<h1 id="single-comfyui-node-that-does-everything-t2i-i2i-inpaint-outpaint-faceswap-camera-angle-bg-removal-and-ai-upscale.-no-wiring.-ever.-v1.4.1"># Single ComfyUI Node That Does Everything — T2I, I2I, Inpaint, Outpaint, Faceswap, Camera Angle, BG Removal, AND AI Upscale. No Wiring. Ever. - v1.4.1</h1>&gt; Free · Open Source · GitHub&gt; Four all-in-one nodes: Z-Image Turbo · Flux.2 Klein 9B · Qwen Image Edit 2511 · Krea 2&gt; Korean / English UI — switch language in Settings---## The Problem With ComfyUI WorkflowsYou 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 → EDIT → FACESWAP → ANGLE → 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.<h2 id="four-nodes-four-powerhouse-models">Four Nodes, Four Powerhouse Models</h2><hr /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/96b3658d-6a02-4ee2-afe0-9411f83e43e9/original=true/96b3658d-6a02-4ee2-afe0-9411f83e43e9.jpeg" /><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 region (DifferentialDiffusion)OUTPAINTExpand canvas in any direction (Up/Down/Left/Right px)RE-BGBiRefNet cuts the subject, you write a new background — no seamsCONTROLNETUnion 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) — multi-reference editing 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)<h3 id="qwen-image-edit-2511-one-studio-(tj)">Qwen Image Edit 2511 ONE STUDIO (TJ)</h3>Qwen2.5-VL 7B — image editing with multi-image context and camera angle controlModeWhat it doesT2IText to image with Qwen2.5-VL architectureI2IImage transformation with denoise controlEDITEdit images with text + up to 3 reference images (FluxKontext workflow)PAINTInpaint + Outpaint — mask editor + canvas expansionFACESWAPBFS LoRA face swap with auto promptANGLEDrag H/V/Z controls to rotate camera angle in 3D sceneUPSCALESeedVR2 AI upscale<h3 id="krea-2-one-studio-(tj)">Krea 2 ONE STUDIO (TJ)</h3>Krea AI generation modelModeWhat it doesT2IText to imageI2IImage-to-image transformationUPSCALESeedVR2 AI upscale<hr /><h2 id="features-that-actually-matter">Features That Actually Matter</h2>No wiring required — the node builds the ComfyUI graph internally. Set parameters, click Generate, done.Send to — one button passes your result to any other mode as source image. T2I → Inpaint → Upscale in 3 clicks.Language support — switch between 한국어 and English in Settings. All UI text, help docs, and error messages update on reload.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.Scroll zoom — mouse wheel to zoom into any result.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.Lightning LoRA — 4-step fast generation LoRA support for Klein, QE2511, and Krea2. Auto-sets Steps=4, CFG=1.Camera Angle LoRA — QE2511 ANGLE mode: configure once in Settings, applies automatically.Prompt Templates — save and recall prompt presets per node.Settings persist — model selection, prompt suffix, save folder, everything saves and survives restarts.<hr /><h2 id="seedvr2-upscale-built-in-to-all-nodes">SeedVR2 Upscale — Built In to All Nodes</h2>All four nodes include a full SeedVR2 AI upscale mode sharing the same model folder.<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</li><li>Color correction (LAB space)</li></ul><hr /><h2 id="bug-fixes-(v1.2)">Bug Fixes (v1.2)</h2><ul><li>Outpaint color param crash — <code>ImagePadKJ</code> color was sent as <code>[R,G,B]</code> array which ComfyUI interpreted as a node link → "Bad linked input" error. Fixed to <code>"R, G, B"</code> string format.</li><li>Klein outpaint generating new images — <code>GetImageSize</code> was reading the unscaled padded image, causing <code>EmptyFlux2LatentImage</code> and <code>VAEEncode</code> dimensions to mismatch → model ignored reference conditioning and drew entirely new content. Fixed to read from <code>ImageScaleToTotalPixels</code> output.</li><li>Help overlay links — URLs in all help docs are now clickable hyperlinks.</li><li>Language selector position — Language selector now appears right after Refresh Models (consistent across all 4 nodes).</li></ul><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><h3 id="2.-install-required-custom-nodes">2. Install required custom nodes</h3>Windows — just double-click:<pre><code>install_requirements.bat</code></pre>Mac / Linux:<pre><code>chmod +x install_requirements.sh
