<p>This <code>ComfyUi</code> <code>fl2va</code> | <code>t2va</code> workflow use <code>Turbo 4steps</code> with <code>Latent Upscaling</code> to generate <code>1.0 mp</code> good quality result with short generating time.</p><p></p><p><strong>If this workflow works great for you, please support me with ⚡⚡buzz⚡⚡, so I can do more testing and share more works, thank you!🦉</strong>❤️<strong>☕</strong></p><p></p><hr /><p></p><h3 id="generating-pipeline:">Generating Pipeline:</h3><ol><li><p>Turbo generating with <code>0.2mp</code> in the first 2 steps.</p></li><li><p>Upscale the H3 AV<code>Latent</code> to <code>1.0mp</code> using trained neural network.</p></li><li><p>Re-sample the H3 AV latent at higher resolution in the last 2 steps.</p></li></ol><p>This ensures decent generating speed + Low <code>VRAM</code> requirement + Naturel fine upscaling details.</p><p></p><hr /><h3 id="features:">Features:</h3><ul><li><p>🚀<code>Turbo</code> <code>4 steps</code> generating (2+2)</p></li><li><p>🔍<code>Latent</code> Upscaling</p></li><li><p>🗒️Multiple <code>LoRa</code> support</p></li><li><p>🖥️Runtime <code>Preview</code> when generating</p></li><li><p>🗜️<code>GGUF</code> models</p></li></ul><hr /><p></p><p>Tested on <code>RTX3080(laptop) 16GB VRAM</code> .</p><ul><li><p>Choose <code>Q8</code> <code>Base model</code> and <code>Q5</code> <code>text encoder</code> if you have <code>16GB VRAM</code>.</p></li></ul><ul><li><p>Choose <code>Q4</code> <code>Base model</code> and <code>Q4</code> <code>text encoder</code> if you have less than <code>16GB VRAM</code>.</p></li></ul><p></p><hr /><p></p><p></p><h2 id="model-links">Model Links</h2><hr /><h3 id="base-model-minimaxh3turbogguf_fl2vturbo4stepv10768.gguf(q8)">- <span style="color:rgb(250, 176, 5)">Base model</span> -<br /><a target="_blank" rel="ugc" href="https://civarchive.com/models/2851657/minimax-h3-turbo-gguf?modelVersionId=3220323">minimaxH3TurboGGUF_fl2vTurbo4stepV10768.gguf</a>(Q8)</h3><ul><li><p>place it under: <code>models\unet\</code></p></li><li><p><code>Choose Q8 for 16GB+ VRAM, Q4 for less</code></p></li></ul><p>Thanks <a target="_blank" rel="ugc" href="https://civarchive.com/user/molbal"><strong>molbal </strong></a>for his awesome model!<br /></p><hr /><p></p><h3 id="text-encoder-minimax-h3-encoder-q5_k_m.gguf(q5)">- <span style="color:rgb(250, 176, 5)">Text Encoder</span> -<br /><a target="_blank" rel="ugc" href="https://civarchive.com/models/2834385/minimax-h3-text-encoder-gguf-qwen3-vl">MiniMax-H3-encoder-Q5_K_M.gguf</a>(Q5)</h3><h3 id="make-sure-also-download:-minimax-h3-encoder-mmproj-f16.gguf-from-that-page-and-place-it-under-the-same-folder.">Make sure also download: <code>MiniMax-H3-encoder-mmproj-F16.gguf</code> from that page and place it under the same folder.</h3><ul><li><p>place them all under: <code>models\text_encoders\</code></p></li><li><p><code>Choose Q5 for 16GB+ VRAM, Q4 for less</code></p></li></ul><p>Thanks <a target="_blank" rel="ugc" href="https://civarchive.com/user/joeygambino"><strong>joeygambino </strong></a>for his awesome model!