<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/34a584b9-b6a5-4266-959b-cc03653e3fca/original=true/34a584b9-b6a5-4266-959b-cc03653e3fca.jpeg" /><h2 id="no-gpu-run-it-on-runpod-serverless">🚀 No GPU? Run it on Runpod Serverless</h2><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/5000cb1b-6161-49d6-ab0e-76616bbd4422/original=true/5000cb1b-6161-49d6-ab0e-76616bbd4422.jpeg" />ForgeHub — desktop app (macOS / Windows / Linux) that drives this stack: wildcard quick-insert (<code>__qwen21/*</code>), reference image upload, QwenVL prompt enhancer, and a chat that writes edit prompts for you.⬇️ <a target="_blank" rel="ugc" href="https://github.com/huchukato/ForgeHub/releases">Download ForgeHub</a> · ⚡ <a target="_blank" rel="ugc" href="https://console.runpod.io/hub/listing/huchukato/runpod-qwen21?ref=ioakclrv">Deploy the worker on Runpod</a>Deploy the endpoint → attach a ≥60 GB network volume → paste the endpoint ID into ForgeHub settings → generate. All models auto-download on first boot.Prefer a full ComfyUI pod instead? Same stack, one click:<ul><li>🟣 <a target="_blank" rel="ugc" href="https://console.runpod.io/hub/template/piawvhe127?ref=ioakclrv">RunPod — ComfyUI Qwen Image Edit 2.1 pod template</a></li><li>🔵 <a target="_blank" rel="ugc" href="https://cloud.vast.ai/?ref_id=749134&amp;creator_id=749134&amp;name=OneClick%20-%20ComfyUI%20-%20Qwen%20Image%20Edit%202.1%20-%20Qwen3.5%20AutoPrompt">Vast.ai — OneClick ComfyUI Qwen Image Edit 2.1 - Qwen3.5 AutoPrompt</a></li></ul><hr /><h1 id="qwenimageedit21">QwenImageEdit21</h1>The Only Workflow You Will Need<h3 id="wildcard-edits-or-multi-reference-or-gguf-support">🎨 Wildcard Edits | Multi-Reference | GGUF Support</h3>ComfyUI-QwenVL-Mod — Qwen Image Edit 2.1 with Wildcard-driven auto-prompting<hr /><h2 id="what-is-this">🎯 What Is This?</h2>A Qwen Image Edit 2.1 workflow pack built for reference-driven editing — swap faces, heads, outfits, poses, expressions, backgrounds and more using natural-language edit prompts pulled from a curated wildcard library.The prompt pipeline:<pre><code>WildcardProcessor.processed_text → QwenVL_Unified.prompt
QwenVL_Unified.RESPONSE → PixaromaShowText → TextEncodeQwenImage21.prompt
Load Image 1 / Load Image 2 → QwenVL_Unified + TextEncodeQwenImage21
</code></pre><ul><li>🎲 Wildcard edits: type <code>__qwen21/sheet__</code> and a full professional editing prompt appears — no prompt-writing skill needed</li><li>🧠 QwenVL auto-enhance: the enabled QwenVL Unified node uses the dedicated <code>IMG › Qwen Edit</code> preset to rewrite and polish the expanded edit instruction before it reaches the text encoder</li><li>🖼️ Multi-reference: <code>&lt;image1&gt;</code> and <code>&lt;image2&gt;</code> inside wildcards bind to the two connected images, in order</li><li>💾 GGUF support: run the Qwen Image diffusion model in GGUF (Q8_0 → Q3_K_M) on low-VRAM GPUs or Apple Silicon</li></ul><hr /><h2 id="required-models">📦 Required Models</h2>Drop into <code>ComfyUI/models/</code> — pick one diffusion model, one text encoder, the VAE:<h3 id="diffusion_models">diffusion_models</h3><ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/diffusion_models/qwen_image_2.1_bf16.safetensors">qwen_image_2.1_bf16.safetensors</a> — full quality</li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/diffusion_models/qwen_image_2.1_int8_convrot.safetensors">qwen_image_2.1_int8_convrot.safetensors</a> — lighter</li></ul><h3 