<h1 id="minimax-h3-refmod-prism-(singularity-10eros)-continue-+-outfit-lock">MiniMax H3 RefMod · PRISM (Singularity × 10Eros) — Continue + Outfit Lock</h1><hr /><h2 id="description">Description</h2>Ref2VA workflow for MiniMax H3 with RefMod face files, running the community PRISM merge (Singularity v2 × 10Eros turbo baked in — no extra turbo LoRA). Built for clip continuation: last frame → AddGuide at frame 0, plus an optional separate outfit/bikini reference (<code>ref_image_0</code> / <code>&lt;Picture 1&gt;</code>) so wardrobe does not drift when the continue frame is far/back-facing. Manual prompt (no LLM). Auto last-frame PNG next to the saved MP4. Optional SageAttention.RefMod setup: use <a target="_blank" rel="ugc" href="https://civarchive.com/models/3002052/minimax-h3-refmod-the-same-face-in-every-clip-no-lora">StableYogi — MiniMax H3 RefMod (same face, no LoRA)</a> Part 1 to create face mods. Put required face / first-frame / outfit images in <code>ComfyUI/input/</code> before loading them in this graph.<hr /><h2 id="features">Features</h2><ul><li>PRISM Ref2VA UNET — Singularity + 10Eros turbo in one checkpoint (keep 4-step / 8-step turbo LoRA nodes bypassed)</li><li>RefMod — load face/body <code>.safetensors</code> mods (Part 1 files) → Apply into conditioning</li><li>Continue clips — <code>FIRST FRAME</code> → MiniMaxH3AddGuide only (pose/continuity)</li><li>Outfit lock — separate <code>OUTFIT / BIKINI REF</code> → <code>ref_image_0</code> = <code>&lt;Picture 1&gt;</code> (clear front/wardrobe), independent from the last frame</li><li>Manual prompt — duration/width/height meta + your prompt → ReferenceToVideo</li><li>Resolution + duration — ResolutionSelector + seconds → frames at 24fps (<code>17n+5</code>)</li><li>Last frame PNG — <code>VHS_SelectImages (-1)</code> + SaveImage (same date stem as the video)</li><li>Optional — Power LoRA rack (off by default), Ref Audio (bypassed), SageAttention (KJ)</li></ul><hr /><h2 id="how-to-use">How to use</h2><ol><li>Install custom nodes (below) and restart ComfyUI.</li><li>Download models into the folders listed under Models.</li><li>Open <code>H3RefMod-PRISM-comfyui.json</code>.</li><li>RefMod faces (StableYogi Part 1) — follow <a target="_blank" rel="ugc" href="https://civarchive.com/models/3002052/minimax-h3-refmod-the-same-face-in-every-clip-no-lora">StableYogi — MiniMax H3 RefMod (same face, no LoRA)</a> to create the face <code>.safetensors</code>. Place those files where the loader finds them, and put any source face / reference images the Part‑1 workflow asks for into <code>ComfyUI/input/</code>. Then in this graph’s Face files, pick <code>mod_1</code> / <code>mod_2</code>… (press R if the list is empty).</li><li>FIRST FRAME — last frame of the previous clip into <code>ComfyUI/input/</code> (continuity → AddGuide only).</li><li>OUTFIT / BIKINI REF — clear front photo of the same outfit/face also in <code>ComfyUI/input/</code> → <code>ref_image_0</code>. In the prompt, lock wardrobe with <code>&lt;Picture 1&gt;</code>.</li><li>Write the prompt in Prompt Manual (<code>subject_definitions</code> + scene; use <code>&lt;Subject 1&gt;</code> matching RefMod names).</li><li>Set duration (seconds) and Resolution Selector.</li><li>Queue. Do not enable the turbo LoRA bypass nodes (turbo is already in PRISM).