</p><hr /><p></p><h3 id="vae">- <span style="color:rgb(250, 176, 5)">VAE</span> -</h3><h3 id="video-vae"><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/vae/minimax_h3_video_vae_fp16.safetensors?download=true">Video VAE</a></h3><h3 id="audio-vae"><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/vae/minimax_h3_audio_vae_fp32.safetensors?download=true">Audio VAE</a></h3><ul><li><p>place them all under: <code>models\vae\</code></p></li></ul><hr /><p></p><h3 id="preview-vae">- <span style="color:rgb(250, 176, 5)">Preview VAE</span> -</h3><h3 id="tine-vae"><a target="_blank" rel="ugc" href="https://huggingface.co/Kijai/MiniMax-H3-TAE/resolve/main/vae_approx/taeh3.safetensors?download=true">Tine VAE</a></h3><ul><li><p>place it under: <code>models\vae_approx\</code></p></li></ul><p>Thanks <a target="_blank" rel="ugc" href="https://huggingface.co/Kijai">Kijai </a>for his awesome model!</p><hr /><p></p><h3 id="upscaler">- <span style="color:rgb(250, 176, 5)">Upscaler</span> -</h3><h3 id="minimax_h3_latent_upscaler_3d_fp16.safetensors"><a target="_blank" rel="ugc" href="https://huggingface.co/LBH-123-AI/Minimax_h3_latent_Upscaler/resolve/main/minimax_h3_latent_upscaler_3d_fp16.safetensors?download=true">minimax_h3_latent_upscaler_3d_fp16.safetensors</a></h3><ul><li><p>place it under: <code>models\latent_upscale_models\</code></p></li></ul><p>Thanks <a target="_blank" rel="ugc" href="https://huggingface.co/LBH-123-AI">LBH-123-AI </a>or his awesome model!<br /></p><hr /><p></p><h2 id="comfyui-custom-nodes">ComfyUi Custom Nodes</h2><h3 id="comfyui-h3-multishot">- <span style="color:rgb(250, 176, 5)">ComfyUI-H3-Multishot</span></h3><p>Used for <code>GGUF</code> base model loader, <code>GGUF</code> text encoder loader, Multi <code>LoRa</code> loader,</p><p>Github install URL: <a target="_blank" rel="ugc" href="https://github.com/jlucasmcrell/ComfyUI-H3-Multishot">https://github.com/jlucasmcrell/ComfyUI-H3-Multishot</a></p><p></p><h3 id="comfyui_minimax_h3_latent_upscaler">- <span style="color:rgb(250, 176, 5)">Comfyui_Minimax_h3_latent_Upscaler</span></h3><p>Used for <code>Latent</code> upscale</p><p>Github install URL: <a target="_blank" rel="ugc" href="https://github.com/LBH-123-AI/Comfyui_Minimax_h3_latent_Upscaler">https://github.com/LBH-123-AI/Comfyui_Minimax_h3_latent_Upscaler</a></p><p></p><p><span style="color:rgb(253, 126, 20)"><strong>Make sure update your ComfyUi to latest version!</strong></span></p><p></p><hr /><h2 id="tips">Tips</h2><p></p><h3 id="how-to-use-last-frame">- How to use Last Frame -</h3><ul><li><p>Connect <code>resized image (0.2mp)</code> to <code>last_frame</code> of <code>MiniMax H3 Image to Video</code></p></li><li><p>Connect <code>resized image (1mp)</code> to <code>last_frame</code> of <code>Upscaler Condition</code></p></li></ul><hr /><p></p><h3 id="how-to-switch-to-t2va">- How to switch to [t2va] -</h3><ul><li><p>Bypass the following nodes: <code>First Frame</code>, <code>Resized Image (0.2mp)</code>, <code>Resized Image (1mp)</code></p></li></ul><hr /><p></p><h3 id="balance-between-motion-and-surface-details">- Balance between motion and surface details -</h3><ul><li><p>Lower the <code>Shift_video</code> value of <code>ModelSamplingMiniMaxH3</code> node will make the model focus more on physics, increase it will make it focus more on surface details.</p></li></ul><hr /><p></p><p></p>

Files with SHA256: 6c481be0f28a399832c59ffc89c87b4062ada5748b04e7a81bdedd58994beb0e