id="text_encoders">text_encoders</h3><ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/text_encoders/qwen3vl_8b_bf16.safetensors">qwen3vl_8b_bf16.safetensors</a> — full</li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/text_encoders/qwen3vl_8b_int8_convrot.safetensors">qwen3vl_8b_int8_convrot.safetensors</a> — lighter</li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/text_encoders/qwen3vl_8b_w4a8.safetensors">qwen3vl_8b_w4a8.safetensors</a> — smallest, low VRAM</li></ul><h3 id="vae">vae</h3><ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/vae/qwen_image_2.1_vae_bf16.safetensors">qwen_image_2.1_vae_bf16.safetensors</a></li></ul><h3 id="gguf-option-(low-vram-mac)">🐋 GGUF option (low VRAM / Mac)</h3>From <a target="_blank" rel="ugc" href="https://huggingface.co/Abiray/Qwen-Image-2.1-GGUF">Abiray/Qwen-Image-2.1-GGUF</a> — drop in <code>models/diffusion_models/</code> and load with Unet Loader (GGUF):<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Abiray/Qwen-Image-2.1-GGUF/resolve/main/qwen_image_2.1_Q8_0.gguf">Q8_0</a> · <a target="_blank" rel="ugc" href="https://huggingface.co/Abiray/Qwen-Image-2.1-GGUF/resolve/main/qwen_image_2.1_Q6_K.gguf">Q6_K</a> · <a target="_blank" rel="ugc" href="https://huggingface.co/Abiray/Qwen-Image-2.1-GGUF/resolve/main/qwen_image_2.1_Q5_K_M.gguf">Q5_K_M</a> · <a target="_blank" rel="ugc" href="https://huggingface.co/Abiray/Qwen-Image-2.1-GGUF/resolve/main/qwen_image_2.1_Q4_K_M.gguf">Q4_K_M</a> · <a target="_blank" rel="ugc" href="https://huggingface.co/Abiray/Qwen-Image-2.1-GGUF/resolve/main/qwen_image_2.1_Q4_K_S.gguf">Q4_K_S</a> · <a target="_blank" rel="ugc" href="https://huggingface.co/Abiray/Qwen-Image-2.1-GGUF/resolve/main/qwen_image_2.1_Q3_K_M.gguf">Q3_K_M</a></li></ul>Repo: <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Qwen-Image-2.1">Comfy-Org/Qwen-Image-2.1</a> · <a target="_blank" rel="ugc" href="https://modelscope.cn/models/Comfy-Org/Qwen-Image-2.1">ModelScope mirror</a><h3 id="loras-(optional-for-__qwen21enhance__)">loras (optional — for <code>__qwen21/enhance__</code>)</h3><ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/reverentelusarca/elusarcas-qwen-2.1-detail-enhancer-lora/resolve/main/elusarcas-qwen2-1-detailer-v1.safetensors">elusarcas-qwen2-1-detailer-v1.safetensors</a> → <code>models/loras/</code> — detail enhancement, creative upscale, photo restoration. Only ~80MB; add a Load LoRA node between the diffusion model loader and the sampler.</li></ul><hr /><h2 id="wildcard-edit-library">🎲 Wildcard Edit Library</h2>Type <code>__qwen21/&lt;name&gt;__</code> for a specific edit, or <code>__qwen21/*__</code> for a random one. <code>&lt;image1&gt;</code> and <code>&lt;image2&gt;</code> are the two connected reference images, in order.