UPDATED to v100!
Digital Media Workflow:
Anime Workflow:
🎨 The Gege Workflow Color Key
🟢 Green | The Engine: This is your core hardware. These are the Samplers and Schedulers—the fundamental nodes that drive the standard Stable Diffusion process. If the Engine is off, nothing moves.
🟣 Purple | The Anchors: These are the nodes of Order. Their job is to maintain structural integrity, lock in detail, and ensure your outputs stays faithful to your vision. They don't change the "what," they just make the "what" look perfect. (🔴 Red | Banded Tracker is just an advanced "anchor".)
🟡 Yellow | The Igniters: This is where the Mad Science happens. These nodes are the "Chemical Reagents" that completely reshape your output. They provide the power of a brand-new model while keeping the DNA of your current one, kicking the latent manifold into entirely new aesthetic territories.
Gege Nodes: The 2026 Standard.
Welcome to the bleeding edge of ComfyUI control. This isn't just a collection of upscalers or basic samplers; this is a fully integrated, feedback-aware pipeline designed for pristine anatomy, immaculate detail, and flawless color absorption. Built and optimized to squeeze maximum professional fidelity out of standard consumer hardware (like an 8GB GPU), the Gege Suite gives you laboratory-level control over the latent space, yet the detail is effortless.
Gege Nodes achieves this by allowing the user to abuse high denoise parameters alongside high steps on the second pass to squeeze the maximum amount of detail that your GPU can produce. You literally see it while the image is generating, "why can't it just be good?", well it can be... you just have to add a little bit of elbow grease.
With the new updates, we can achieve even more with our 1st pass generation.
Ergodic Explorer, Its job is to stop the AI from taking the "easy path." In AI sampling, the model usually settles on a predictable composition very quickly. This node introduces Targeted Entropy early in the process to shake the latent vectors and force them to explore "adjacent manifolds"—basically, it's hunting for more interesting versions of your image that the AI would normally ignore.
Pendulum and Scalpel introduce a new way to manipulate your initial generation beyond prompting and puts conditioning into the hands of the operator.
If you're an anime creator, these will be the most important nodes in your life (or at least until the next big anime architecture drops).
None of the images previewed have post-processing upscaling.
All previewed images are the latents produced from the 2nd sampler and a face Adetailer.
Previews also are made with WAI-ePred, and WAI-v160 and Noob Stabilizer LoRA.
⚡ Gege Agentic Euler (The Engine)
The Gege Agentic Euler: Manifesting the Inevitable ⚡
While standard samplers blindly follow a linear path, the Agentic Euler utilizes Dual-Trajectory Latent Forcing to simulate two parallel futures for every single step. It forces the latent seeds to sprout into a solid anatomical trunk before they are allowed to grow any pixel-leaves, effectively hallucinating the final structure of the image in the earliest stages of the run.
Designed specifically to defeat the memory bottlenecks of your consumer GPU, it uses a batch trajectory engine that fuses dual passes into a single GPU call. Wrapped in FP16 Autocast, it survives high-resolution runs on 8GB of VRAM while leveraging JIT-compiled PyTorch math to perform heavy-duty SLERP calculations at the speed of light. It takes the "guesswork" out of the first ten steps, so the rest of the generation can focus purely on the aesthetic finish.
New Core Features
Variance-Preserving Macro-Blend: The ghosting issues of previous versions have been eradicated. The sampler now uses a Standard Deviation Sync during blending. If the dual trajectories begin to diverge, the sampler mathematically clamps their variance to the original pixel distribution, ensuring your edges stay sharp and your characters stay singular.
Resolution Stability Factor: The normal VRAM consumer GPUs can be temperamental with high resolutions. The Agentic Euler now dynamically checks your latent pixel area and adjusts its shift_intensity automatically. If you're pushing a 2K render, it lowers the "heat" to prevent the manifold from fracturing.
Temporal Decoupling Mode: You now have the option to "uncouple" time. When active, the sampler draws the anatomical structure using the future’s noise levels while maintaining the current noise level for the colors. It allows for anatomical locking without losing the soft gradients of early-stage diffusion.
