<h1 id="minimax-h3-longforge-v2.0-fl2va-and-ref2va">MiniMax H3 LongForge v2.0 — FL2VA &amp; REF2VA</h1><h3 id="support-the-project">Support the project</h3>If LongForge helps you, you can support future development through the Tip button on my Civitai profile or an optional donation:USDT · TRON (TRC20) <code>THLU3yu8VpoPqP4bMLCp86FRhoCweJ2Qz4</code>Please send only USDT through the TRON (TRC20) network. Thank you!<hr />One workflow to generate, continue, review and export long videos with audio in ComfyUI.LongForge transfers saved video and audio latent context between scenes to help preserve motion, appearance, camera movement and ambience. Everything stays in one editable graph with visible loaders and manual controls.Version 2.0 brings a redesigned scene editor, a built-in reference library, persistent scene previews and automatic slicing of long audio references.<h3 id="what's-new-in-v2.0">What’s new in v2.0</h3><ul><li>Redesigned Scene Studio: a searchable scene list, clear scene statuses, and an integrated prompt editor and video player.</li><li>Persistent scene previews: return to earlier fragments after generating additional scenes.</li><li>Built-in Reference Library: add multiple images, videos and audio files through drag-and-drop or + Add files. Preview, reorder, trim and enable individual sources.</li><li>Follow scenes audio mode: automatically supply the correct part of a long audio reference to each scene, accounting for overlap, continuation and regeneration.</li><li>Reference audio switch: disable audio references without muting the generated film.</li><li>Resolution and duration selectors: convenient choices with Custom controls for manual values.</li><li>Native model and LoRA loaders: two optional LoRAs, with no forced Turbo mode or fixed step count.</li><li>Clearer render feedback: generation progress and processing information in Render &amp; Export.</li><li>Saving and interface fixes: improved project saving and corrected settings shifting after a page reload.</li></ul><h3 id="two-generation-modes">Two generation modes</h3><ul><li>FL2VA — text-to-video, image-to-video and optional first/last-frame guidance.</li><li>REF2VA — up to 9 images, 3 videos with paired soundtracks, and 3 standalone audio references.</li></ul>Switch modes inside the same workflow and select the matching H3 model.Resolution, frames, overlap, steps, sampler, scheduler, CFG and seed remain manually adjustable. There is no 1 MP resolution cap; practical limits depend on your hardware and model.<h3 id="long-audio-references">Long audio references</h3>In Reference Library, add an audio file and choose:Audio use → Follow scenes · REF2VALongForge advances through the source automatically, supplying each scene with its corresponding audio window. There is no need to reconnect files or manually seek between generations.Start, End and Extra offset · frames control the source range. Follow scenes works throughout the film even when fixed image/video references apply only to the first scene.This is reference guidance for H3, not a replacement soundtrack. The model generates its own audio and may change the source’s voice, wording, music or rhythm.<h3 id="download-and-installation">Download &amp; installation</h3>Download both archives:<ul><li>ComfyUI-H3-LongForge-NodePack.zip — the custom nodes and installation README.</li><li>H3_LongForge_FL2VA_REF2VA_PROMPT_GUIDE.zip — the unified workflow and one complete Russian/English user and prompt guide.</li></ul><ol><li>Close ComfyUI and extract the node-pack folder into <code>ComfyUI/custom_nodes/</code>. When updating, replace the previous LongForge installation without merging versions. Keep your saved projects in <code>ComfyUI/output/</code>.</li><li>Follow the README to install dependencies. The supplied graph uses native MiniMax H3 support, ComfyUI-KJNodes and compatible SageAttention.</li><li>Restart ComfyUI, refresh the browser with Ctrl+F5, and open <code>H3_LongForge_FL2VA_REF2VA.json</code>.</li><li>Select your diffusion model, text encoder and both VAEs.</li><li>Both native LoRA loaders start bypassed. Select a LoRA and press Ctrl+B on its node to enable it when needed. Optional image loaders also start bypassed.</li></ol>Models: <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3/tree/main">Comfy-Org / MiniMax-H3</a> Text encoders: <a target="_blank" rel="ugc" href="https://huggingface.co/ethanfel/Qwen3-VL-32B-Ultra-Heretic-H3-ComfyUI-INT8-ConvRot/tree/main">INT8 ConvRot</a> · <a target="_blank" rel="ugc" href="https://huggingface.co/sakamakismile/Qwen3-VL-32B-Heretic-MiniMax-H3-NVFP4/tree/main">NVFP4</a> Attention patch nodes: <a target="_blank" rel="ugc" href="https://github.com/kijai/ComfyUI-KJNodes">ComfyUI-KJNodes</a>Reference media loading is included in LongForge; no separate media-loader node pack is required.<h3 id="generate-your-film">Generate your film</h3>Write one prompt per scene, using + Scene to add the next:<pre><code>VIDEO: Describe the subject, action, setting and camera.
SOUND: Describe ambience, effects and dialogue.