<ul><li><code>__qwen21/sheet__</code> (1 img) — quick character sheet: front / side / back + headshot</li><li><code>__qwen21/prosheet__</code> (1 img) — pro reference sheet (image2 optional as extra detail ref)</li><li><code>__qwen21/newscene__</code> (1 img) — subject into a new random scene (location wildcard)</li><li><code>__qwen21/backgroundswap__</code> (2 img) — replace background of image1 with image2</li><li><code>__qwen21/enhance__</code> (1 img) — detail enhance / restore / creative upscale — pairs with the <a target="_blank" rel="ugc" href="https://huggingface.co/reverentelusarca/elusarcas-qwen-2.1-detail-enhancer-lora">detailer LoRA</a> (optional, see below)</li><li><code>__qwen21/faceswap__</code> (2 img) — face of image2 ← identity from image1</li><li><code>__qwen21/headswap__</code> (2 img) — head + hair of image2 ← head from image1</li><li><code>__qwen21/eyeswap__</code> (2 img) — eyes of image2 ← eyes from image1</li><li><code>__qwen21/makeupswap__</code> (2 img) — makeup of image2 onto image1 — identity kept</li><li><code>__qwen21/hairswap__</code> (2 img) — hair of image2 ← hair from image1</li><li><code>__qwen21/outfitswap__</code> (2 img) — dress image1 subject with the outfit from image2</li><li><code>__qwen21/poseswap__</code> (2 img) — copy body pose from image2 onto image1</li><li><code>__qwen21/expressionswap__</code> (2 img) — copy expression from image2 onto image1</li><li><code>__qwen21/multiref__</code> (3 img) — persons from image1 + image2 together inside image3</li><li><code>__qwen21/multiedit__</code> (1 img) — applies every edit you describe at once</li><li><code>__qwen21/timeweather__</code> (1 img) — change time/weather to what you describe</li><li><code>__qwen21/removeobject__</code> (1 img) — remove the object you name + inpaint</li><li><code>__qwen21/insertobject__</code> (1 img) — insert the object you describe</li><li><code>__qwen21/textedit__</code> (1 img) — replace visible text with yours</li><li><code>__qwen21/novelview__</code> (1 img) — camera angle you describe</li><li><code>__qwen21/anime2real__</code> (1 img) — anime → photorealistic human</li><li><code>__qwen21/real2anime__</code> (1 img) — photo → manhwa / webtoon illustration</li></ul><h3 id="how-it-works">How It Works</h3><ol><li>The WildcardProcessor node (TagForge) sits before the QwenVL enhancer</li><li>At queue time each <code>__wildcard__</code> token is replaced by the matching entry from <code>wildcards/pmp/qwen21.yaml</code></li><li>The expanded text goes through QwenVL Unified → <code>PixaromaShowText</code> → <code>TextEncodeQwenImage21</code></li><li>Different seed = different wildcard picks — pin the seed for reproducible results</li></ol><h3 id="keep-qwenvl-unified-enabled">⚠️ Keep QwenVL Unified enabled</h3>The workflow ships with QwenVL Unified enabled and the dedicated <code>IMG › Qwen Edit</code> preset selected. The node receives the expanded wildcard instruction and both source images, then returns an edit prompt that preserves the <code>&lt;image1&gt;</code> / <code>&lt;image2&gt;</code> bindings for <code>TextEncodeQwenImage21</code>.Do not bypass QwenVL Unified in this workflow. Because the node has IMAGE inputs and a STRING output, ComfyUI's generic bypass can forward an image tensor to <code>PixaromaShowText</code> instead of the generated text. If Show Text displays <code>Tensor shape=(...)</code>, re-enable QwenVL Unified and confirm that <code>IMG › Qwen Edit</code> is selected.The preset requires ComfyUI-QwenVL-Mod 2.9.8 or newer. Update to the current 2.10.2 release, restart ComfyUI completely, and refresh the browser if the preset is missing.