Gege Guidance Integration: The sampler is now "Heat-Aware." If paired with the Phasic Entropy suite, it checks for spikes in the latent entropy map and triggers a Heun/DPM2 Refinement pass only where needed, saving precious VRAM by skipping unnecessary math on empty backgrounds.
🧬 Gege Fusion (The Power)
The central "Neural Conductor" of the Gege Suite. This node doesn't just blend noise maps like DDIM and SGM via a Sigmoid Router—it now features a Spatial Brain that observes the UNet's attention blocks in real-time. By calculating a 2D Volatility Heatmap, it identifies exactly where the model is struggling with complex anatomy and broadcasts those "thoughts" telepathically to the rest of the suite. It uses Entropy Injection for organic "Beautiful Data" textures and a Dynamic Coherence Scope to anchor the structure before the handoff.
Spatial Brain Toggle: This feature activates a surgical variance tracker that observes the UNet’s attention blocks in real-time. It identifies where the model "struggles" (hands, feet, complex lace) versus where it is confident (sky, flat skin).
Telepathic Link: Instead of using messy UI wires, the Brain writes its spatial heatmap directly into the model's subconscious memory (model_options), allowing other nodes to "read its thoughts" instantly.
"We’re just stitching two different dreams together... eventually we'll see which one is more real. 🧪"
🕵️Gege ASA Agent (The Legacy Specialist)
The original Adaptive Scheduling Agent. It uses "Hard-Switching" based on your workload intensity. Use with 1st pass.
The Portfolio: It maps Latency to Simple, Balance to Karras, and Throughput to Exponential.
The Strategy: It’s a set-and-forget tool for batch runs where speed and consistency are the priority over granular artistic blending.
🕵️♂️Gege ASA Agent V2 (Complete Edition)
The Continuous Intelligence engine. This is the crown jewel of the suite, designed to "grab the horns" of Illustrious-based realism models. Use with 1st pass.
The "Cheese Dip" Logic: Instead of jumping between schedulers, it uses an Intelligence Slope to create a smooth, sliding transition between three "Experts" (Latency ⇄ Balanced ⇄ Precision).
The Flavor Profiles:
Pristine Anime: Optimized for clean cel-shading.
Flat Cel-Shade: Strips out Karras/Exponential to prevent over-rendering.
Beta Meta: The "Realism Hack." Uses SGM Uniform to pull out micro-details that standard anime schedulers usually ignore.
Classic & Deep Illustrative: The heavy hitters for digital media, focusing on maximum texture lock-in during the final steps.
📆 Gege Hybrid Scheduler (The Icing on the Cake)
A structural node that physically blends two entirely different noise curves (like the aggressive drop of Karras and the smooth slope of Normal) using a custom Blend Ratio. It creates a custom hybrid schedule, allowing you to tailor the exact pacing of your diffusion process for specialized art styles. Use this for the 2nd pass.
"Why use only one scheduler when it's 2026? We can use 2! 📆"
🥽 Gege Observer (The Shock Absorber)
A physics-based utility node that applies classical momentum and drag formulas to smooth out the mathematical "bumps" in a noise schedule. By observing the velocity of the denoising steps, it prevents jittery artifacts and keeps the diffusion trajectory completely stable during complex, high-CFG runs. Use with 1st pass.
"Things are about to start flying all over the place. Let's put on some three-dimensional goggles first. 🥽 "
🏗️ Gege Tile (The Tiler)
The Tile is the master architect of the Gege Suite, specifically engineered to break the "Resolution Ceiling" during your Second Pass. When you're pushing a generation into high-definition territories, standard AI models often get confused, leading to "hallucinations" like double heads, distorted limbs, or repetitive patterns. The Tile solves this by dividing the image into a smart, algebraic grid that handles the heavy lifting in sections without losing the "Big Picture." Use with 2nd pass.