</li></ol><h3 id="prompt-tip-(continue-+-outfit)">Prompt tip (continue + outfit)</h3><ul><li>Opening state = first_frame / AddGuide (do not call the last frame <code>&lt;Picture 1&gt;</code>).</li><li><code>&lt;Picture 1&gt;</code> = outfit/bikini/face lock from the OUTFIT REF node only.</li></ul><hr /><h2 id="defaults-(this-graph)">Defaults (this graph)</h2>SettingValueSteps8Scheduler<code>simple</code>Sampler<code>euler</code>Sigma shift12 / 3FPS24RefModstrength 1.0 (adjust copies as needed)Turbo LoRAsBYPASSSageAttentionON (optional speed)<hr /><h2 id="custom-nodes-(dependencies)">Custom nodes (dependencies)</h2>Install via ComfyUI Manager or <code>git clone</code> into <code>ComfyUI/custom_nodes/</code>.PackRequiredNodes usedLinksComfyUI (native MiniMax H3)Yes<code>MiniMaxH3ReferenceToVideo</code>, <code>MiniMaxH3AddGuide</code>, <code>MiniMaxH3SigmaShift</code>, VAE/CLIP/UNET loaders, CreateVideo, SaveVideo, ResolutionSelector, …<a target="_blank" rel="ugc" href="https://github.com/comfyanonymous/ComfyUI">ComfyUI</a> · <a target="_blank" rel="ugc" href="https://docs.comfy.org/">MiniMax H3 docs</a>ComfyUI-MiniMaxH3ModYes<code>MiniMaxH3RefModsLoader</code>, <code>MiniMaxH3RefModApply</code><a target="_blank" rel="ugc" href="https://github.com/Luisacaotica/ComfyUI-MiniMaxH3Mod">GitHub</a> · <a target="_blank" rel="ugc" href="https://comfy.icu/extension/Luisacaotica__ComfyUI-MiniMaxH3Mod">comfy.icu</a>ComfyUI-VideoHelperSuiteYes<code>VHS_SelectImages</code>, <code>VHS_LoadAudioUpload</code> (optional audio)<a target="_blank" rel="ugc" href="https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite">GitHub</a>ComfyUI-KJNodesRecommended<code>PathchSageAttentionKJ</code>, <code>SomethingToString</code><a target="_blank" rel="ugc" href="https://github.com/kijai/ComfyUI-KJNodes">GitHub</a>rgthree-comfyOptional<code>Power Lora Loader (rgthree)</code><a target="_blank" rel="ugc" href="https://github.com/rgthree/rgthree-comfy">GitHub</a>ComfyUI-Custom-ScriptsOptional<code>PlaySound|pysssss</code><a target="_blank" rel="ugc" href="https://github.com/pythongosssss/ComfyUI-Custom-Scripts">GitHub</a><blockquote>Need a current ComfyUI build with native MiniMax H3 support. RefMod does not replace the H3 core nodes.</blockquote><h3 id="manager-search-keywords">Manager search keywords</h3><code>MiniMaxH3Mod</code> · <code>Video Helper Suite</code> · <code>KJNodes</code> · <code>rgthree</code> · <code>Custom Scripts</code> / <code>pythongosssss</code><hr /><h2 id="models">Models</h2><h3 id="required">Required</h3>RoleSuggested path under <code>ComfyUI/models/</code>SourceUNET / diffusion (PRISM)<code>diffusion_models/MiniMax-H3/singularityV2X10eros_prism.safetensors</code> (name as in this workflow — rename or re-pick after download)<a target="_blank" rel="ugc" href="https://civarchive.com/models/3019421/singularity-v2-x-10eros-max">https://civarchive.com/models/3019421/singularity-v2-x-10eros-max</a>Text encoder (CLIP type <code>minimax</code>)<code>text_encoders/</code> e.g. <code>qwen3vl_32b_heretic_minimax_h3_nvfp4.safetensors</code> (or official H3 Qwen3-VL)<a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3">Comfy-Org/MiniMax-H3</a>Video VAE<code>vae/minimax_h3_video_vae_fp16.safetensors</code><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3">Comfy-Org/MiniMax-H3</a>Audio VAE<code>vae/minimax_h3_audio_vae_fp32.safetensors</code><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3">Comfy-Org/MiniMax-H3</a>RefMod