Add MiniMax H3 LongVideos candidate

This commit is contained in:
Morpheus
2026-09-28 14:37:51 +00:00
parent 93c444be26
commit 0064d7d6be
2 changed files with 48 additions and 0 deletions
@@ -0,0 +1,11 @@
{
"1": {"inputs": {"unet_name": "minimax_h3_hybrid_fl2va_ref2va_b25-49-int8.safetensors", "weight_dtype": "default"}, "class_type": "UNETLoader"},
"2": {"inputs": {"clip_name": "qwen3vl_32b_h3_ultra_uncensored_heretic_int8_convrot.safetensors", "type": "minimax"}, "class_type": "CLIPLoaderGGUF"},
"3": {"inputs": {"vae_name": "minimax_h3_video_vae_int8_convrot.safetensors"}, "class_type": "VAELoader"},
"4": {"inputs": {"vae_name": "MiniMax-H3-audio_vae.safetensors"}, "class_type": "VAELoader"},
"5": {"inputs": {"image": "reference.png"}, "class_type": "LoadImage"},
"6": {"inputs": {"value": "A bright sunny public park in warm natural daylight, with green grass, mature trees, a wooden park bench, gentle shadows, and a calm friendly atmosphere.\n\nAn adult male with the reference character's identity, beard, white cap, white clothing, tattoos, layered chains, jewelry, and proportions sits comfortably on the bench and feeds small pieces of bread to red-breasted robins. Show natural hand movement, realistic bird flight and landing, stable identity, coherent sunlight, natural camera movement, and clean realistic detail. Preserve the reference character without redesign or extra people."}, "class_type": "PrimitiveStringMultiline"},
"7": {"inputs": {"model": ["1", 0], "clip": ["2", 0], "vae": ["3", 0], "audio_vae": ["4", 0], "prompt": ["6", 0], "resolution": "16:9", "megapixels": 0.55, "shot_seconds": 10.0, "steps": 8, "sampler_name": "res_multistep", "scheduler": "simple", "seed": 20260927, "ref_image_1": ["5", 0], "shift_video": 12.0, "shift_audio": 3.0, "ref_noise_aug": 0.999, "latent_upscale": "off", "latent_upscale_scale": 2.0, "upscale": "off", "upscale_model": "none", "upscale_target_short_edge": 0, "shot_length": "from the beat", "auto_remove": true, "restart_after_removal": true, "hold_restraints": true, "plan_only": false, "anchor": "", "character_memory": "", "pace": 1.0, "ambient_level": 0.25, "foley_level": 0.35, "speech_lead_seconds": 0.5, "speech_tail_seconds": 2.0, "hold_levels": 0.8}, "class_type": "H3LongVideos"},
"8": {"inputs": {"images": ["7", 0], "audio": ["7", 1], "fps": 24.0, "bit_depth": 8, "color_space": "sRGB"}, "class_type": "CreateVideo"},
"9": {"inputs": {"video": ["8", 0], "filename_prefix": "video/MiniMax_h3_longvideos_baseline", "format": "auto", "codec": "auto"}, "class_type": "SaveVideo"}
}
+37
View File
@@ -0,0 +1,37 @@
# MiniMax H3 LongVideos candidate
This is the next R2V candidate after rejecting connected-chunk continuation.
It generates the clip directly from the character reference and a
paragraph-driven shot plan; it does not load or encode a previous video.
## Fixed baseline
- Reference: `reference.png`
- Prompt: `prompt.txt` equivalent bright park scene, expressed as a scene
paragraph plus one action paragraph in the API graph
- Resolution target: 960×544 landscape (`0.55` megapixels, 16:9)
- Duration: 10 seconds
- Steps: 8
- Sampler: `res_multistep`
- Scheduler: `simple`
- Seed: 20260927
- Shift video/audio: 12 / 3
- Reference noise augmentation: 0.999
- Latent and post upscaling: disabled for this baseline
## Files
- `h3-longvideos-api.json` — direct API workflow
- `reference.png` — fixed character reference
- `prompt.txt` — shared reference-to-video prompt source
## Why this candidate
The installed `H3LongVideos` node supports paragraph-driven shot chaining,
synchronized audio, identity references, and later latent/RTX upscale paths.
It is the appropriate next workflow for multi-shot generation without loading
the previous clip into VRAM.
The imported source workflow remains at
`models/minimax/minimax-h3-longvideos-smite79/workflow.json`. This optimized
API graph is intentionally reduced to the fixed reference-video baseline.