Ecocoro Preview 1

eyecatch

日本語版の説明

Ecocoro is an image generation full-finetuned model based on black-forest-labs/FLUX.2-klein-base-4B.

Ecocoro is designed for generating expressive illustrations and anime-style images from text prompts while preserving the flexibility and accessibility of the FLUX.2 klein 4B family.

Usage

There are two ways to use this model. The first is with ComfyUI, and the second is with diffusers.

Using with ComfyUI

This section explains how to use the model with ComfyUI.

First, prepare an environment where you can run black-forest-labs/FLUX.2-klein-base-4B by following the guide below:

ComfyUI Flux.2 Klein 4B Guide

Next, download the model from the following link:

ecocoro-preview-1-fp8 (light weight)

Or you can download the BF16 model from the following link:

ecocoro-preview-1 (standard weight)

After downloading it, place the model file in ComfyUI’s diffusion_models folder.

Then, select ecocoro-preview-1.safetensors instead of flux-2-klein-base-4b-fp8.safetensors. Once this is done, the setup is complete.

Change the prompt as follows:

recent, highreso, 1girl, solo, close-up, brown hair, blue eyes, white T-shirts, ahoge, looking at viewer, blue background, simple background

After that, run the workflow to generate an image.

Using with diffusers

This section explains how to use the model with diffusers.

Set up the required environment, then run the following script:

import torch
from diffusers import Flux2KleinPipeline

model_id = "alfredplpl/ecocoro-preview-1"

pipe = Flux2KleinPipeline.from_pretrained(
  model_id,
  torch_dtype=torch.bfloat16,
)

pipe.enable_model_cpu_offload()

prompt = "recent, highreso, 1girl, solo, close-up, brown hair, blue eyes, white T-shirts, ahoge, looking at viewer, blue background, simple background"

image = pipe(
  prompt=prompt,
  height=1024,
  width=1024,
  num_inference_steps=28,
  guidance_scale=3.5,
).images[0]

image.save("ecocoro_sample.png")

Please adjust inference settings such as resolution, number of steps, guidance scale, and random seed depending on your environment and desired output style.

  • Recommended Settings
    • Resolution: 1024 x 1024
    • Steps: 20 to 50
    • Guidance scale: 3.0 to 4.5
    • Precision: bfloat16 where supported

These values are only starting points. Different prompts and workflows may benefit from different settings.

Prompting

Ecocoro adopts the prompting conventions of circlestone-labs/Anima. It supports Danbooru-style tags, natural language captions, and mixed prompts.

Recommended tag order:

[period/safety tags/highreso] [subject count] [general tags]

Natural language descriptions can be combined freely with tags. Clear descriptions of appearance, pose, composition, clothing, background, and style usually produce better results than very short prompts.

Model Details

  • Model name: Ecocoro Preview 1
  • Base model: black-forest-labs/FLUX.2-klein-base-4B
  • Model type: Text-to-image generation model
  • Primary use: Illustration and anime-style image generation
  • License: Apache 2.0

Base Model

Ecocoro is built on top of FLUX.2 klein base 4B, a compact image generation and editing model released by Black Forest Labs.

Intended Use

Ecocoro is intended for generating anime-style and illustration-style images from text prompts.

Possible use cases include creative prototyping, character design, scene visualization, concept art, research, and experimentation with image generation workflows.

Limitations

Ecocoro may produce incorrect anatomy, hands, text, symbols, or small details.

  • It may misinterpret complex prompts or spatial relationships.
  • It may generate inconsistent character details across multiple images.
  • It may reflect biases inherited from the base model and training process.
  • It may produce lower-quality results for prompts far outside its intended illustration-oriented use cases.

Users should review generated outputs carefully before publication or commercial use.

Ethical Considerations

Please use this model responsibly.

Do not use Ecocoro to generate illegal, harmful, deceptive, harassing, or non-consensual content. Users are responsible for ensuring that generated images comply with applicable laws, platform policies, and the rights of third.

Origin of Ecocoro

The name Ecocoro is inspired by the Japanese word egokoro, meaning an artistic sense or a feel for drawing. The name combines E and cocoro, suggesting both image-making and heart in a form that is easy to pronounce globally.

Gallery

sample_1

prompt:

recent, highreso, 1girl, solo, close-up, brown hair, blue eyes, white T-shirts, ahoge, looking at viewer, blue background, simple background

negative prompt:

bad anatomy, low quality, bad hands, extra fingers

sample_2

prompt:

newest, highres, close-up portrait of a face divided in two, left half is a sleek humanoid robot, right half is a gentle young anime girl, their faces perfectly aligned as one, mechanical eye and human eye looking at viewer, cold metal and warm skin, contrast between artificial intelligence and human emotion, glowing circuits, soft eyelashes, symmetrical composition, cinematic lighting, dark simple background, highly detailed, sharp focus, emotional, beautiful anime illustration

negative prompt:

bad anatomy, low quality, bad hands, extra fingers

sample_3

prompt:

a canvas in a quiet artist studio, the word "Ecocoro" written softly on the canvas, warm afternoon light, scattered sketches, brushes and paint tubes, a gentle feeling of creative inspiration, anime illustration style

negative prompt:

bad anatomy, low quality, bad hands, extra fingers

License

This model is licensed under the Apache License, Version 2.0. See the LICENSE file for details.

Copyright 2026 Yasunori Ozaki

Acknowledgements

Ecocoro is based on black-forest-labs/FLUX.2-klein-base-4B by Black Forest Labs.

Contact

If you would like to discuss ways to reduce potential risks associated with using this model, please contact:

yasunori.ozaki@outlook.com

Downloads last month
44
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for alfredplpl/ecocoro-preview-1

Finetuned
(17)
this model

Spaces using alfredplpl/ecocoro-preview-1 2