<h1 id="i-made-a-single-comfyui-node-that-does-everything-t2i-i2i-inpaint-outpaint-controlnet-face-redraw-faceswap-bg-removal-and-ai-upscale.-no-wiring.-ever.">I Made a Single ComfyUI Node That Does Everything — T2I, I2I, Inpaint, Outpaint, ControlNet, Face Redraw, Faceswap, BG Removal, AND AI Upscale. No Wiring. Ever.</h1>Free · Open Source · GitHub Supports Z-Image Turbo (flow-matching) and Flux.2 Klein 9B<hr /><h2 id="the-problem-with-comfyui-workflows">The Problem With ComfyUI Workflows</h2>You download a workflow. It works — until you want to inpaint the result. So you load another workflow. Then you want to upscale that. Another workflow. Then redo the background. Another one.You end up with 8 browser tabs, 6 workflow files, and you're manually copy-pasting filenames between them like it's 2019.This node fixes that.<hr /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/80f333d6-a472-41cb-863f-3ec857b5c53e/original=true/80f333d6-a472-41cb-863f-3ec857b5c53e.jpeg" /><h2 id="what-is-this">What Is This</h2>TJ NODE STUDIO ONE is a ComfyUI UI extension built around one idea: you shouldn't need to be a node graph engineer just to generate and iterate on images.One node on the canvas. Every mode lives inside its panel. No wiring, no routing, no "wait which output goes into which input again."<pre><code>T2I → I2I → INPAINT → OUTPAINT → RE-BG → CONTROLNET → FACE REDRAW → UPSCALE</code></pre>The entire pipeline — generation, editing, background removal, upscaling — in a single self-contained panel. Hit Send to on any result to pass it to the next mode instantly. That's it.This node is about workflow convenience, not magic. It handles all the node wiring and parameter routing so you don't have to. The image quality you get depends entirely on the model you choose and how you use it — good prompts, well-tuned LoRAs, and appropriate settings will always produce better results. The node just gets out of your way so you can focus on that.<hr /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3cc12bf3-1eb5-4b59-901f-64496702ef57/original=true/3cc12bf3-1eb5-4b59-901f-64496702ef57.jpeg" />Two Nodes, Two Powerhouse Models<h3 id="z-image-one-studio-(tj)">Z-Image ONE STUDIO (TJ)</h3>Z-Image Turbo — Alibaba's flow-matching model (fast, high quality)ModeWhat it doesT2IText → image, LoRA stack (up to 3), resolution presetsI2IDenoise-controlled image transformationINPAINTDraw mask directly in the node, regenerate regionOUTPAINTExpand canvas in any direction (Up/Down/Left/Right px)RE-BGBiRefNet cuts the subject, you write a new backgroundCONTROLNETUnion ControlNet — Depth / Canny / Pose / etc in one modelFACE REDRAWYOLO detects faces → crop → regenerate → seamlessly blend backUPSCALESeedVR2 AI upscale (3B / 7B DiT model)<h3 id="flux.2-klein-one-studio-(tj)">Flux.2 Klein ONE STUDIO (TJ)</h3>FLUX.2 Klein 9B (Black Forest Labs) — the new beastModeWhat it doesT2IFlux.2 Klein generation, LoRA stackI2IImage-to-image with Flux.2EDITMulti-reference editing — feed up to 5 reference imagesPAINTInpaint + Outpaint in one modeFACESWAPFace transfer via BFS LoRA (Best Face Swap)UPSCALESeedVR2 AI upscale (same models, shared)<hr /><h2 id="features-that-actually-matter">Features That Actually Matter</h2>No wiring required — the node builds the ComfyUI graph internally. You set parameters, click Queue, done.Send to — one button passes your result to any other mode as source image. T2I result → Inpaint → Upscale → done, all in 3 clicks.Gallery — every generated image is saved and browsable inside the node. Star favorites, delete failures, send any past result to any mode.Compare slider — drag