./install_requirements.sh</code></pre>This installs all 8 required nodes in one shot, skips anything already installed, and runs <code>pip install -r requirements.txt</code> for each. Restart ComfyUI when done.<hr />Or install manually via ComfyUI Manager:NodeRequired for<a target="_blank" rel="ugc" href="https://github.com/ltdrdata/ComfyUI-Impact-Pack">ComfyUI-Impact-Pack</a>Face Redraw mode (Z-Image)<a target="_blank" rel="ugc" href="https://github.com/ltdrdata/ComfyUI-Impact-Subpack">ComfyUI-Impact-Subpack</a>Required alongside Impact Pack<a target="_blank" rel="ugc" href="https://github.com/kijai/ComfyUI-KJNodes">ComfyUI-KJNodes</a>Klein KV Cache · ImagePadKJ · QE2511 FluxKontext nodes<a target="_blank" rel="ugc" href="https://github.com/numz/ComfyUI-SeedVR2_VideoUpscaler">ComfyUI-SeedVR2</a>UPSCALE mode (all nodes)<a target="_blank" rel="ugc" href="https://github.com/cubiq/ComfyUI_FaceAnalysis">ComfyUI_FaceAnalysis</a>Faceswap mode (Klein · QE2511)<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 (Z-Image)<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>Qwen Image Edit 2511:<ul><li>All model files: <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Qwen2.5-VL-7B-Image-Edit">Comfy-Org QE2511</a> → <code>models/diffusion_models/</code> + <code>models/text_encoders/</code> + <code>models/vae/</code></li></ul>Krea 2:<ul><li>All model files: <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Krea2">Comfy-Org Krea2</a> → <code>models/diffusion_models/</code> + <code>models/text_encoders/</code> + <code>models/vae/</code></li></ul>SeedVR2 (UPSCALE mode — all nodes):<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/numz/SeedVR2_comfyUI/tree/main">SeedVR2-models</a> → <code>models/SEEDVR2/</code></li><li>You need: 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><hr /><h2 id="the-other-half:-tj_node-wireless-workflow-architecture-toolkit">The Other Half: TJ_NODE — Wireless Workflow Architecture Toolkit</h2><a target="_blank" rel="ugc" href="https://github.com/designloves2/ComfyUI-TJ_NODE">ComfyUI-TJ_NODE</a> — the node package this studio was built to work with.If TJ NODE STUDIO ONE is the cockpit, TJ_NODE is the engine room.NodeWhat it doesPrompt Studio (TJ)Unified LLM prompt workflow — Auto / Enhancer / Image-to-Prompt modesScene Maker (TJ)Visual Beat-based prompt architecture — KO/EN/JP/CN auto-translateWireless ArchitectureMulti Router, Embedded Get/Set, Batch Workflow, Save Pipeline→ <a target="_blank" rel="ugc" href="https://github.com/designloves2/ComfyUI-TJ_NODE">github.com/designloves2/ComfyUI-TJ_NODE</a><hr /><h2 id="faq">FAQ</h2>Q: Which models does each node support? A: Z-Image ONE → Z-Image Turbo. Klein ONE → Flux.2 Klein 9B/4B. QE2511 ONE → Qwen2.5-VL 7B Image Edit. Krea2 ONE → Krea AI model. Each node is architecture-specific.Q: Do I need all the custom nodes? A: Only the ones for modes you use. SeedVR2 for upscale, FaceAnalysis for faceswap, RMBG for RE-BG, KJNodes for Klein/QE2511 outpaint. Core T2I/I2I works without most of them.Q: VRAM requirements? A: Flux.2 Klein 9B ~18–24GB; use CPU offload for lower VRAM. Z-Image Turbo is lighter. QE2511 7B ~16–20GB. Krea2 varies by model. SeedVR2 3B FP8 ~12GB.Q: My results don't look great. Is something broken? A: Probably not. Output quality is determined by the model, prompts, and LoRA choices — not the node. The node is just a UI shell.Q: Can I wire this into a normal ComfyUI workflow? A: The node internally generates its own graph. 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 /><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 — Wireless Workflow Architecture Toolkit: <a target="_blank" rel="ugc" href="https://github.com/designloves2/ComfyUI-TJ_NODE">https://github.com/designloves2/ComfyUI-TJ_NODE</a><hr /><h2 id="acknowledgements">Acknowledgements</h2>The core idea and architecture of this node family originated from <a target="_blank" rel="ugc" href="https://github.com/yanokusnir-ai">yanokusnir-ai</a>'s open-source work:<a target="_blank" rel="ugc" href="https://github.com/yanokusnir-ai/one-node-flux-2-klein">one-node-flux-2-klein</a> — The original single-node, all-in-one concept for Flux.2 KleinTJ NODE STUDIO ONE (Z-Image · Klein · QE2511 · Krea2) grew out of that idea — expanding it to multiple models and modes. Deep thanks to the original author for the inspiration and for keeping it open source. 🙏<hr />MIT License. Free forever. PRs welcome.