Topological Nudge (Smart Gridding): Unlike basic tilers that use a rigid, "cookie-cutter" grid, the Tile features Spatial Gravity. If it detects a complex subject (like a character), the grid lines can actually warp and bend toward that density. This ensures the most important parts of your image get the most "mathematical attention," keeping your main subject crisp while the background stays stable.
Dilated Context Expansion (The Seam-Killer): The biggest fear with tiling is seeing "seams" where the squares meet. The Tile fixes this by giving each section "Peripheral Vision." Every tile sees a feathered overlap of its neighbors. This communication ensures that a long object, like a sword or a strand of hair spanning multiple tiles, remains one continuous, perfect line.
Attention-Weighted CFG: The Tile is "Context-Aware." It automatically boosts the prompt strength (CFG) for the high-detail areas of your grid. This makes your main subject "pop" with clarity and sharpness, while preventing the simpler background tiles from becoming "deep-fried" or over-processed.
Global Crystallization: As your Second Pass nears the finish line, the Tile slowly dissolves its own grid boundaries. This allows the model to perform a final "Global Pass," crystallizing the separate sections into one perfectly seamless, high-fidelity masterpiece.
"We aren't just breaking the image into pieces; we're teaching the pieces how to talk to each other. It’s not just a upscale—it’s a structural evolution. 🏗️📐"
🌈 Gege Enforcer (The Color)
The Enforcer is the structural bedrock of the Gege Suite. Its mission is to ensure that your high-resolution generations remain faithful to your original vision. In standard upscaling, the AI often "drifts," changing poses or losing the essence of the first pass. The latent forcer prevents this by forcefully tethering the current generation to a Source Latent using advanced spherical math, ensuring your composition stays locked while the details flourish.
Spherical Latent Forcing (SLERP): Unlike basic blending that can cause "ghosting" or transparency, the Anchor uses SLERP (Spherical Linear Interpolation). This moves the pixels along the mathematical curve of the latent manifold rather than cutting across it, keeping your edges sharp and your character's soul intact.
The Golden Phi Mask: Using the Golden Ratio (ϕ), this node performs a multi-scale analysis to identify the most important parts of your image (like eyes and focal points). It creates a "Precise Shield" around these areas, protecting them from distortion while allowing the background to evolve naturally.
Color Manifold Matching: The Anchor features a complete "Laboratory" of color tools:
Anime Saturated (RGB): Forces the colors to stay vibrant and prevents the "Latent Gray Death."
CMYK Subtractive: Mimics the look of physical ink on paper for a rich, matte aesthetic.
Legacy Reverse AdaIN: A fan-favorite "exploit" that weaponizes noise to create a bright, blooming pastel style.
Momentum Protection: When paired with a Causal History, the Anchor becomes "self-aware." If it senses the image is changing too fast (which usually causes artifacts), it "steps on the brakes" to stabilize the anatomy, ensuring a smooth and intentional render.
"It’s not an overlay; it’s a permanent ink-soak into the soul of the image. 🎨"
🛡️ Gege Manifold Governor (The Protection)
This is the Structural Skeleton that lives inside the math. It acts as a real-time Manifold Stabilizer, using spatial edge detection to calculate the latent "tension" of your image. By leveraging a complex Trajectory Engine, it allows you to manipulate the path of denoising through modes like Smooth -> Sharpen or Blended Force, essentially acting as a structural shock absorber for your pixels. Use with 2nd pass.
Instead of the traditional, VRAM-heavy process of a multi-pass SD Upscale, the Governor performs a "Shadow Upscale" within a single sampler. It uses Grid Coherence to pool and interpolate attention queries, ensuring the image doesn't drift into chaos even at massive resolutions. It provides the sharpening and structural reinforcement of a second pass, but confines it to the existing GPU memory footprint—allowing your home GPU to render professional-scale detail without an impractical amount of oversized data that would take you to long to process, or you wouldn't process otherwise, due to VRAM limitations.
ADetailer Auto-Bypass: Features a built-in sequence length shield that instantly detects cropped latents, preventing the Governor from turning ADetailer face-fixes into over-sharpened artifacts.