to compare source vs result side by side. State is preserved when you switch modes.Scroll zoom — mouse wheel to zoom into any result. Works in all modes including upscale and compare view.Model Override Slots — expose input slots to wire in Primitive nodes for model filenames. <code>.gguf</code> filenames auto-switch to GGUF loaders.LoRA stack — up to 3 LoRAs per generation. Trigger word auto-detection from file metadata.Prompt Templates — save and recall prompt presets per node.Settings persist — model selection, prompt suffix, save folder, everything saves to localStorage and survives restarts.LLM Prompt Studio integration — pair with <a target="_blank" rel="ugc" href="https://github.com/designloves2/ComfyUI-TJ_NODE">TJ_NODE LLM PROMPT STUDIO</a> for AI-assisted prompt generation.<hr /><h2 id="seedvr2-upscale-built-in">SeedVR2 Upscale — Built In</h2>Both nodes include a full SeedVR2 AI upscale mode.<ul><li>DiT model: 3B (FP16 / FP8) or 7B FP8</li><li>Resolution up to 4096px</li><li>Attention mode: sdpa / flash_attn_2 / flash_attn_3 / sageattn_2 / sageattn_3</li><li>Color correction (LAB space)</li><li>Cache model + offload to CPU/CUDA</li></ul>Upload any image, pick your DiT + VAE, hit Queue. The result lands in the gallery and you can send it straight to inpaint or I2I.<hr /><h2 id="about-image-quality">About Image Quality</h2>The node does not generate images. The model does.This is important to understand before you use it:<ul><li>Results depend on your model choice — Z-Image Turbo and Flux.2 Klein 9B both produce excellent output, but they have different strengths and respond differently to prompts.</li><li>Results depend on your LoRA selection — a well-trained LoRA for your target style will dramatically improve quality. A mismatched LoRA will hurt it.</li><li>Results depend on your prompts and settings — steps, CFG, sampler, denoise strength all matter and need tuning for your use case.</li></ul>This node's job is to make all of that easier to iterate on — switch modes, tweak, compare, send to next step — without rebuilding a workflow every time. It doesn't replace good prompting or model knowledge. It just removes the wiring friction so you can focus on the creative part.If your results look off, the answer is almost always: better prompt, different LoRA, adjust denoise/CFG, or try a different model variant. The node itself is just a UI shell.<hr /><h2 id="install">Install</h2><h3 id="1.-install-the-node">1. Install the node</h3><pre><code>cd ComfyUI/custom_nodes
git clone https://github.com/designloves2/ComfyUI-TJ_NODE_STUDIO_ONE.git</code></pre>Or search <code>ComfyUI-TJ_NODE_STUDIO_ONE</code> in ComfyUI Manager.<h3 id="2.-install-required-custom-nodes-(via-comfyui-manager)">2. Install required custom nodes (via ComfyUI Manager)</h3>NodeRequired for<a target="_blank" rel="ugc" href="https://github.com/kijai/ComfyUI-KJNodes">ComfyUI-KJNodes</a>Flux.2 Klein KV Cache<a target="_blank" rel="ugc" href="https://github.com/kijai/ComfyUI-SeedVR2">ComfyUI-SeedVR2</a>UPSCALE mode (both nodes)<a target="_blank" rel="ugc" href="https://github.com/cubiq/ComfyUI_FaceAnalysis">ComfyUI_FaceAnalysis</a>Faceswap mode (Klein)<a target="_blank" rel="ugc" href="https://github.com/1038lab/ComfyUI-RMBG">ComfyUI-RMBG</a>RE-BG mode (Z-Image)<a target="_blank" rel="ugc" href="https://github.com/Fannovel16/comfyui_controlnet_aux">comfyui_controlnet_aux</a>ControlNet preprocessors<a target="_blank" rel="ugc" href="https://github.com/city96/ComfyUI-GGUF">ComfyUI-GGUF</a>GGUF quantized models (optional)<h3 id="3.