Simple Image Upscaling Workflow for low VRAM user <span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">(Extreme FAST)<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">Low VRAM專用簡單image放大Workflow (極快)Once again, what I'm sharing today is an ultra-simple workflow. It follows my usual style: simple, practical, completely spaghetti-free, and 100% Low VRAM friendly.I’ve covered the rest of the details directly inside the workflow, so I won't repeat them here.The purpose of separating the Upscale model from the rest of the workflow is to reduce the VRAM load. It also provides users with options, as there is no need to rely on a single model for every task. For instance, you can generate images with KREA 2 and then upscale them using Klein 9B. This keeps the interface much cleaner and more organized.把Upscale model與生成圖片的Workflow分開目的，就是為了VRAM負擔亦提供用家選項, 沒必要把所有任務都交由同一model負責你可以用KREA 2生成圖片, 然後用Klein 9B放大令介面更簡潔乾淨今次分享的亦是超簡單Workflow。亦遵循我一貫作風, 簡單, 實際, 不會意粉化, Low VRAM能用其他詳細在workflow中有說明就不再重覆了Make sure you had selected 1 or 2 in the green node before click RUN記得先選好1或2再按RUN1: SeedVR2+RTX2: Upscale with model + Ultra SD Upscale<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/bf85d71c-25b5-403d-9fa5-5a69ea4cf184/original=true/bf85d71c-25b5-403d-9fa5-5a69ea4cf184.jpeg" />time difference on my PC:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/d7297032-3be3-4e07-95bd-c2257ae5e68f/original=true/d7297032-3be3-4e07-95bd-c2257ae5e68f.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/c6ff2349-5aed-485a-a0e6-50edd5aaa87e/original=true/c6ff2349-5aed-485a-a0e6-50edd5aaa87e.jpeg" />(有中文)Why Does AI Always Give You a Close-Up? How to Control Framing with a Four-Stage Ladder 📷✨<a target="_blank" rel="ugc" href="https://civarchive.com/articles/34198/why-does-ai-always-give-you-a-close-up-how-to-control-framing-with-a-four-stage-ladder">https://civarchive.com/articles/34198/why-does-ai-always-give-you-a-close-up-how-to-control-framing-with-a-four-stage-ladder</a>Video Upscale simple workflow for low VRAM (Extreme FAST)Low VRAM專用簡單影片放大Workflow (極快)<a target="_blank" rel="ugc" href="https://civarchive.com/models/2871874/video-upscale-workflow-for-low-vram-extreme-fast">https://civarchive.com/models/2871874/video-upscale-workflow-for-low-vram-extreme-fast</a>How to merge videos with FFmpeg in ComfyUI Desktop (2026)?如何使用FFmpeg在ComfyUi Desktop(2026)中合拼片段<a target="_blank" rel="ugc" href="https://civarchive.com/articles/34119/how-to-merge-videos-with-ffmpeg-in-comfyui-desktop-2026">https://civarchive.com/articles/34119/how-to-merge-videos-with-ffmpeg-in-comfyui-desktop-2026</a>This is a simple workflow for VoxCPM2, featuring a single-speaker setup and a multi-speaker setup.這是作為輸助語音生成VoxCPM2的Workflow<a target="_blank" rel="ugc" href="https://civarchive.com/models/2865730/voxcpm2-simple-workflow-for-low-v-ram">https://civarchive.com/models/2865730/voxcpm2-simple-workflow-for-low-v-ram</a>I've also written an article about VoxCPM2—feel free to check it out if you're interested!