NOOB2 Project-Character Reference Bypass Injector Research

This is the initial public research-preview release of Anima Native Reference V2 E180 .

Compared with a normal Anima workflow, this version adds an integrated Native Reference V2 path for prompt-controlled generation from two ordered reference images.

What is included

Integrated checkpoint: anima-native-ref-v2-e180-step64080-256px.safetensors

Checkpoint step: 64,080

Two explicitly ordered reference inputs

Prompt-defined reference semantics

Pure-noise generation

Native Reference V2 sequence and attention paths

Standard ComfyUI workflow

Local /prompt API workflow

Balanced, high-VRAM, and CPU text-encoder modes

Runtime, prompt-embedding, and reference-latent caches

Checkpoint-layout validation

Optional release SHA-256 verification

Integration tests and validation reports

Required custom node

Install the custom node before using this model:

https://github.com/HK416-TYPED/ComfyUI-Anima-Native-Reference

Required files

In addition to the E180 checkpoint downloaded from this version, you need:

qwen_3_06b_base.safetensors qwen_image_vae.safetensors

Download them from the Anima repository:

https://huggingface.co/circlestone-labs/Anima

Important usage note

This is a complete integrated diffusion checkpoint, not a LoRA.

Do not load an additional stock Anima diffusion checkpoint alongside it.

No external LoRA, IP-Adapter, or separate reference-adapter weight file is required.

Recommended starting settings

Resolution: 256 × 256 Steps: 40 CFG: 1.0 Flow shift: 5.0 Reference scale: 1.0 Offload mode: balanced

Larger output sizes are accepted when their dimensions are multiples of 16, but they remain experimental.

Current status

This version is intended as a research preview rather than a final production release. Please report reproducible problems through the GitHub repository and include your workflow, prompt, reference setup, hardware, and error log.