face filesPut the <code>.safetensors</code> mods where the RefModsLoader expects them (see StableYogi pack / your Part‑1 export folder), then pick them in Face filesCreate with <a target="_blank" rel="ugc" href="https://github.com/Luisacaotica/ComfyUI-MiniMaxH3Mod">ComfyUI-MiniMaxH3Mod</a> · guide + pack: <a target="_blank" rel="ugc" href="https://civarchive.com/models/3002052/minimax-h3-refmod-the-same-face-in-every-clip-no-lora">StableYogi — MiniMax H3 RefMod (Civitai)</a><h3 id="optional-(in-graph-bypassed-by-default)">Optional (in graph, bypassed by default)</h3>RoleExampleNotes4-step turbo LoRA<code>loras/Minimax-h3/minimax_h3_ref2v_lightx2v_turbo_4step_…</code>Keep bypassed on PRISM8-step turbo LoRA<code>loras/Minimax-h3/minimax_h3_ref2v_turbo_8step_…</code>Keep bypassed on PRISMPower LoRA rackvarious NSFW/motion LoRAsAll off unless neededRef audioany short <code>.wav</code>/<code>.mp3</code>/<code>.m4a</code>Node bypassedDo not stack LightX2V / Comfy turbo LoRA on top of the PRISM UNET.<h3 id="related-base-models-(if-you-ab)">Related base models (if you A/B)</h3><ul><li>Stock Ref2VA: <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3">Comfy-Org/MiniMax-H3</a></li><li>Singularity: <a target="_blank" rel="ugc" href="https://huggingface.co/WarmBloodAban/Minimax-h3_Singularity">WarmBloodAban/Minimax-h3_Singularity</a></li></ul><hr /><h2 id="node-map-(this-workflow)">Node map (this workflow)</h2>StepNodePurpose1UNETLoaderPRISM Ref2VA2CLIPLoaderMiniMax text encoder3–4VAELoader ×2Video + audio VAE—MiniMaxH3SigmaShiftMotion dial (12/3)—Power Lora LoaderOptional LoRAs6MiniMaxH3ReferenceToVideoPrompt, size, length, refs7MiniMaxH3RefModsLoaderFace files8MiniMaxH3RefModApplyInject mods into cond—MiniMaxH3AddGuideFirst-frame lock @ t=0—LoadImage (FIRST FRAME)Continue last frame → AddGuide only—LoadImage (OUTFIT REF)Clear wardrobe → <code>ref_image_0</code> / <code>&lt;Picture 1&gt;</code>—SamplerCustomAdvanced8-step euler—VAEDecode + VAEDecodeAudio → CreateVideo → SaveVideoOutput MP4—VHS_SelectImages (−1) → SaveImageLast-frame PNG<hr /><h2 id="outputs">Outputs</h2><ul><li>Video: <code>output/video/H3RefMod-PRISM/H3RefMod-PRISM-&lt;date&gt;_&lt;time&gt;_….mp4</code></li><li>Last frame: same stem <code>.png</code> (copy to <code>input/</code> for the next continue)</li></ul><hr /><h2 id="credits">Credits</h2><ul><li>MiniMax H3 — MiniMax / <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3">Comfy-Org</a></li><li>PRISM merge — <a target="_blank" rel="ugc" href="https://civarchive.com/models/3019421/singularity-v2-x-10eros-max">https://civarchive.com/models/3019421/singularity-v2-x-10eros-max</a></li><li>Singularity — <a target="_blank" rel="ugc" href="https://huggingface.co/WarmBloodAban/Minimax-h3_Singularity">WarmBloodAban</a></li><li>RefMod nodes — <a target="_blank" rel="ugc" href="https://github.com/Luisacaotica/ComfyUI-MiniMaxH3Mod">Luisacaotica/ComfyUI-MiniMaxH3Mod</a></li><li>Graph inspired by StableYogi-style RefMod Part 2 continue layouts</li></ul><hr /><h2 id="license-nsfw">License / NSFW</h2>Follow the licenses of MiniMax H3, PRISM/Singularity/10Eros, and any LoRAs you enable. This graph can generate NSFW content; use responsibly and only with adult subjects.

Files with SHA256: 61f3735165db5c7fe92f5aea60c71568f01f03a0fc936fe8563742efeb7c8150