還有關於VoxCPM2的文章, 有興趣可以看看<a target="_blank" rel="ugc" href="https://civarchive.com/articles/34023/why-do-we-still-need-a-audio-generation-model-even-minimax-h3-could-produce-audio">https://civarchive.com/articles/34023/why-do-we-still-need-a-audio-generation-model-even-minimax-h3-could-produce-audio</a>Krea 2 Quadview Simple workflowQaudview might lock the character appearence during mulit reference in long video generation.Krea 2 多角度參考圖 Workflow多角度參考圖可以在影片生成時鎖死角色外表, 特別在長片分多次生成中作為參考圖使用<a target="_blank" rel="ugc" href="https://civarchive.com/models/2867878/krea-2-quadview-simple-workflow-for-low-vram">https://civarchive.com/models/2867878/krea-2-quadview-simple-workflow-for-low-vram</a>Just post a quick guide for the newbie like me to install triton/Sage Attention/Flash Attention in new ComfyUI Desktop 2026.Check it if you need it.寫了三篇簡單教學如何在新ComfyUI Desktop 2026中安裝triton/Sage Attention/Flash Attention有需要的朋友可以看一下.<a target="_blank" rel="ugc" href="https://civarchive.com/articles/33944/how-to-install-trition-in-new-comfyui-desktop-2026">https://civarchive.com/articles/33944/how-to-install-trition-in-new-comfyui-desktop-2026</a><a target="_blank" rel="ugc" href="https://civarchive.com/articles/33949/how-to-install-sage-attention-in-new-comfyui-desktop-2026">https://civarchive.com/articles/33949/how-to-install-sage-attention-in-new-comfyui-desktop-2026</a><a target="_blank" rel="ugc" href="https://civarchive.com/articles/33950/how-to-install-flash-attention-in-new-comfyui-desktop-2026">https://civarchive.com/articles/33950/how-to-install-flash-attention-in-new-comfyui-desktop-2026</a>In the green options, Select 1 is SeedVR2, and Select 2 is Ultra SD Ultimate (or Ultimate SD Upscale).<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/ff543936-623d-4b0e-9a35-cfd751869407/original=true/ff543936-623d-4b0e-9a35-cfd751869407.jpeg" />For users with Low VRAM, using SeedVR2 (the purple node) for image upscaling is an excellent choice.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/34af0a54-e818-4b12-b216-77e33ecf437f/original=true/34af0a54-e818-4b12-b216-77e33ecf437f.jpeg" />Here are several critical parameter settings to keep in mind when configuring SeedVR2:1. Upscaler Model: Low VRAM users must select <code>seedvr2_ema_3b-Q4KM.gguf</code>. This is the key to achieving fast speeds on low VRAM systems.2. Offload Device: Set all offload device options to CPU.3. Tile Size: This value must be a multiple of 8. It is currently configured to 896.4. Attention Mode: Your choice here significantly impacts speed. While you can select SDPA if you prefer not to install anything extra, Sage Attention 2 is a much better choice.Sage Attention installation tutorial:<a target="_blank" rel="ugc" href="https://civarchive.com/articles/33949/how-to-install-sage-attention-in-new-comfyui-desktop-2026">https://civarchive.com/articles/33949/how-to-install-sage-attention-in-new-comfyui-desktop-2026</a>5. Resolution: I have set this to 2144 because I want the shorter side of the upscaled image to be 2144 pixels. For example, if your original image is 640x960 and your target 4K resolution is 2160x3840, you should change this parameter to 2160.6. Memory Management: Since this is optimized for low VRAM, the offload device must be set to CPU. Any overflowing data will be offloaded to your system RAM to prevent Out of Memory (OOM) errors.7. Process Flow: The image is first slightly restored using RTX Denoise. RTX does not apply any scaling; the entire upscaling workload is handled by SeedVR2.---### Ultra SD Upscale Configuration (Blue Node)<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3f34055a-52e5-4e9f-b0e2-96f25d5fb2a0/original=true/3f34055a-52e5-4e9f-b0e2-96f25d5fb2a0.jpeg" />* Model Selection: You can freely choose your preferred model, such as Krea 2, Z-image, Klein 9B, or any other model. Just pair it with the appropriate VAE and Text Encoder.