Optimization: FP8, BF16, and FP32 modes with torch.compile.
At times, it may feel like this node isn't obeying you. When that occurs, turn off q_mode, turn down anchors and all boosts; test. All of the nodes are essentially performing a dance with each other, sometimes when toes or fingers converge and are unmoved by parameters, it's simply because "the dance is too much".
"I don't just see pixels; I see the weight of the world they’re holding up. 🛡️"
🏺 Gege Pitcher (The Detail)
The Pitcher is the "Aesthetic Refiner" of the Gege Suite. Its logic is based on the fluid dynamics of a water pitcher: it metaphorically "pours water in" (detail) and "takes water out" (noise) through a process called a Van Der Pol limit cycle. By oscillating the frequencies of the latent space, it forces the AI to find a perfect balance between sharp, defined lines and soft, atmospheric colors. This node is a mandatory standard for anyone looking to increase Second Pass detail without making the image look over-sharpened or "crunchy." Use with 2nd pass.
The Fluid Logic (In and Out): Imagine the image is a pitcher of water. This node "stirs" the water, pushing the "ink" (edges and fine details) to the top while keeping the "water" (gradients and colors) smooth at the bottom. It applies a mathematical "friction" that prevents the image from becoming too flat or too noisy, ensuring that the details feel naturally poured into the composition.
The Pastel Sprayer: This is the node's secret weapon for high-end aesthetics. It allows you to "tint the water" before it's poured. You can inject vibrant, soft palettes like Sakura Pink or Mint Breeze directly into the low-frequency layers of the image. Because it only tints the "water" and not the "ink," you get gorgeous, vibrant colors without destroying your sharp lineart or deep shadows.
It's not a filter, it's beautiful data injection. A node included named 'Gege Primer' gives this ability for 1st pass usage.Shadow Depth & Limit Cycles: The Pitcher is a guardian of contrast. It identifies the "shadow mass" of your image and applies extra friction there, preventing the dark areas from "washing out." This ensures that even in a pale, pastel render, your blacks stay deep and your structure stays solid.
Stochastic Jitter (The Blind Pour): By adding a tiny amount of "blind" noise during the process, the node forces the AI to constantly re-evaluate the details. This prevents the model from getting "lazy" during the second pass, resulting in a dramatic increase in micro-textures like skin pores, fabric weave, or stray hairs.
"We aren't just sharpening an image; we're pouring the details into a vessel that’s built to hold them. Stir the water, protect the ink. 🏺🎨"
📡 Gege Banded Attention Tracker (The Tether)
The Band is the "Neural Memory" of the Gege suite. Its primary job is to bridge the gap between your First Pass and your Second Pass. In standard AI workflows, the second pass (upscaling/refinement) often "forgets" the original composition or introduces messy new details that ruin the vibe. The Band fixes this by recording exactly where the model was "thinking" during the first pass and forcefully anchoring the high-resolution pass to that foundation.
Phasic Memory Recording (Mode 1): In its first mode, the node acts like a digital observer. It watches the UNet’s attention layers and creates a "Spatial Brain Map." It identifies the "Bands" of high-importance data—like the specific curve of a face or the intricate fold of a garment—and saves that structural DNA.
Steiner-Lanczos Frequency Substitution (Mode 2 & 3): This is the "Mad Science" core. During the second pass, the node performs a surgical swap. It takes the solid, reliable foundation from your First Pass (the "Truth") and injects it into the Second Pass. It then uses Lanczos Filtering to sharpen only the new details, ensuring your upscale is crisp without becoming a "hallucinated" mess.
The "Ultra-Sharp" Edge: When set to Mode 3, the node uses a Laplacian filter to find every structural edge in the image. It then "locks" those edges in place using Harmonic Strictness, meaning no matter how high you push your denoise or CFG, the character's core features will never shift or distort.
