<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/1db2172d-dc55-467d-861b-bcc9c13b4dac/original=true/1db2172d-dc55-467d-861b-bcc9c13b4dac.jpeg" />Generated image with BHS Lora, Upscaled Latent, not a real actor or character This image made in the workflow Olemara's High Quality 2pass Qwen 2.1 Image Workflow v.1.01 and based on generated before image below <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/2213b1b1-932b-4c7e-af85-f01a6427e6ab/original=true/2213b1b1-932b-4c7e-af85-f01a6427e6ab.jpeg" /><h3 id="high-quality-2pass-upscaling-qwen-2.1-image-workflow-v.1.01">High Quality 2pass-Upscaling Qwen 2.1 Image Workflow v.1.01</h3><h3 id="why-this-workflow">Why this workflow</h3>Most people still bolt old face-swap stacks onto Qwen (InstantID-style helpers, separate swap pipelines, extra detectors). On Qwen-Image 2.1 that is mostly unnecessary.This graph is a Turbo DiT setup with an experimental but stable LoRA stack, a two-pass latent upscale, and one clear job in the default prompt path:*<code>&lt;image1&gt;</code> = full base frame (body, light, background)** *<code>&lt;image2&gt;</code> = donor head** BHS (Best Head Swap) LoRA does the head replacement inside the same generate — no second tool.On top of that:- Lightning / turbo LoRA — native few-step sampling (about 4–8 steps)- Multiple Angles LoRA — <code>&lt;mva&gt;</code> + azimuth / elevation labels (camera orbit without reshooting the ref)- Detail enhancer LoRA — <code>enhance this image</code> / <code>1_enhance this image</code>- Natural Exposure LoRA — more believable light and skin response- Optional NSFW / MysticXXX LoRAs when you need adult anatomy without fighting the base model- Wildcard prompt node so one graph can randomize angle + shot height every run- Texture-fix VAE + 2-pass latent upscale so the second pass tightens skin, edges, and micro-contrast instead of only blowing up pixelsFree to use. If something breaks on your machine, ask in the comments — happy to help. ---<h3 id="core-idea-(head-swap-done-right)">Core idea (head swap done right)</h3>1. Load base as <code>&lt;image1&gt;</code> (pose, wardrobe, room, lighting). 2. Load face/head donor as <code>&lt;image2&gt;</code>. Optionaly - background &lt;image3&gt;3. Keep the head_swap block in the prompt (see below). 4. Run. BHS LoRA strength usually 1.0; if it copies the whole ref instead of swapping, drop to ~0.8.You are not running a separate “face replace” app. The diffusion pass is the swap.---<h3 id="models-(what-the-graph-expects)">Models (what the graph expects)</h3><ul><li> DiT / UNET | <code>qwen-image-2.1-turbo-int8_convrot.safetensors</code> or <code>…-fp8.safetensors</code> - Turbo quant from Qwen-Image-2.1-Turbo-Quantized, Can be used with 8 steps without any lightning loras Text encoder | <code>qwen3vl_8b_nvfp4_heretic.safetensors</code> | CLIP type: <code>qwen_image</code> (Abliterated / Heretic NVFP4) |</li><li>VAE | <code>texture_fix_vae_for_qwen_image_2.1_bf16.safetensors</code> | Prefer this over a random VAE |</li></ul><h3 id="loras-used-in-the-workflow"> LoRAs used in the workflow</h3><ul><li> Lightning (few-step) | <code>pytorch_lora_weights.safetensors</code> | Optional Enables stable 4–8 step turbo sampling |</li><li> BHS – Best Head Swap | <code>bfs_head_v1.1_alternative_qwen_2.1.safetensors</code> | <code>head_swap:</code> block; strength 1.0 → 0.8 if needed |</li><li>Detail enhancer | <code>elusarcas-qwen2-1-detailer-v1.safetensors</code> | <code>enhance this image</code> / <code>1_enhance this image</code> |</li><li> Multiple angles | <code>multiple_angles_step1000.safetensors</code> | Start prompt with <code>&lt;mva&gt;</code> + labeled angle |</li><li> Natural Exposure | <code>Qwen-Image-2.1-Natural-Exposure-LoRA-4000.safetensors</code> | Natural light / exposure response |</li><li> MysticXXX (optional) | <code>Qwen-MysticXXX-v1.safetensors</code> | Adult / denser detail when needed |</li><li>NSFW 2.1 (optional) | <code>nsfw_lora_2.1_v2.safetensors</code> | Adult anatomy helper |</li></ul>---<h3 id="default-wildcard-style-prompt-(as-in-the-graph)"> Default wildcard-style prompt (as in the graph)</h3>Uses [custom-willdcard-prompt](<a target="_blank" rel="ugc" href="https://github.com/eliseipanov/custom-willdcard-prompt">https://github.com/eliseipanov/custom-willdcard-prompt</a>) so <code>{a\|b\|c}</code> picks one option per run:&lt;mva&gt; {front view 0°|front-right view 30°|front-right quarter view 60°|right side view 90°|back-right quarter view 120°|back-right view 150°|back view 180°|back-left view 210°|back-left quarter view 240°|left side view 270°|front-left quarter view 300°|front-left view 330°}, {high-angle shot|top-down shot|eye-level shot}.head_swap: start with &lt;image1&gt; as the base image, keeping its lighting, environment, and background. Remove completely the head of the Woman from &lt;image1&gt; and replace it with the head of the Woman from &lt;image2&gt;, strictly preserving the hair, eye color, nose structure from &lt;image2&gt;. Copy the direction of the eye, head rotation, micro expressions from &lt;image1&gt;, high quality, sharp details, 4k. Keep the colour scheme of &lt;image1&gt;camera