save route prefers the refined image and falls back to the base image if the upscale result is unavailable.The Folder Path node saves under ComfyUI's output directory using:<code>images/&lt;current date&gt;</code>The default date format is <code>%Y-%m-%d</code>. Generation Data collects the prompt, negative prompt, seed, model and VAE names, LoRAs, dimensions, steps, CFG, sampler, scheduler, and CLIP skip. Save Images uses those values for its metadata and filename placeholders.Review the Folder Path and Save Images nodes before a long batch if you want a different folder structure, filename, image format, embedded workflow policy, or preview-only output.<hr /><h2 id="quick-start">Quick Start</h2><h3 id="first-run">First run</h3><ol><li>Install the three custom-node packages listed below and restart ComfyUI.</li><li>Load <code>iGEN_Simple.json</code>.</li><li>In Smart Model Loader, leave the template action on Load and select <code>Krea2_GPT</code>.</li><li>Resolve any model marked <code>(missing)</code>. Use Download from CivitAI for the referenced primary model when offered, and install the external CLIP and VAE in their indicated model folders.</li><li>Enter or populate a positive prompt.</li><li>Leave the default sampler and upscale settings unchanged for the first test.</li><li>Queue the workflow, select an image if Image Selector pauses for a choice, then inspect the comparison and saved output.</li></ol><h3 id="use-a-different-model-template">Use a different model template</h3><ol><li>Select another saved or bundled Smart Model Loader template.</li><li>Confirm every required model component is available.</li><li>Review the template's width, height, sampler, scheduler, steps, CFG, and guidance.</li><li>Adjust the Fast Mode Toggle for the new architecture—especially Flux Guidance and Negative Prompt.</li><li>Test the base pass before increasing the upscale multiplier or refinement denoise.</li></ol><h3 id="generate-without-the-refinement-pass">Generate without the refinement pass</h3>Mute or bypass the nodes in the Upscale group. The included fallback allows Save Images to accept the base <code>init_images</code> value when no refined image is published.<hr /><h2 id="troubleshooting">Troubleshooting</h2><h3 id="smart-model-loader-reports-a-missing-model">Smart Model Loader reports a missing model</h3>Reload the saved template so its filename and identity are restored. If Download from CivitAI appears, press it to enqueue the exact referenced primary file. External CLIP, VAE, LoRA, and upscale-model files may need to be installed separately in the folder shown by the relevant widget.<h3 id="integrity-verification-reports-a-mismatch">Integrity verification reports a mismatch</h3>The installed file is not the file identified by the saved hash or CivitAI reference. Use the displayed CivitAI download action to acquire the verified file again, or deliberately select and save a template for the local variant you intend to use. Do not rename an unrelated model to match the expected filename.<h3 id="a-node-is-missing-when-the-workflow-loads">A node is missing when the workflow loads</h3>Install or update the package named in Custom Node Packages Used below, then restart ComfyUI and reload the workflow. The <code>[Eclipse]</code> suffix alone does not identify package ownership: the seven loader-to-sampler pipeline nodes come from Smart Model Loader, while the routing, prompt, UI, upscale, comparison, and save utilities come from ComfyUI Eclipse.<h3 id="the-negative-prompt-or-flux-guidance-has-no-effect">The negative prompt or Flux guidance has no effect</h3>Open Fast Mode Toggle in the Sampler group and enable the corresponding feature. Both are disabled in the saved workflow.<h3 id="the-upscale-pass-runs-out-of-vram">The upscale pass runs out of VRAM</h3>Enable tiled decode in Eclipse KSampler (Pipe), lower its tile size, reduce the base resolution, lower the upscale multiplier, or leave the upscale model set to <code>None</code> and use ordinary resampling.<h3 id="the-workflow-pauses-after-the-base-render">The workflow pauses after the base render</h3>Image Selector is waiting for a selection. Choose the image or images that should continue to the upscale stage.<h3 id="output-is-not-where-expected">Output is not where expected</h3>Check Folder Path first. By default it creates <code>images/&lt;current date&gt;</code> inside ComfyUI's configured output directory. Then check the custom output and filename features in Save Images.<hr /><h2 id="custom-node-packages-used">Custom Node Packages Used</h2><ul><li><a target="_blank" rel="ugc" href="https://github.com/r-vage/ComfyUI_SmartModelLoader">ComfyUI Smart Model Loader</a> — Smart Model Loader, IO Checkpoint Loader, basic and Advanced CLIP encoding, Conditioning Zero Out, and the pipe KSampler.</li><li><a target="_blank" rel="ugc" href="https://github.com/r-vage/ComfyUI_Eclipse">ComfyUI Eclipse</a> — Folder and save tools, Set/Get routing, wildcard prompts, Smart Sampler Settings, LoRA stack, mode toggles, Image Selector, image upscale, and Image Comparer.</li><li><a target="_blank" rel="ugc" href="https://github.com/Jonseed/ComfyUI-Detail-Daemon">ComfyUI Detail Daemon</a> — Optional sigma-schedule detail enhancement around the base sampler.</li></ul>ComfyUI's built-in nodes provide the empty latent, advanced sampler, guider, scheduler, noise, VAE decode, and Flux guidance primitives.<hr />Load a model, write a prompt, and let the compact two-pass path do the rest. 🌒

iGEN - Simple (Save with Generation Data)

<ul><li>update to Eclipse 4.4.26</li><li>update smart model loader 1.0.28</li><li>added qwen image 2.1 support and multi prompts by adding wildcard list</li></ul>

Files with SHA256: bd6b47e26f50b38d7d2664981bb56b524748618330e1cbbfb1651d2884d1bfac