Correct MiniMax LongVideos baseline graph

This commit is contained in:
Morpheus
2026-09-28 14:48:57 +00:00
parent c935becc3d
commit 5dd3e330b1
4 changed files with 46 additions and 29 deletions
+9 -3
View File
@@ -7,8 +7,8 @@ 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
- Prompt: two-shot bright park scene: robins in shot one, then a squirrel
joins in shot two
- Resolution target: 960×544 landscape (`0.55` megapixels, 16:9)
- Duration: 10 seconds
- Steps: 8
@@ -17,6 +17,11 @@ paragraph-driven shot plan; it does not load or encode a previous video.
- Seed: 20260927
- Shift video/audio: 12 / 3
- Reference noise augmentation: 0.999
- SLA Attention: sparsity 0.85, block 32
- Model Attention Backend: Comfy Kitchen attention
- Low-VRAM Attention: 4 head chunks
- Chunk FeedForward: 2 chunks, sequence threshold 4096
- Turbo LoRA: enabled at strength 1.0
- Latent and post upscaling: disabled for this baseline
## Files
@@ -35,4 +40,5 @@ 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.
graph now retains the proven memory/attention chain and explicitly wires the
scene anchor and character memory fields.