-download-models">3. Download models</h3>Z-Image Turbo:<ul><li>Diffusion: <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/z_image_turbo/resolve/main/split_files/diffusion_models/z_image_turbo_bf16.safetensors">z_image_turbo_bf16.safetensors</a> → <code>models/diffusion_models/</code></li><li>Text Encoder: <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/z_image_turbo/resolve/main/split_files/text_encoders/qwen_3_4b.safetensors">qwen_3_4b.safetensors</a> → <code>models/text_encoders/</code></li><li>VAE: <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/z_image_turbo/resolve/main/split_files/vae/ae.safetensors">ae.safetensors</a> → <code>models/vae/</code></li></ul>Flux.2 Klein 9B:<ul><li>Diffusion: <a target="_blank" rel="ugc" href="https://huggingface.co/collections/black-forest-labs/flux2">Black Forest Labs HF Collection</a> → <code>models/diffusion_models/</code></li><li>Text Encoder + VAE: <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/vae-text-encorder-for-flux-klein-9b/tree/main/split_files">Comfy-Org 9B pack</a> → <code>models/text_encoders/</code> + <code>models/vae/</code></li></ul>SeedVR2 (UPSCALE mode):<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/kijai/SeedVR2-models/tree/main">kijai/SeedVR2-models</a> → <code>models/SEEDVR2/</code></li><li>You need: a DiT file (<code>seedvr2_ema_3b_fp16</code> or <code>_fp8</code> or <code>7b_fp8</code>) + <code>ema_vae_fp16.safetensors</code></li></ul><h3 id="4.-load-the-workflow-+-open-settings">4. Load the workflow + open Settings</h3>Open the included <code>.json</code> workflow → double-click the node → click ⚙ Settings → pick your models → Queue.<hr /><h2 id="github">GitHub</h2>TJ NODE STUDIO ONE (this): <a target="_blank" rel="ugc" href="https://github.com/designloves2/ComfyUI-TJ_NODE_STUDIO_ONE">https://github.com/designloves2/ComfyUI-TJ_NODE_STUDIO_ONE</a>TJ NODE — LLM Prompt Studio: <a target="_blank" rel="ugc" href="https://github.com/designloves2/ComfyUI-TJ_NODE">https://github.com/designloves2/ComfyUI-TJ_NODE</a><hr /><h2 id="screenshots-demo">Screenshots / Demo</h2>(attach your screenshots here — T2I result, Inpaint canvas, Gallery view, Compare slider, Upscale before/after)<hr /><h2 id="faq">FAQ</h2>Q: Does this work with other models (SDXL, SD 1.5)? A: No — built specifically for Z-Image Turbo and Flux.2 Klein 9B/4B architectures.Q: Do I need all the custom nodes? A: Only the ones for the modes you use. SeedVR2 for upscale, FaceAnalysis for faceswap, RMBG for RE-BG, etc. Core T2I / I2I / Inpaint works without any of them.Q: VRAM requirements? A: Flux.2 Klein 9B needs ~18–24GB without offloading; use Offload to CPU for lower VRAM. Z-Image Turbo is lighter. SeedVR2 3B FP8 runs on ~12GB.Q: My results don't look great. Is something broken? A: Probably not. Output quality is 100% determined by the model, your prompts, and your LoRA choices — not the node. Try refining your prompt, adjusting CFG/steps, or using a style LoRA suited to your target. The node is just a UI layer; it passes your settings to the model exactly as you set them.Q: Can I wire this into a normal ComfyUI workflow? A: The node internally generates its own graph on Queue. You can connect <code>prompt_override</code> or model override input slots, but the node is self-contained by design.Q: Why is there no separate workflow file for each mode? A: That's literally the whole point.<hr />MIT License. Free forever. PRs welcome.

I Made a Single ComfyUI Node That Does Everything — T2I, I2I, Inpaint, Outpaint, ControlNet, Face Redraw, Faceswap, BG Removal, AND AI Upscale. No Wiring. Ever.

Files with SHA256: 5b7718101b50bcd9782fa93ddd130334d3360128cbddcb67be62f3d9d71aadde