MUSIC: No music.</code></pre>These headings are optional. In continuation prompts, describe the next stage of the action while preserving the scene’s established appearance and movement.For REF2VA, use native markers such as <code>&lt;Picture 1&gt;</code>, or define optional names such as <code>hero=image1</code> and use <code>{hero}</code> in the prompt.<ul><li>GENERATE NEXT + ONE SCENE — generate one next scene per Run.</li><li>GENERATE NEXT + ALL PENDING — generate all remaining scenes in order.</li><li>Preview scene — review a selected saved fragment.</li><li>REGENERATE SELECTED — replace from the selected scene and rebuild its continuation.</li><li>EXPORT FILM — assemble the active saved scenes into one MP4 with generated audio.</li></ul>Each successful scene is saved automatically. Reopen your project to continue later.<h3 id="optional-post-processing">Optional post-processing</h3>For additional enhancement, upscaling or frame interpolation, you can process the exported film with <a target="_blank" rel="ugc" href="https://github.com/Merserk/dlss5-visual-enhancer">DLSS 5 Visual Enhancer</a>. It is a separate application and is not required by LongForge.<a target="_blank" rel="ugc" href="https://github.com/Merserk/dlss5-visual-enhancer/releases">Download DLSS 5 Visual Enhancer</a><h3 id="my-telegram-channel">My Telegram channel</h3><a target="_blank" rel="ugc" href="https://t.me/aibobpublic">AI Bob Public</a>===========================================================<h1 id="minimax-h3-longforge-v1.0-fl2va-and-ref2va">MiniMax H3 LongForge v1.0 — FL2VA &amp; REF2VA</h1><h3 id="support-the-project">Support the project</h3>If LongForge helps you, you can support future development through the Tip button on my Civitai profile or an optional donation:USDT · TRON (TRC20)<code>THLU3yu8VpoPqP4bMLCp86FRhoCweJ2Qz4</code>Please send only USDT through the TRON (TRC20) network. Thank you!One workflow for generating, continuing and exporting long videos with synchronized audio in ComfyUI.LongForge carries saved video and audio latent context between scenes to help preserve motion, appearance, camera movement and ambience. Everything stays in one editable graph, with visible model loaders and manual controls.<h3 id="two-generation-modes">Two generation modes</h3><ul><li>FL2VA — text-to-video, image-to-video, and optional first/last-frame guidance.</li><li>REF2VA — image, video and audio references: up to 9 images, 3 videos with paired soundtracks, and 3 standalone audio references.</li></ul>Switch modes inside the same workflow and select the appropriate H3 model.<h3 id="main-features">Main features</h3><ul><li>One scene editor: add prompts with + Scene.</li><li>Flexible generation: process one scene or all pending scenes.</li><li>Automatic continuation: each new scene starts from the saved film end.</li><li>Saved projects: reopen and continue your film later.</li><li>Scene regeneration: replace a selected scene and rebuild its continuation.</li><li>Manual settings: resolution, frames, overlap, steps, sampler, scheduler, CFG and seed.</li><li>Two optional LoRA loaders: use either, both, or leave them on <code>NONE</code>.</li><li>Visible attention patches: KJ Sage Attention and MiniMax H3 Memory Efficient Sage Attention.</li><li>Fragment previews and full-film export in the same graph.</li></ul><h3 id="download-and-installation">Download &amp; installation</h3>Download both archives:<ul><li>ComfyUI-H3-LongForge-NodePack.zip — required custom nodes and installation README.</li><li>H3_LongForge_FL2VA_REF2VA_PROMPT_GUIDE.zip — the workflow and one Russian/English prompt guide.</li></ul><ol><li>Extract the node-pack folder into <code>ComfyUI/custom_nodes/</code>.</li><li>Follow its README to install dependencies. The supplied graph requires native MiniMax H3 support, ComfyUI-KJNodes and compatible SageAttention.</li><li>Restart ComfyUI and open <code>H3_LongForge_FL2VA_REF2VA.json</code>.</li><li>Select your diffusion model, text encoder, video VAE and audio VAE. Choose optional LoRAs and enable any image loaders you need.</li></ol>Models DiT: <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3/tree/main">Comfy-Org / MiniMax-H3</a>Text Encoders: <a target="_blank" rel="ugc" href="https://huggingface.co/ethanfel/Qwen3-VL-32B-Ultra-Heretic-H3-ComfyUI-INT8-ConvRot/tree/main">INT8 ConvRot </a>or <a target="_blank" rel="ugc" href="https://huggingface.co/sakamakismile/Qwen3-VL-32B-Heretic-MiniMax-H3-NVFP4/tree/main">NVFP4</a>Other Custom Nodes: <a target="_blank" rel="ugc" href="https://github.com/kijai/ComfyUI-KJNodes">ComfyUI-KJNodes</a> — required only for SAGE Attention and MiniMax H3 Memory Efficient Sage Attention.<h3 id="generate-your-film">Generate your film</h3>Write one prompt per scene. A simple structure works well:<pre><code>VIDEO: Describe the subject, action, setting and camera.
SOUND: Describe ambience, effects and dialogue.
MUSIC: No music.</code></pre>These headings are optional. The included guide explains both modes, continuation prompts and reference numbering.In REF2VA, use native markers such as <code>&lt;Picture 1&gt;</code>, or optional names such as <code>{hero}</code>. References can apply to the first scene or every scene.Choose GENERATE NEXT and press Run to generate. When ready, choose EXPORT FILM and press Run to assemble the saved scenes into one MP4 with audio.<h2 id="optional-final-post-processing">Optional final post-processing</h2>If additional upscaling or frame interpolation is required, process the completed video afterward with:<a target="_blank" rel="ugc" href="https://github.com/Merserk/dlss5-visual-enhancer">DLSS 5 Visual Enhancer</a>It supports RTX Video Super Resolution, DLSS Neural Rendering, and optional frame generation. It is a separate application and is not required for these ComfyUI workflows.<a target="_blank" rel="ugc" href="https://github.com/Merserk/dlss5-visual-enhancer/releases">Download DLSS 5 Visual Enhancer</a>My Telegram channel:<a target="_blank" rel="ugc" href="https://t.me/aibobpublic">https://t.me/aibobpublic</a>

Files with SHA256: a1e5ef79dc4e848a16cfee141b96651e872b9456992caf6f45617ad918105597