JoyAI-Echo Multishot Workflow - one character, many shots, same face + voice — v1.3.1

v1.3.1 — documentation release. No code changes; same nodes and workflow as v1.3. The zip now contains INSTRUCTIONS.md: a full step-by-step guide written for people who could not get v1.3 working, covering the things the notes files never explained properly:

• Where the reference image goes. There is no image node in the shipped graph because the reference is optional — the guide shows the 30-second wiring (Add Node > image > Load Image, connect to the Generate node's reference_image input), explains what it actually does (it pre-seeds the cross-shot memory bank as a permanent identity anchor — every shot gets conditioned on that face), and how reference_zoom controls whether shot 1 opens ON your image (set 1.0 for classic i2v-style behavior) or just inherits the identity (the 1.2 default).
• What this workflow is and is not — multi-shot text-to-video with joint audio and a memory bank, not a classic i2v graph. The byte-identical identity sentence rule, spelled out with the reasoning.
• LoRAs: the loader takes one LoRA today, fused at load. Workaround for multiple (merge first), and multi-LoRA loading is on the roadmap.
• Performance settings by VRAM — sequential offload, resident_blocks guidance per card size, which toggles are 40/50-series only.
• A troubleshooting table for every failure mode reported so far: the GGUF checkpoint_path error, duplicated/garbled characters (wrong text encoder — use true base Gemma, not the circulating "joyecho" one), face drift between shots, British accent, invented music, widgets resetting after restart, and the master video arriving a few minutes after the run ends.

If v1.3 works for you already, there is nothing new to download — this is the same pack with a much better manual.