Download storyflow/three-frame-in-context/README.md from Coffmeteor/Krea2-Model-Library: direct link, hf CLI and curl.
- Browser
- Download file 1.72 kB
-
https://huggingface.co/Coffmeteor/Krea2-Model-Library/resolve/main/storyflow/three-frame-in-context/README.md
- Command line
-
hf download hf://Coffmeteor/Krea2-Model-Library/storyflow/three-frame-in-context/README.md
-
curl -L -o README.md https://huggingface.co/Coffmeteor/Krea2-Model-Library/resolve/main/storyflow/three-frame-in-context/README.md
Storyflow Three-Frame In-Context
Published by Coffmeteor for Krea 2.
Original Civitai release · Version: v0.1
The files below preserve the exact published Civitai bytes. SHA-256 checksums are included.
Downloads
| File | Type |
|---|---|
| krea2-meridian-storyboard-v003-final.safetensors | Model |
Usage
The Civitai trigger field is empty; follow the published usage notes below.
Verified default adapter strength: 0.85.
Experimental Krea 2 three-frame storyboard LoRA, inspired by in-context LoRA training. It jointly generates three consecutive states on one vertical canvas to maintain characters, clothing, setting, props and lighting while changing action, position or framing.
Published settings: Krea 2 Turbo BF16; strength 0.85 (try 0.7–1.0); 512×1536; 8 steps; CFG 1.0; Euler Ancestral sampler; Beta scheduler.
[SHOT-SEQUENCE] Shared characters, clothing, setting, props and lighting.
[FRAME-1] Initial state.
[FRAME-2] Intermediate state.
[FRAME-3] Final state.
Start at the training resolution, 512×1536, then upscale the panels separately. Subtle gaze, expression and gesture changes may be weak; complex hand/prop interactions may fail or duplicate props. Excessive strength may make the later frames too similar, and distant faces have limited detail.
Provenance and terms
The original public descriptions, exact trigger field, permission flags and file metadata are retained in source.json. Original model and component terms remain applicable; this mirror does not grant a different license.