and lighting: Cinematic realistic lighting, soft even light, realistic skin texture, realistic skin highlights, high detail. No text, no logos, no watermark. Shot made by fullframe Nikon camera with prime lens 85mm f/1.8. film grain, Dramatic lighting, cold long shadows.enhance this image: make all details of skin better, improve micro contrast, the point of black and point of white, sharp manual focus and white balance*<code>&lt;mva&gt;</code> rules (Multiple Angles LoRA)** - Put <code>&lt;mva&gt;</code> at the start of the prompt. - Use labels only (not raw degrees like <code>47°</code>). - Azimuth examples: <code>front view 0°</code>, <code>right side view 90°</code>, <code>back view 180°</code>, … - Elevation: <code>eye-level shot</code>, <code>elevated shot 30°</code>, <code>high-angle shot</code>, <code>top-down shot</code>. - Optional: add <code>close-up</code> at the end of the angle line. - LoRA strength ~**0.8–1.0** (lower if it pastes the ref instead of changing camera).BHS tip: if the head “sticks” wrong, lower BHS toward 0.8 and keep the <code>head_swap:</code> sentence explicit about hair / eyes / nose from <code>&lt;image2&gt;</code> and pose / gaze from <code>&lt;image1&gt;</code>.---<h3 id="sampler-quality-(practical)">Sampler / quality (practical)</h3>- Turbo model in workflow can be used as is with 8 steps. - Turbo model + Lightning Lora: about 4-5 steps, low CFG (graph example ~**1.1**), <code>euler</code> + <code>beta</code> is a known-good combo on this stack. You also can use er_sde + sgm_uniform or Euler_ancestral / Beta *(or sgm_uniform) - Always run the second pass / latent upscale if you care about skin pores, wet highlights, fabric, hair edges - Output size: keep multiples of 32; ResolutionSelector-style ~1.5 MP 3:2 is a solid default before upscale ---<h3 id="nodes-you-need">Nodes you need</h3>- ComfyUI build that includes TextEncodeQwenImage21 (Qwen-Image 2.1 path) - UNETLoader / CLIPLoader <code>qwen_image</code>) / VAELoader - LoadImage ×2 minimum <code>&lt;image1&gt;</code> base, <code>&lt;image2&gt;</code> head) - KSampler (+ second pass / latent upscaler node and model as in the graph) - Wildcard / custom-willdcard-prompt (or paste a fixed angle line) - Optional: LUT-from-reference (Just load the image you want you generated photo looks like by style) + light grade nodes if you keep the color group ---<h3 id="honest-limitations">Honest limitations</h3>- Quantized Turbo is fast, not magic — garbage refs in → garbage swap out - <code>&lt;mva&gt;</code> wants named angles; random degree numbers are unreliable - Stacking every LoRA at 1.0 can muddy the image; if detail collapses, turn optional NSFW/Mystic down first, not BHS - VRAM: INT8/FP8 Turbo is the point; the 2nd latent pass + texture VAE is what usually hurts small GPUs ---<h3 id="license-tone"> License / tone</h3>Workflow and notes are shared free for the community. Remix it, post results, tag the LoRA authors. Questions welcome in the comments — especially if BHS or <code>&lt;mva&gt;</code> misbehave on your exact quant.---### Workflow Notes (short install list)Weights1. DiT: <code>qwen-image-2.1-turbo-int8_convrot.safetensors</code> or FP8 twin → <code>models/diffusion_models</code> 2. TE: <code>qwen3vl_8b_nvfp4_heretic.safetensors</code> → <code>models/text_encoders</code> (type <code>qwen_image</code>) 3. VAE: <code>texture_fix_vae_for_qwen_image_2.1_bf16.safetensors</code> → <code>models/vae</code> 4. LoRAs → <code>models/loras/</code>: - <code>pytorch_lora_weights.safetensors</code> (Lightning Lora from ) - <code>bfs_head_v1.1_alternative_qwen_2.1.safetensors</code> (BHS) - <code>elusarcas-qwen2-1-detailer-v1.safetensors</code> - <code>multiple_angles_step1000.safetensors</code> - <code>Qwen-Image-2.1-Natural-Exposure-LoRA-4000.safetensors</code> - optional: <code>Qwen-MysticXXX-v1.safetensors</code>, <code>nsfw_lora_2.1_v2.safetensors</code> Custom nodes- Qwen-Image 2.1 encoder support (stock recent Comfy or pack that ships <code>TextEncodeQwenImage21</code>) - [custom-willdcard-prompt](<a target="_blank" rel="ugc" href="https://github.com/eliseipanov/custom-willdcard-prompt">https://github.com/eliseipanov/custom-willdcard-prompt</a>) for <code>{a|b|c}</code> - Whatever upscale/2pass nodes are already wired in the uploaded graph Run order1. <code>&lt;image1&gt;</code> base, <code>&lt;image2&gt;</code> head, also optional (disabled) &lt;image3&gt; for background2. Confirm BHS + Lightning + (optional) angles/detail/exposure strengths 3. Generate few-step turbo → let 2pass latent upscale finish 4. Only then push grade/LUT Support: comment with GPU, quant (INT8/FP8), and which LoRA strengths you used if the head swap or <code>&lt;mva&gt;</code> drifts.

Olemara's High Quality 2pass Qwen 2.1 Image Workflow v.1.01

Free and rady for use Workflow Olemara's High Quality 2pass Qwen 2.1 Image Workflow v.1.01, working without any bugs with VRAM 24 Gb + (For example on vast.ai instances here <a target="_blank" rel="ugc" href="https://cloud.vast.ai/?ref_id=398237">https://cloud.vast.ai/?ref_id=398237</a>)

Files with SHA256: f90d10780e7ea04f5a74618439b11752eb65e0839f473f43c49985a34df11790