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

v1.4.1 (2026-06-28)Bug Fixes[Critical] LoRA 트리거 워드가 자꾸 지워지는 버그 수정 — 전체 4개 노드v1.4 (2026-06-28)New FeaturesPrompt Studio (LLM) 통합 — 프롬프트 확대창에 AI 강화 기능 추가TJ_NODE의 Prompt Studio를 4개 노드(Klein · QE2511 · Z-Image · Krea2) 프롬프트 확대창(<code>🔍</code>)에 직접 내장. TJ_NODE가 설치된 경우에만 활성화되며, 미설치 시 기존 편집 기능은 그대로 동작.새 탭 구조:<ul><li><code>✏️ Edit</code> — 기존 전체화면 텍스트 편집 (변경 없음)</li><li><code>✨ Enhance</code> — 현재 프롬프트를 GGUF LLM으로 강화 → 결과 팝업에서 [교체하기] / [닫기] 선택</li><li><code>🖼 Image→Prompt</code> — 이미지 업로드 또는 URL 다운로드 → 비전 LLM으로 프롬프트 생성 → 현재 모드 프롬프트로 전송</li></ul>설정 자동 기억 (localStorage):<ul><li>마지막 선택한 GGUF 모델, mmproj 파일, Vision Task, Model Format, Aesthetic, GPU Layers, n_ctx, Max Tokens, Temperature, Seed를 브라우저에 저장</li><li>4개 노드가 동일한 설정을 공유 — 한 번 설정하면 모든 노드에서 재사용</li></ul>TJ_NODE 미설치 시 설치 배너:<ul><li>미설치 상태에서 Enhance / Image→Prompt 탭 진입 시 설치 안내 UI 표시</li><li>GitHub 링크(<code>designloves2/ComfyUI-TJ_NODE</code>) 클릭 가능</li><li><code>⬇ 지금 설치하기</code> 버튼 → 서버에서 <code>git clone</code> + <code>pip install -r requirements.txt</code> 자동 실행</li><li>설치 완료 후 ComfyUI 재시작 안내</li></ul>