Beautiful Data Injection: Instead of just sharpening the whole image (which can look grainy), the Band uses the recorded memory to decide exactly where to amplify detail. It creates "Beautiful Data" only in the areas that matter, leaving flat backgrounds smooth and keeping the focus on your subject.
"We aren't just upscaling pixels; we're enforcing the memory of the original dream. The first pass provides the soul; the Band provides the skeleton. 📡🧠"
🦋 Gege Lorenz Expression Filter (The Pendulum)
The "Chaos Engine" of the Gege Suite. Based on the mathematics of the Butterfly Effect, this node introduces controlled, non-linear turbulence into your conditioning. Instead of static weights, it projects your prompts into a 3D Lorenz Attractor, causing the "meaning" of your words to oscillate and drift through a chaotic manifold. This creates a "swinging" effect where the model explores unexpected artistic variations—shifting between subtle expressions and high-intensity drama—without ever losing the core structure.
Use with 1st pass.
Chaotic Drift (The Butterfly Effect): Using the σ (Sigma), ρ (Rho), and β (Beta) parameters, the node calculates a mathematical "swing." A tiny change in your prompt text can lead to a massive, beautiful divergence in the final render, mimicking the way a butterfly flapping its wings can cause a storm.
Inversion Polarity: When enabled, the Invert Negative toggle pushes the negative prompt's math in the opposite direction of the positive. This creates a "repelling" force that ensures your negative prompts (like "deformed hands") are mathematically fleeing away from your beautiful positive results.
The Gege Scope: Features a live visualizer that renders a 3D "Rainbow Trace" of the chaos. You can literally watch the Pendulum swing as it maps out the mathematical path of your prompts before the first latent is ever sampled.
"Chaos isn't a mess; it's just a pattern we haven't decoded yet. Let the Pendulum swing until the beauty emerges. 🦋🌈"
✂️ Gege Latent Scalpel (The Goal)
If The Pendulum is the chaotic soul of the Gege suite, The Scalpel is its cold, clinical hands. While its sibling uses the chaos of the butterfly effect to drift through the latent space, the Scalpel performs surgical strikes on the actual mathematical vectors of your tokens. It is designed for "Absolute Control," allowing you to carve away unwanted conceptual bleed and suture together a more cohesive structure. This is color coded yellow as well within the demo workflows. This is because despite the description, this is a huge shake-up to the 1st pass, even more so than pendulum. Use with 1st pass.
Latent Surgery (Contrast & Cohesion): Using a precision-weighted gradient, this node can either "repel" concepts to create high-contrast definition or "stitch" neighboring tokens together for smoother transitions. It identifies the Mean Concept of your prompt and allows you to surgically isolate tokens so they don't "smear" into each other.
The Void Modulator: What happens to the empty space in your prompt? Usually, it's just dead noise. The Scalpel transforms this "Unused Latent Real Estate" into a tool:
Vacuum: Sucks the energy out of the padding to prevent "deep-fried" artifacts. (2nd recommended.)
Phantom Echo: Reflects the core meaning of your prompt into the background for thematic resonance. (Recommended for everything.)
Abyssal Static: Injects raw, textured entropy into the empty vectors. (Experimental.)
Dark Mirror: Inverts the prompt's influence to create a subtractive, moody atmosphere. (Experimental.)
Surgical Repulsion: When Repel Negative is active, the Scalpel uses the average "vibe" of your positive prompt as a mathematical weapon, violently pushing the negative tokens away to ensure that "bad data" never infects your "good data."
"The Pendulum dreams the vision, but the Scalpel carves the reality. One provides the swing; the other provides the edge. ✂️🩸"
🧭 Gege Ergodic Explorer (The Explorer)
The Explorer is the "Gonzo Journalist" of the Gege Suite. While other nodes focus on stability and rules, this node is designed to inject Targeted Entropy into the sampling process. Its mission is to prevent the AI from taking the "easy path." By applying "Heat" to the latent manifold, it forcefully pushes the model to explore adjacent possibilities—meaning it hunts for more interesting compositions, poses, and lighting that the model would normally settle past. Use with 1st pass. (Gently)
Ergodic Temperature (Manifold Heat): Think of this as the "Chaos Dial." At low temperatures, it gives the model a gentle nudge to improve textures. At high temperatures, it delivers a violent kick that can completely reshape the scene, forcing the AI to wander through different "versions" of your prompt before it settles.