* Turbo LoRA: This is set to Bypass by default. If you wish to use a RAW or base model, you can load an appropriate Turbo LoRA here.* LoRA Slots: Since this workflow is intended for anime image upscaling, multiple LoRA slots are reserved for character outfits and styles. You can leave them blank if they are not needed.* Upscale Scale: The default scale is set to 2.2x purely based on my personal preference. You can adjust this size freely.#### Download Links for Upscale ModelsYou can choose between <code>4xClearRealityV1</code>, <code>4xUltraSharpV2</code>, or <code>4xAnimeSharp</code>. Download them from the links below and place them into your <code>comfyUI/models/upscale_models/</code> directory:UltraSharpV2:<a target="_blank" rel="ugc" href="https://openmodeldb.info/models/4x-UltraSharpV2">https://openmodeldb.info/models/4x-UltraSharpV2</a>AnimeSharp:<a target="_blank" rel="ugc" href="https://openmodeldb.info/models/4x-AnimeSharp">https://openmodeldb.info/models/4x-AnimeSharp</a>ClearrealityV1:<a target="_blank" rel="ugc" href="https://openmodeldb.info/models/4x-ClearRealityV1">https://openmodeldb.info/models/4x-ClearRealityV1</a>#圖片放大簡單 Workflow在綠色選項中，**select 1** 是 <code>SeedVR2</code>，**select 2** 是 <code>Ultra SD Ultimate</code>。對於 Low VRAM 的使用者來說，使用 <code>SeedVR2</code>（紫色 Node）來放大圖片是一個非常不錯的選擇。---一、 SeedVR2 參數設定注意事項在設定 <code>SeedVR2</code> 的參數時，有以下幾項必須注意的要點：1. Upscaler model：Low VRAM 使用者務必選擇 <code>seedvr2_ema_3b-Q4KM.gguf</code>，這是確保 Low VRAM 運行速度快的關鍵。2. Offload device：請全選設定為 CPU。3. Tile size：必須是 8 的倍數，目前設定值為 <code>896</code>。4. Attention_mode：是否選擇 <code>Sage attention</code> 對速度影響極大。如果不想另外安裝可以選擇 <code>SDPA</code>，但 <code>Sage Attention2</code> 會是比較推薦的優質選擇。5. Resolution：目前設定為 <code>2144</code>，目的是希望放大後的**短邊**長度為 2144 像素。* 舉例：若原圖是 <code>640x960</code>，而目標 4K 解析度為 <code>2160x3840</code>，則此處參數應修改為 <code>2160</code>。6. 記憶體管理：由於是提供給 Low VRAM 使用者，**offload device 必須設定為 CPU**。所有溢出的數據都會 offload 到系統 RAM 中，以避免發生 OOM（記憶體不足）錯誤。7. 處理流程：圖片會先由 <code>RTX denoise</code> 進行輕微修復（RTX 不包含放大倍數），後續的放大工作則全權交給 <code>SeedVR2</code> 處理。---二、 Ultra SD Upscale (藍色 Node) 設定說明* Model 選擇：可以自由挑選。無論是使用 <code>Krea 2Z-imageKlein 9B</code> 或其他任意模型皆可，只需配合適當的 <code>VAE</code> 和 <code>Text encoder</code> 即可正常運作。* Turbo LoRA：預設狀態為 Bypass。如果想使用 RAW 或 base model，可以自行放入適當的 Turbo LoRA。* LoRA 欄位：因為主要用於 Anime（動漫）影像放大，工作流中預留了較多欄位供角色衣服和 Style（畫風）使用。不需要的部分直接留空即可。* 放大倍數：預設為 <code>2.2倍</code>（純屬個人使用習慣），您可以根據需求隨意調節大小。---三、 放大模型下載資訊您可以選擇 <code>4xClearRealityV14xUltraSharpV2</code> 或 <code>4xAnimeSharp</code>。請點擊下方連結下載，並將檔案放置於您的 ComfyUI 目錄中：👉 路徑<code>comfyUI/models/upscale_models/</code>UltraSharpV2:<a target="_blank" rel="ugc" href="https://openmodeldb.info/models/4x-UltraSharpV2">https://openmodeldb.info/models/4x-UltraSharpV2</a>AnimeSharp:<a target="_blank" rel="ugc" href="https://openmodeldb.info/models/4x-AnimeSharp">https://openmodeldb.info/models/4x-AnimeSharp</a>ClearrealityV1:<a target="_blank" rel="ugc" href="https://openmodeldb.info/models/4x-ClearRealityV1">https://openmodeldb.info/models/4x-ClearRealityV1</a>

Simple Image Upscaling Workflow for low VRAM (Extreme FAST)

add a green switch to choose the upscale modelput all unnecessary node into subgraphkeep the interface clean and tidy, spaghetti-free.

Files with SHA256: d75fe7dce5f625fe62b9fda6a754ddc27433313613a190658b4bdc6b1ba52f66