Hybrid Anima 2.9B → KREA2 AutoPrompt Upscale Workflow

This workflow automatically analyzes an input illustration, generates an English tag-based prompt from the image, and then uses both the generated prompt and the source image in a two-stage Anima 2.9B → KREA2 img2img pipeline.

It automatically detects elements such as the character, outfit, pose, and composition. The image is first regenerated with Anima 2.9B, then refined with KREA2 to improve details, edges, text, textures, and overall image quality.

You can also append your own preset prompts to the automatically generated tags. This makes it possible to analyze each image automatically while maintaining a consistent style or quality specification across generations.

The degree of similarity to the original image can be adjusted using the Anima denoise value, while the degree of similarity to the Anima-generated image can be controlled using the KREA2 denoise value.

Recommended Use Cases

This workflow is suitable if you want to:

• Reuse the character, outfit, pose, and composition of an input image

• Automatically create prompts directly from an image

• First regenerate the image with Anima using a combination of natural language + Danbooru / e621 tags

• Then use KREA2 to refine details, text, edges, and textures

• Improve image quality semi-automatically without manually writing long prompts every time

Workflow Overview

This workflow uses ExistingCLWD14Prompt to analyze the input image.

The resulting tags are automatically sent to the positive prompts of both Anima 2.9B and KREA2.

In addition to the automatically generated tags, you can append your own Preset Prompt.

This allows you to combine:

• Automatic image-specific analysis

• Consistent style and quality instructions

within the same workflow.

Workflow Structure

1. Input Illustration Analysis

First, load any illustration and analyze it using:

Existing CL Tagger v2 + WD14 v3 → English Prompt

This step mainly extracts features such as:

• Character

• Hairstyle and hair color

• Outfit

• Pose

• Facial expression

• Background

• Composition

These features are automatically converted into an English tag-based prompt.

For example, an input image may produce:

1girl, pink hair, blue eyes, sitting, white skirt, looking at viewer

2. Stage 1: ANIMA 2.9B Reference img2img

The automatically generated tags are sent to:

ANIMA 2.9B Positive Prompt ← AUTO TAGS + PRESET

At this stage,

Automatically generated tags + preset prompt

are combined and used as the positive prompt for Anima.

The original input illustration itself is also used as the img2img reference image for Anima.

Default Setting

• denoise = 0.70

This value greatly affects how closely the result follows the original image.

Denoise Guidelines

• 0.10–0.25
  Preserves most of the original image

• 0.30–0.45
  Strong adherence to the original image
  Recommended when you want to preserve the original pose and composition as much as possible

• 0.50–0.65
  Maintains the original image while allowing a moderate amount of redrawing

• 0.70–0.80
  Gives Anima more freedom to reinterpret the source image

• 0.80–1.00
  Produces major changes from the source
  The input image becomes only a loose reference

If you want to preserve the original pose and composition more strongly, starting around 0.50 is recommended.

3. Stage 2: KREA2 Detail / Text / Texture Refinement

The image generated by Anima is decoded, then encoded again with the VAE before being passed to KREA2.

This stage is mainly used to improve:

• Fine details

• Small objects and patterns

• Edges and outlines

• Textures

• Text

• Typography

• Minor visual artifacts

KREA2 therefore acts primarily as the final refinement and high-detail enhancement stage while using the Anima output as its base image.

Auto Prompt + Preset Prompt

This workflow supports not only automatically analyzed tags, but also user-defined preset prompts.

For example, if the automatic analysis produces:

1girl, pink hair, blue eyes, sitting

and your preset prompt is:

masterpiece, best quality, detailed anime illustration

the actual prompt sent to the model becomes:

1girl, pink hair, blue eyes, sitting, masterpiece, best quality, detailed anime illustration

Separate preset prompts can be configured for Anima and KREA2.

This allows each model to have a different role.

Anima

Use prompts that emphasize:

• Character identity

• Anime-style rendering

• Overall composition and appearance

KREA2

Use prompts that emphasize:

• Fine details

• Textures

• Clean edges

• Text rendering

• Typography

Main Features

• Input image → automatic English prompt generation

• Image analysis with CL Tagger v2 + WD14 v3

• Automatic extraction of character, outfit, pose, and composition

• Automatic prompt input for Anima

• Automatic prompt input for KREA2

• Combination of automatic tags + preset prompts

• Use of the input image as Anima's img2img reference

• Two-stage Anima → KREA2 generation

• KREA2-based improvement of details, edges, and text rendering

• Optional FaceDetailer

• Optional Remacri finishing process

Recommended Applications

This workflow is especially useful for:

• Improving the quality of existing illustrations

• Redrawing an image while maintaining the original character

• Improving the style while preserving pose and composition

• Avoiding the need to manually write long prompts

• Automatically generating prompts from images

• Combining Anima's character rendering with KREA2's fine-detail capabilities

• Regenerating old or low-resolution illustrations

• Turning rough artwork into something closer to a finished illustration

• Reinterpreting character images with different models

Optional Post-Processing

FaceDetailer

FaceDetailer detects facial regions and regenerates facial details.

Enable it only when needed.

Remacri

The KREA2 output can optionally be upscaled using:

4x_foolhardy_Remacri

The image is first enlarged by 4× and then downscaled to 0.25× using Lanczos.

This process is intended to reduce:

• Fine noise

• Rough linework

• Jagged edges

• Small image artifacts

while keeping edges relatively sharp.

Main Required Models

This workflow primarily uses the following models.

ANIMA

• Anima 2.9B

KREA2

• liquidmix_krea2_turbo_fp8

Text Encoders

• qwen_3_06b_base

• qwen3vl_4b_bf16

VAE

• qwenimageVAE_liquid1127

Optional

• Anima LoRA

• KREA2 LoRA

• 4x_foolhardy_Remacri

Model files themselves are not included with this workflow.

Please obtain them from their respective distribution pages.

Custom Nodes

Depending on your environment, this workflow uses the following custom nodes.

Main Custom Nodes

• ExistingCLWD14Prompt

• ApplyKrea2NegPiP

• ComfyUI-Custom-Scripts

Optional

• FaceDetailer

• Image Comparer

• Anima29LegacyLoRALoader

If you do not use certain optional features, the corresponding nodes can be disabled.

Notes

This workflow is not a simple upscaler.

Instead of merely enlarging the original image, it performs the following process:

Image Analysis
→ Prompt Generation
→ Anima Regeneration
→ KREA2 Refinement

Because of this, the final result will not necessarily be identical to the source image.

If you want to preserve the character, pose, and composition more strongly, lower the Anima denoise value.

Processing Flow

INPUT ILLUSTRATION
        ↓
CL Tagger v2 + WD14 v3
        ↓
English Auto Tags
        ↓
Auto Tags + Preset Prompt
        ↓
ANIMA 2.9B
Reference Img2Img
        ↓
ANIMA Output
        ↓
KREA2
Detail / Text Refinement
        ↓
KREA2 Output
        ↓
Optional FaceDetailer
        ↓
Optional Remacri Cleanup
        ↓
FINAL IMAGE