Orthogonal Kick (Gram-Schmidt Math): This is the secret "Gege" sauce. Unlike standard noise that can make an image look "deep-fried" or gray, the Orthogonal mode uses advanced math to ensure the noise is at a 90-degree angle to your existing data. It adds massive variety and "vibe" without destroying the core structural integrity or exposure of your image.
Surgical Timeline (Logarithmic Scaling): The Explorer understands that not all steps are equal. In Corrected (Surgical) mode, it uses logarithmic math to ensure that the "chaos window" feels natural. It knows that the first few steps are where the "big ideas" happen, giving you precise control over exactly when the exploration stops and the refinement begins.
Decaying Entropy: As the generation reaches your Exploration Cutoff, the "Heat" doesn't just vanish—it decays. This allows the model to smoothly transition from a state of chaotic discovery back to the "Causal Anchor," ensuring your final pixels are clean and intentional.
"We aren't just sampling; we're wandering the map. If the AI wants to take the shortcut, the Explorer throws it into the wilderness to find something better. 🧭🔥"
📡 Gege Overseer (The Sensor)
The "Structural Pulse" of the Gege Suite. Moving away from standard policing, the Overseer acts as a live sonar for the latent manifold. It doesn't just wait for a mistake; it observes the UNet’s attention blocks in real-time to map out the "Skeletal Energy" of your generation. It identifies the exact moment the model’s "thoughts" become fractured or chaotic—like when it tries to manifest 12 fingers in a space meant for 5—and uses that data to broadcast a structural correction. Use with 1st pass.
Magnitude Sensing: The Overseer monitors the "subconscious" energy levels of the model. By tracking these Topological Peaks, it can tell the difference between a high-detail face and a "fractured" anatomical error before a single pixel is ever drawn.
Targeted Entropy Kick: When the Overseer detects a structural fracture (like a tangled mess of limbs), it doesn't just blur the area. It injects a precise "Kick" of noise back into that specific spot. This forces the AI to "re-dream" that part of the body until it obeys the laws of physics.
Patrol Windows: You decide when the Overseer is on duty. Typically, it patrols the critical middle steps (0.1 to 0.5) when the core anatomy is being "calcified." Once the structure is locked and safe, the Overseer stands down so the model can focus on the fine textures and colors.
"We aren't just fixing mistakes; we're monitoring the pulse of the latent space. If the geometry fractures, the Overseer mends the bone. 📡⚖️"
🗿 Gege Relational Decoder (The Bailiff)
The "Anatomical Enforcer" of the Gege Suite. Moving beyond the "casual" reliance on external ControlNets, this node performs Topological Peak Analysis directly on the latent manifold. It utilizes a Laplacian of Gaussian (LoG) filter to identify terminal spikes (fingertips/toes) and maps them to their corresponding Semantic Anchors (palms/soles). By enforcing Mereological Logic, it ensures that "morphisms" (branches) never exceed the categorical limit, effectively upholding the Law of Five before a single pixel is ever rendered. Use with 2nd pass.
Dynamic Mean-Shift: A floating thresholding math that adapts to the current noise floor. It ignores the chaos of early steps but becomes surgically precise as the manifold cools, catching "hallucinated" digits the moment they attempt to manifest.
Relational Discrepancy Map: Instead of a simple mask, it generates a high-level topological blueprint. This "Veto Map" identifies orphaned peaks and illegal clusters, broadcasting a spatial "Order of Execution" to the Causal Anchor.