singleComfyuiNodeThatDoesEverythingT2II2I_v141.zip

v1.4.1

v1.3

v1.0

<h1 id="single-comfyui-node-that-does-everything-t2i-i2i-inpaint-outpaint-faceswap-camera-angle-bg-removal-and-ai-upscale.-no-wiring.-ever.-v.1.4.1">Single ComfyUI Node That Does Everything — T2I, I2I, Inpaint, Outpaint, Faceswap, Camera Angle, BG Removal, AND AI Upscale. No Wiring. Ever. v.1.4.1</h1>Free · Open Source · GitHub Four all-in-one nodes: Z-Image Turbo · Flux.2 Klein 9B · Qwen Image Edit 2511 · Krea 2 Korean / English UI — switch language in Settings<hr /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/d4e21f94-dea8-40aa-b2df-e27414431ab3/original=true/d4e21f94-dea8-40aa-b2df-e27414431ab3.jpeg" /><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 /><h2 id="what-is-this">What Is This</h2>TJ NODE 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 → EDIT → FACESWAP → ANGLE → 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.<hr /><h2 id="four-nodes-four-powerhouse-models">Four Nodes, Four Powerhouse Models</h2><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 region (DifferentialDiffusion)OUTPAINTExpand canvas in any direction (Up/Down/Left/Right px)RE-BGBiRefNet cuts the subject, you write a new background — no seamsCONTROLNETUnion 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) — multi-reference editing 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)<h3 id="qwen-image-edit-2511-one-studio-(tj)">Qwen Image Edit 2511 ONE STUDIO (TJ)</h3>Qwen2.5-VL 7B — image editing with multi-image context and camera angle controlModeWhat it doesT2IText to image with Qwen2.5-VL architectureI2IImage transformation with denoise controlEDITEdit images with text + up to 3 reference images (FluxKontext workflow)PAINTInpaint + Outpaint — mask editor + canvas expansionFACESWAPBFS LoRA face swap with auto promptANGLEDrag H/V/Z controls to rotate camera angle in 3D sceneUPSCALESeedVR2 AI upscale<h3 id="krea-2-one-studio-(tj)">Krea 2 ONE STUDIO (TJ)</h3>Krea AI generation modelModeWhat it doesT2IText to imageI2IImage-to-image transformationUPSCALESeedVR2 AI upscale<hr /><h2 id="features-that-actually-matter">Features That Actually Matter</h2>No wiring required — the node builds the ComfyUI graph internally. Set parameters, click Generate, done.Send to — one button passes your result to any other mode as source image. T2I → Inpaint → Upscale in 3 clicks.Language support — switch between 한국어 and English in Settings. All UI text, help docs, and error messages update on reload.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.Scroll zoom — mouse wheel to zoom into any result.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.Lightning LoRA — 4-step fast generation LoRA support for Klein, QE2511, and Krea2. Auto-sets Steps=4, CFG=1.Camera Angle LoRA — QE2511 ANGLE mode: configure once in Settings, applies automatically.Prompt Templates — save and recall prompt presets per node.Settings persist — model selection, prompt suffix, save folder, everything saves and survives restarts.<hr /><h2 id="seedvr2-upscale-built-in-to-all-nodes">SeedVR2 Upscale — Built In to All Nodes</h2>All four nodes include a full SeedVR2 AI upscale mode sharing the same model folder.<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</li><li>Color correction (LAB space)</li></ul><hr /><h2 id="bug-fixes-(v1.2)">Bug Fixes (v1.2)</h2><ul><li>Outpaint color param crash — <code>ImagePadKJ</code> color was sent as <code>[R,G,B]</code> array which ComfyUI interpreted as a node link → "Bad linked input" error. Fixed to <code>"R, G, B"</code> string format.</li><li>Klein outpaint generating new images — <code>GetImageSize</code> was reading the unscaled padded image, causing <code>EmptyFlux2LatentImage</code> and <code>VAEEncode</code> dimensions to mismatch → model ignored reference conditioning and drew entirely new content. Fixed to read from <code>ImageScaleToTotalPixels</code> output.</li><li>Help overlay links — URLs in all help docs are now clickable hyperlinks.</li><li>Language selector position — Language selector now appears right after Refresh Models (consistent across all 4 nodes).</li></ul><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_ONE.git</code></pre><h3 id="2.-install-required-custom-nodes">2. Install required custom nodes</h3>Windows — just double-click:<pre><code>install_requirements.bat</code></pre>Mac / Linux:<pre><code>chmod +x install_requirements.sh
./install_requirements.sh</code></pre>This installs all 8 required nodes in one shot, skips anything already installed, and runs <code>pip install -r requirements.txt</code> for each. Restart ComfyUI when done.<hr />Or install manually via ComfyUI Manager:NodeRequired for<a target="_blank" rel="ugc" href="https://github.com/ltdrdata/ComfyUI-Impact-Pack">ComfyUI-Impact-Pack</a>Face Redraw mode (Z-Image)<a target="_blank" rel="ugc" href="https://github.com/ltdrdata/ComfyUI-Impact-Subpack">ComfyUI-Impact-Subpack</a>Required alongside Impact Pack<a target="_blank" rel="ugc" href="https://github.com/kijai/ComfyUI-KJNodes">ComfyUI-KJNodes</a>Klein KV Cache · ImagePadKJ · QE2511 FluxKontext nodes<a target="_blank" rel="ugc" href="https://github.com/numz/ComfyUI-SeedVR2_VideoUpscaler">ComfyUI-SeedVR2</a>UPSCALE mode (all nodes)<a target="_blank" rel="ugc" href="https://github.com/cubiq/ComfyUI_FaceAnalysis">ComfyUI_FaceAnalysis</a>Faceswap mode (Klein · QE2511)<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 (Z-Image)<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>Qwen Image Edit 2511:<ul><li>All model files: <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Qwen-Image-Edit_ComfyUI/tree/main/split_files">Comfy-Org QE2511</a> → <code>models/diffusion_models/</code> + <code>models/text_encoders/</code> + <code>models/vae/</code></li></ul>Krea 2:<ul><li>All model files: <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Krea2">Comfy-Org Krea2</a> → <code>models/diffusion_models/</code> + <code>models/text_encoders/</code> + <code>models/vae/</code></li></ul>SeedVR2 (UPSCALE mode — all nodes):<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/numz/SeedVR2_comfyUI/tree/main">SeedVR2-models</a> → <code>models/SEEDVR2/</code></li><li>You need: 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><hr /><h2 id="the-other-half:-tj_node-wireless-workflow-architecture-toolkit">The Other Half: TJ_NODE — Wireless Workflow Architecture Toolkit</h2><a target="_blank" rel="ugc" href="https://github.com/designloves2/ComfyUI-TJ_NODE">ComfyUI-TJ_NODE</a> — the node package this studio was built to work with.If TJ NODE ONE is the cockpit, TJ_NODE is the engine room.NodeWhat it doesPrompt Studio (TJ)Unified LLM prompt workflow — Auto / Enhancer / Image-to-Prompt modesScene Maker (TJ)Visual Beat-based prompt architecture — KO/EN/JP/CN auto-translateWireless ArchitectureMulti Router, Embedded Get/Set, Batch Workflow, Save Pipeline→ <a target="_blank" rel="ugc" href="https://github.com/designloves2/ComfyUI-TJ_NODE">github.com/designloves2/ComfyUI-TJ_NODE</a><hr /><h2 id="faq">FAQ</h2>Q: Which models does each node support? A: Z-Image ONE → Z-Image Turbo. Klein ONE → Flux.2 Klein 9B/4B. QE2511 ONE → Qwen2.5-VL 7B Image Edit. Krea2 ONE → Krea AI model. Each node is architecture-specific.Q: Do I need all the custom nodes? A: Only the ones for modes you use. SeedVR2 for upscale, FaceAnalysis for faceswap, RMBG for RE-BG, KJNodes for Klein/QE2511 outpaint. Core T2I/I2I works without most of them.Q: VRAM requirements? A: Flux.2 Klein 9B ~18–24GB; use CPU offload for lower VRAM. Z-Image Turbo is lighter. QE2511 7B ~16–20GB. Krea2 varies by model. SeedVR2 3B FP8 ~12GB.Q: My results don't look great. Is something broken? A: Probably not. Output quality is determined by the model, prompts, and LoRA choices — not the node. The node is just a UI shell.Q: Can I wire this into a normal ComfyUI workflow? A: The node internally generates its own graph. 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 /><h2 id="github">GitHub</h2>TJ NODE ONE (this): <a target="_blank" rel="ugc" href="https://github.com/designloves2/ComfyUI-TJ_NODE_ONE">https://github.com/designloves2/ComfyUI-TJ_NODE_ONE</a>TJ NODE — Wireless Workflow Architecture Toolkit: <a target="_blank" rel="ugc" href="https://github.com/designloves2/ComfyUI-TJ_NODE">https://github.com/designloves2/ComfyUI-TJ_NODE</a><hr />MIT License. Free forever. PRs welcome.

