How AI Drawing? Creating Anime Illustrations with HiDream × SDXL Workflow
- SDXL: Soft and natural skin tones
- HiDream: High texture quality
- Use the strengths of both
Introduction
Hello, I'm Easygoing.
This time, I’ll introduce a workflow for creating anime illustrations using HiDream.
From Noise to Illustration!
I’ve created a video showcasing how AI generates images.
AI gradually transforms noise into illustrations, and even if you don’t usually use generative AI, I hope the video gives you a sense of how it works!
The background music, as in my previous post, is provided by Shiki’s AiPeacefulMoment channel.
The channel offers a collection of soothing tracks perfect for background music while working, so please check it out!
First, the Workflow!
Here’s the workflow used this time.
flowchart LR
subgraph SDXL
A1(Rough Sketch)
end
B1(HDR Processing<br>SuperBreasts)
subgraph HiDream
C1(Redrawing)
end
subgraph Upscaler
D1(Upscaling<br>+ Noise Removal)
end
subgraph SDXL
E1(Detailer<br>Face Redrawing)
end
A1-->B1
B1-->C1
C1-->D1
D1-->E1
Models
The video includes an additional step of redrawing eyes using Flux.1[dev], but this was omitted in the distributed version due to its time-consuming processing.
Leveraging the Strengths of HiDream and SDXL!
This time, I used the following two models to generate images:
- HiDream: Released in April 2025, exceptional texture quality
- SDXL: Released in July 2023, high-quality anime illustrations
By combining both models, I aimed to achieve expressions that neither model could accomplish alone!
Rough Sketch: SDXL’s Soft Expressions
First, I created a rough sketch using the previous-generation SDXL model.
SDXL’s anime models excel in soft character expressions and natural color rendering, particularly for skin tones.
Among SDXL’s anime models, the blue_pencil-XL series by blue_pen5805 is convenient because it supports natural language prompts without negative prompts, making it ideal for use with next-generation models like this time.
Adding Presence with HDR Processing!
The orthodox e-pred model in SDXL struggles with high-contrast, richly colored expressions.
To address this weakness, I applied HDR processing.
ComfyUI-SuperBeasts makes HDR processing simple by passing through a custom node, emphasizing darker colors to create a weighty, refined expression.
Enhancing Texture with HiDream!
Next, I redrew the illustration using HiDream.
HiDream stands out among next-generation models for its high parameter count and superior texture quality.
Redrawing with HiDream significantly improved the texture of clothing and backgrounds, and the perspective was refined, resulting in a three-dimensional illustration.
Upscaling + Noise Removal
Next, I upscaled the illustration using an AI upscaler.
For the AI upscaler, I used the IllustrationJaNai_V1 model, which balances anime-style coloring and noise removal.
- IllustrationJaNai_V1_ESRGAN_135k (Lightweight model)
- IllustrationJaNai_V1_DAT2_190k (High-precision model)
By upscaling and removing noise, the background became detailed and clear.
Redrawing Faces with Detailer!
HiDream has decent anime illustration capabilities as a base model, but since it’s not specialized for anime, eyes tend to lose detail, and skin tone expressions are better handled by the previous-generation SDXL model.
So, I used Detailer’s partial correction to redraw faces with SDXL.
Using Detailer resulted in natural skin tones and expressions.
The distributed workflow ends here, but the video includes an additional step of redrawing eyes using the Flux.1 model.
Flux.1 excels at drawing small, low-resolution objects, making it particularly suited for redrawing details like eyes.
Completion!
Here’s the final illustration created with this workflow.
I believe the clothing and background leverage HiDream’s high texture quality, while the expressions and skin tones reflect SDXL’s soft rendering!
Summary: Mix and Match Models!
- SDXL: Soft and natural skin tones
- HiDream: High texture quality
- Use the strengths of both
This time, I introduced a workflow combining HiDream and SDXL.
Although the workflow is titled with HiDream, it’s more about complementing SDXL’s weaknesses with HiDream.
Next-generation models like HiDream offer high texture quality, but SDXL, developed by the open community, still surpasses them in anime illustration expressiveness.
I’m grateful for the many models freely available and will continue exploring unique expressions by combining them.
Thank you for reading to the end!