ComfyUI Node
⭐ Star Minimax All In One Advanced
The complete Star Minimax All In One pipeline with three extra optional connectors for externally built inputs (e.g. the BFSNodes shot nodes or any CONDITIONING / LATENT source). Connect positive to sample connected conditioning instead of the internal prompt + reference encode - the text encoder is then not loaded at all and prompt, clip and ref inputs are ignored. Connect negative to sample with a real CFGGuider at the cfg value (at cfg 1.0 the negative is ignored by the sampler anyway). Connect latent to sample a connected latent - aspect ratio, megapixels and duration are then ignored and canvas plus frame count are read back from the latent. Any combination works; RefMod, multiref guides, sigmas, the upscale pass, sound enricher and the model/VAE overrides behave exactly as on the base node.
⭐ Star Minimax All In One Advanced
- model_override
- vae_override
- positive
- negative
- latent
- sound_settings
- sigmas
- options
- ref_mod_settings
- multiref_settings
- preview
- ref_images
- ref_videos
- ref_video_audios
- ref_audios
- IMAGE
- AUDIO
- FPS
- LATENT
- MODEL
- CLIP
- VAE
- AUDIO_VAE
◄modevideo►
◄prompt—►
◄aspect_ratio16:9 (Widescreen)►
◄megapixels0.5►
◄match_ratio_from_imagefalse►
◄duration5.0►
◄ref_image_sizematch►
◄seed0►
◄steps20►
◄sampler_nameres_multistep►
◄schedulersimple►
◄denoise1.00►
◄cfg1.0►
◄diffusion_modelminimax_h3_ref2va_pruned_int8_convrot.safetensors►
◄weight_dtypedefault►
◄clip_nameqwen3vl_32b_minimax_h3_nvfp4_awq.safetensors►
◄clip_typeminimax►
◄clip_devicedefault►
◄vae_nameminimax_h3_video_vae_fp16.safetensors►
◄audio_vae_nameminimax_h3_audio_vae_fp32.safetensors►
◄audio_vae_precisionfp32►
◄audio_vae_devicemain_device►
◄decode_videotrue►
Category⭐StarNodes/Video
Inputs (38)
| Name | Type | Default | Description |
|---|---|---|---|
| mode | COMBO | video | 'video' renders the full clip with audio. 'image' renders one single frame (H3's native still convention) at the selected ratio and size and decodes it with the single-frame still decode; duration and audio decoding are skipped. Reference inputs work exactly like in video mode. |
| prompt | STRING | Reference inputs by tag in connection order, e.g. <Picture 1>, <Video 1>, <Audio 1>, then describe scene, motion and audio. | |
| aspect_ratio | COMBO | 16:9 (Widescreen) | Aspect ratio for the output dimensions. |
| megapixels | COMBO | 0.5 | Target total megapixels (output pixel size). 0.5 MP ~ 960x544 at 16:9; 2.0 MP ~ 1920x1088. Select "audio only" for a fixed 32x32 canvas when you only need audio output. The 3-8 MP presets appear in image mode only (single-frame stills hit their quality sweet spot from 3 MP up). |
| match_ratio_from_image | BOOLEAN | false | If enabled and a reference image is connected, the closest aspect ratio of the first reference image is used at the selected pixel size. |
| duration | FLOAT | 5.00.2–150 | Video duration in seconds at 24 fps. Internally snapped to the model's 17k+5 frame grid (5s -> 124 frames). |
| ref_image_size | COMBO | match | Reference image sizing. 'match' scales each ref (down only, keeping aspect) to the generation's pixel area; 'max' uses a 2048px short edge for best identity fidelity but is much slower. |
| seed | INT | 00–18446744073709550000 | — |
| steps | INT | 201–10000 | — |
| sampler_name | COMBO | res_multistep | 44 options: euler, euler_cfg_pp, euler_ancestral, euler_ancestral_cfg_pp, heun, heunpp2, +38 |
| scheduler | COMBO | simple | 'beta' or 'normal' tends to outperform 'simple' for reference-heavy prompts. |
| denoise | FLOAT | 1.000–1 | — |
| cfg | FLOAT | 1.00–100 | CFG scale for the guider. Only used when a negative conditioning is connected; at 1.0 the negative is ignored by the sampler anyway. |
| diffusion_model | COMBO | minimax_h3_ref2va_pruned_int8_convrot.safetensors | Ignored when a MODEL is connected to the model_override input. |
| weight_dtype | COMBO | default | 4 options: default, fp8_e4m3fn, fp8_e4m3fn_fast, fp8_e5m2 |
| clip_name | COMBO | qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors | 0 options: |
| clip_type | COMBO | minimax | 28 options: stable_diffusion, stable_cascade, sd3, stable_audio, mochi, ltxv, +22 |
| clip_device | COMBO | default | 2 options: default, cpu |
| vae_name | COMBO | minimax_h3_video_vae_fp16.safetensors | Video VAE (e.g. minimax_h3_video_vae_fp16). |