<h3 id="tips">Tips</h3><ul><li>Image order matters: for faceswap/headswap/outfitswap, image1 is the source (identity/outfit), image2 is the target photo</li><li>Edits that need a target: for <code>removeobject</code>, <code>insertobject</code>, <code>textedit</code>, <code>novelview</code>, <code>timeweather</code>, <code>multiedit</code> — describe what to change next to the wildcard, e.g. <code>a red parked car, __qwen21/removeobject__</code></li><li>Combine with text: <code>__qwen21/backgroundswap__, dramatic mood</code> works fine — wildcards expand inline</li><li>Add your own: drop a <code>.txt</code> in <code>ComfyUI/models/wildcards/</code> (one prompt per line) → becomes <code>__filename__</code>; YAML files support nested keys like <code>__name/subkey__</code></li></ul><h3 id="required-custom-nodes">Required Custom Nodes</h3><ul><li>ComfyUI-TagForge — WildcardProcessor + wildcard library: <a target="_blank" rel="ugc" href="https://github.com/huchukato/ComfyUI-TagForge">huchukato/ComfyUI-TagForge</a></li><li>ComfyUI-QwenVL-Mod 2.9.8+ (2.10.2 recommended) — QwenVL Unified enhancer node with the <code>IMG › Qwen Edit</code> preset: <a target="_blank" rel="ugc" href="https://github.com/huchukato/ComfyUI-QwenVL-Mod">huchukato/ComfyUI-QwenVL-Mod</a></li><li>ComfyUI-GGUF (optional, for GGUF quants): <a target="_blank" rel="ugc" href="https://github.com/city96/ComfyUI-GGUF">city96/ComfyUI-GGUF</a></li></ul><hr /><h2 id="why-this-pack">🌟 Why This Pack?</h2><ul><li>Zero prompt-writing: production-grade edit prompts selected with a single wildcard token</li><li>Identity-safe edits: prompts are written to preserve face, proportions and style across swaps</li><li>Flexible: BF16, INT8 ConvRot, W4A8 text encoder or full GGUF — scales from 24GB cards down to unified-memory Macs</li><li>Extensible: your own <code>.txt</code>/<code>.yaml</code> wildcards plug straight in</li></ul><hr /><h2 id="credits">📋 Credits</h2><ul><li>Qwen Image 2.1 — <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Qwen-Image-2.1">Qwen Team / Alibaba</a></li><li>GGUF quants — <a target="_blank" rel="ugc" href="https://huggingface.co/Abiray/Qwen-Image-2.1-GGUF">Abiray/Qwen-Image-2.1-GGUF</a></li><li>Base template — Comfy-Org official editing workflow</li><li>TagForge + QwenVL-Mod — <a target="_blank" rel="ugc" href="https://github.com/huchukato">huchukato</a></li></ul><hr />Built with ❤️ for the ComfyUI community

[Qwen Image 2.1] Multi-Reference Editing 🎨 Wildcards | Faceswap • Headswap • Outfit • Pose | Auto-Enhance + Detailer LoRA | GGUF

v1.3 — T2I workflow + Edit cleanupNew:- QwenImage21-T2I-Wildcards-Qwen3.5 — dedicated text-to-image workflow: tag-style prompt → WildcardProcessor (pmp/* wildcards) → QwenVL prompt enhancer with selectable style tags → Qwen Image 2.1. No input images needed — wildcards and plain text both work.Fixed / Improved:- QwenImageEdit21 reworked — the generation stack now lives in a clean subgraph; removed leftover scaffolding (ComfySwitch, Fast Groups Bypasser, duplicate LoraLoader). The elusarcas detailer LoRA now loads through a single LoraLoaderModelOnly.- Corrected node wiring and model references in both workflows — the wildcards, enhancer and detailer chain now patch and execute correctly end-to-end.Note: the QwenVL enhancer is active by default in both workflows — inthe Edit workflow you can still bypass it via the group switch to sendedit wildcards to the encoder verbatim.

Files with SHA256: 8cef98592a77cb2117ed506ce2cc4aa4144edcf4ac150f3f85c768ae4e4e3973