Single ComfyUI Node That Does Everything — T2I, I2I, Inpaint, Outpaint, Faceswap, Camera Angle, BG Removal, AND AI Upscale. No Wiring. Ever.

v1.4.1 (June 28, 2026)Bug Fixes[Critical] Fixed a bug where LoRA trigger words kept getting deleted — 4 nodes in totalv1.4 (June 28, 2026)New FeaturesPrompt Studio (LLM) integration — Added AI-enhanced features to the prompt expansion windowTJ_NODE’s Prompt Studio is now directly embedded in the prompt expansion panel (🔍) for 4 nodes (Klein · QE2511 · Z-Image · Krea2). This feature is enabled only when TJ_NODE is installed; if not installed, the existing editing functions will continue to work as usual.New tab structure:✏️ Edit — Existing full-screen text editing (no changes)✨ Enhance — Enhance the current prompt using GGUF LLM → Select [Replace] / [Close] in the results pop-up🖼 Image→Prompt — Upload an image or download via URL → Generate a prompt using a vision LLM → Send to the current mode’s promptAutomatic Settings Saving (localStorage):Saves the last selected GGUF model, mmproj file, Vision Task, Model Format, Aesthetic, GPU Layers, n_ctx, Max Tokens, Temperature, and Seed to the browserAll 4 nodes share the same settings — set once, reuse across all nodesInstallation Banner if TJ_NODE is not installed:Displays an installation guide UI when accessing the Enhance / Image→Prompt tab while the software is not installedClickable GitHub link (designloves2/ComfyUI-TJ_NODE)⬇ “Install Now” button → Automatically runs <code>git clone</code> and <code>pip install -r requirements.txt</code> on the serverInstructions to restart ComfyUI after installation*** Translated with www.DeepL.com/Translator (free version) ***

Files with SHA256: 9aea33a27800f3e6c067e5ae94f4a622cacde08bb1a9fabba1c9ee8ea5e0aced