【Anima】Simple T2I Workflow with Upscale, Detailers and ControlNet — v1.0

This Anima text-to-image workflow takes text prompts and produces images through two second-pass upscale routes, with face and hand detailers and a bypassable Anima-LLLite ControlNet.

Choose between Ultimate SD Upscale and HiresUpscale via second-pass resampling for the second pass.

Face and hand detailers are included, and the Anima-LLLite ControlNet can be bypassed.

The main input is a text prompt; the output is an image that can continue through a second-pass upscale route.

Version v1.0 uses Anima as its base model and is published by SAN0.

No license is stated; verify the applicable usage terms before using or sharing it.

Suggestion

The named model files are anima-preview3-base.safetensors (UNet), qwen_3_06b_base.safetensors (text encoder), qwen_image_vae.safetensors (VAE), and 4x_foolhardy_Remacri.pth (upscaler).

It also names Furry Domain .07-000009.safetensors (LoRA) and rgthree-comfy (node pack).

The recommended ControlNet file is anima-lllite-any-test-like-v2.safetensors, and the custom node is ComfyUI-Anima-LLLite.

The description says Ultimate SD Upscale is stable below 0.4 denoise, supports upscaling beyond 2x (4MP), and becomes unstable at high denoise.

For HiresUpscale, the description recommends 1.5x (2.25MP) at 0.65 denoise; if higher denoise causes unnatural transitions, it suggests tile_padding such as 256.

Suggestion · non vérifié

HiresUpscale beyond 2x (>4MP) is described as very unstable, and low denoise may occasionally produce a grainy texture.

Before use, verify that Furry Domain .07-000009.safetensors and 4x_foolhardy_Remacri.pth are the files you intend to use.

Suggestion · non vérifié