| audio_vae_name | COMBO | minimax_h3_audio_vae_fp32.safetensors | Audio VAE (e.g. minimax_h3_audio_vae_fp32). |
| audio_vae_precision | COMBO | fp32 | Precision the audio VAE runs at. fp32 is recommended (same as the KJ loader preset). |
| audio_vae_device | COMBO | main_device | 2 options: main_device, cpu |
| decode_video | BOOLEAN | true | Decode the generated video (or still) with the video VAE. Switch off to skip video decoding entirely - the node then only outputs the LATENT and the AUDIO, e.g. to run an external decode or keep memory free for a follow-up pass. |
| model_overrideopt | MODEL | Optional external MODEL (e.g. a sage-attention patched MiniMax H3). When connected, the internal diffusion_model dropdown is ignored. | |
| vae_overrideopt | VAE | Optional external VAE for video decoding. When connected, the internal vae_name dropdown is ignored. | |
| positiveopt | CONDITIONING | Optional external positive conditioning (e.g. the 'positive' output of a BFSNodes shot conditioning node). When connected, the internal prompt + reference encode is skipped - the clip is not loaded and prompt, clip and ref inputs are ignored. Connect the matching latent too if the conditioning was built for a specific canvas/duration. | |
| negativeopt | CONDITIONING | Optional external negative conditioning. When connected, sampling uses a real CFGGuider at the cfg widget value instead of the positive-only basic guider. | |
| latentopt | LATENT | Optional external latent (e.g. the 'latent' output of a BFSNodes shot conditioning node). When connected, aspect ratio, megapixels and duration are ignored and canvas + frame count are read back from the latent. | |
| sound_settingsopt | SOUND_SETTINGS | Optional sound processing from a 'Star Video Sound Enricher Option' node - the generated soundtrack is cleaned up and enriched with these settings (at least 44.1 kHz, never downsampled) before it leaves the node. Ignored in image mode without audio. | |
| sigmasopt | SIGMAS | Optional exact sigma schedule for the main sampling pass (e.g. from a '⭐ Star Custom Sigmas' node or any SIGMAS output) - overrides the scheduler, steps and denoise widgets. | |
| optionsopt | UPSCALE_SETTINGS | Optional second-pass latent upscale from a 'Star Minimax Latent Upscaler Option' node - the pass-1 video latent is upscaled with the selected latent upscaler model and refined in a short second sampling pass with the same conditioning and the same seed (references are resolution-matched). Ignored when megapixels is 'audio only'. | |
| ref_mod_settingsopt | REF_MOD_SETTINGS | Optional RefMod injection from a 'Star Ref Mod Option' node - appends Apply-H3-RefMod-style reference blocks (ComfyUI-MiniMaxH3Mod pack) to the internal conditioning, after the native reference blocks. Also carried into the upscale refine pass (resolution-matched). | |
| multiref_settingsopt | MULTIREF_SETTINGS | Optional timed keyframe references from a 'Star Minimax Multiref Option' node - reference images/clips anchored at their start seconds on the output timeline, exactly like a chain of core 'Add Guide for MiniMax H3' nodes. Works in video and image mode; the guides are also carried into the upscale refine pass (resolution-matched). | |
| previewopt | STAR_PREVIEW | Optional live sampling preview from a '⭐ Star Preview' node - while this node is sampling, an animated preview of the video latent is shown on the Star Preview node (fixed: 512 px, quality 80, 8 fps). | |
| ref_imagesopt | COMFY_AUTOGROW_V3 | — | |
| ref_videosopt | COMFY_AUTOGROW_V3 | — | |
| ref_video_audiosopt | COMFY_AUTOGROW_V3 | — | |
| ref_audiosopt | COMFY_AUTOGROW_V3 | — |
Outputs (8)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | — |
| AUDIO | AUDIO | — |
| FPS | FLOAT | Fixed frame rate of the generated video (24.0). Connect straight into your video combine/save node. |
| LATENT | LATENT | The combined processed latent from the sampler (NestedTensor with video+audio), before VAE decoding. When an upscale options node is connected, this is the refined second-pass latent. |
| MODEL | MODEL | The diffusion model used for the main (pass-1) sampling. |
| CLIP | CLIP | The loaded text encoder. |
| VAE | VAE | The video VAE. |
| AUDIO_VAE | VAE | The audio VAE (None in image mode without audio references). |