"Geometry is just a suggestion until the Bailiff arrives. We aren't just cleaning pixels; we're enforcing reality. ⚖️"
⚓ Gege Causal Anchor (The Judge)
The "Anti-Swampman" mechanism designed to solve the memoryless failure of standard diffusion. The Anchor functions as a Path-Dependent Governor, tracking the Historical Momentum (ΔH) of every latent step. It listens to the Telepathic Link from the Phasic Fusion and the Relational Map from the Decoder to act as a real-time Executioner. When the UNet attempts a "Swampman" hallucination (an anatomical trajectory that violates its own history), the Anchor intercepts and forces a Causal Veto, overwriting the error with a mathematically sound historical projection. Use with 2nd pass.
Causal Handshake: A sophisticated data handover that preserves the "Rock" (History) for the second pass. It ensures that high-frequency detailing never "shatters" the structural edges already established in the primary pass.
Inertial Dampener: An anisotropic SLERP governor that uses the Structure (Paper) map to decide when to let the model "dream" and when to lock the trajectory to the Causal Anchor. It prevents the "frying" of edges by dampening destructive interference in high-gradient zones.
"Memory is the only thing standing between a masterpiece and a monster. We give the UNet a soul, then we tell it exactly what to remember. 🧪"
These are are an alternative to 1st pass latent forcing. 'CAUSAL_HISTORY' output from Causal Anchor must be plugged in to Temporal Latent Forcer, when used in conjuction.
Computationally more logical, digitally faster, and much cleaner (at the price of less infused data)
than 'latent forcing'.
👮 Relational Injector (The Police)
Think of this as the Live Anatomy Police. While the model is busy dreaming up a person, the Relational Injector is watching the math in real-time to make sure the body actually makes sense. It focuses specifically on the "branching" parts of the body—hands, feet, and limbs—where AI usually gets confused. Use this on the 1st pass.
The "Law of Five" (No More Extra Digits): This is the heart of the node. It scans the latent space and literally counts the fingers or toes attempting to grow in one spot. If it counts more than five, it steps in and tells the model: "Stop. That’s too many," forcing the extras to dissolve before they ever become pixels.
Phantom Limb Removal: Ever see a random hand floating in the background? This node uses Anchor Mass Analysis to check if a hand is actually attached to an arm. If it finds a "limb" floating out in space with no connection to the body, it considers it an "orphaned peak" and deletes it.
Seamless Healing: It doesn't just "cut" things out and leave a mess. It uses a Local Healing technique that blends the deleted area into the background perfectly, so you never even knew an extra limb was trying to ruin your masterpiece.
This is on maximum anchor radius for a very stubborn image. Sadly I can't show rest of image on this post.
Though it added a second hand in a very strategic place when set to maximum. Absolutely amazing piece of tech.
"We aren't just fixing hands; we're enforcing the laws of biology. If it doesn't have a reason to exist, the Injector won't let it stay. 🧬⚖️"
How to use:
Open Gege_Nodes zip file, and drag and drop into the 'custom_nodes' folder in your root directory.
'Anime' and 'Digital Media' workflows are included.
Warnings / Notes:
Using this setup together with Perturbed Attention Guidance or FreeU is not recommended, though you're welcome to experiment.
Even though it's optimized, it will still increase generation time if you're running on 8 GB VRAM (I know this from personal experience).
Is it worth it? Absolutely.
No disrespect to the creators of SD Upscale nodes, but — figuratively speaking— that node is like a BlackBerry keypad phone from the mid-2000s, while Manifold Governor is like an iPhone X on smaller GPUs… and the latest iPhone on modern/high-VRAM GPUs.
That said, I get it—as one AI creator to another. We all wrestle with that prideful instinct, whether it's gatekeeping or withholding praise from fellow creators. (It's okay; we all do it to some degree.)
This time, though, you'll want to give me my roses. Take a chance this one time for an opportunity to strike gold forever.
I'm actually getting super close to upgrading my hardware, and is part of reason why I'm uploading this "free" version. If you'd like to have access to the upcoming update that allows DPM3 usage with Illustrious and has faster and higher quality banded tracker, you can support me on ko-fi. I'd highly appreciate it and it would really mean the world to me.