BLOG

Hero 英雄

by GymDreams
Chinese martial arts, rendered with Stable Diffusion and Virile Fusion, using Control Net with images from Midjourney.
Hero 英雄, Chinese kung fu, rendered with Stable Diffusion and Virile Fusion, using Control Net with images from Midjourney.

These images were influenced by the movie Hero 英雄 (2002), directed, co-written, and produced by Zhang Yimou, a Chinese film director, producer, writer, actor, professor and former cinematographer.

The movie uses strong color palettes to separate the different scenes, and featured Chinese martial arts (kung fu 功夫) in a way that is both realistic and magical, which is a signature unique to Chinese martial arts movies. If you’re not familiar with this genre, perhaps you would be familiar with Kill Bill from Quentin Tarantino, because the fight scenes in Kill Bill were choreographed by Yuen Woo-ping, one of the most influential figures in the world for traditioanl martial arts movies. Tarantino hired him specifically to recreate this style, so there were many similarities there. And if you’re not familiar with Kill Bill, Yuen Woo-ping also choreographed all the fight scenes for The Matrix (1999).

These images were rendered inside Stable Diffusion using txt2img, but the workflow was a bit more complicated than a straight-up prompt. I had an idea of the visuals I wanted, but I couldn’t quite get the results I wanted — especially when it comes to the strength of the color coverage. So I used Midjourney to first generate a sketch that’s close to my vision, then I used that image in two Control Nets: CN0 Reference Only, CN1 Shuffle, both set to CN priority, in order to control the style of the render that I had in mind.

In these image sets, the first one is the final render. The second one is the image I made inside Midjourney that was used to influence the final render. All of the images share a few common elements that define the mood, the figures, and the clothings, but the scenery and colors were modified in each of them, thus creating a harmonizing set that’s related but also very different from each other.

To summarize, here’s the flow:

  • Text prompts in Midjourney v5.2 -> Image
  • Stable Diffusion:
    • Text prompts
    • CN0 Reference Only with Image, CN Priority
    • CN1 Shuffle with Image, CN Priority, 1536 res

Other parameters:

25 steps, DPM++ SDE, CFG 6, 768x768, Virile Fussion v2.0, Denoising 0.5, ADetailer face_yolov8n.pt, Hires 2x (1536x1536), 10 steps, 4x_foolhardy_Remacri. Post: Giagapixel HQ 2x (3072x3072), Lightroom color correction, Photoshop Beta AI fixes (hands).

Images

Cherry Blossom

Cherry Blossom, Hero 英雄, Chinese kung fu, rendered with Stable Diffusion and Virile Fusion, using Control Net with images from Midjourney.
Cherry Blossom, Hero 英雄, Chinese kung fu, rendered with Stable Diffusion and Virile Fusion, using Control Net with images from Midjourney.
Cherry Blossom, Hero 英雄, Chinese kung fu, rendered with Stable Diffusion and Virile Fusion, using Control Net with images from Midjourney.
Cherry Blossom, Hero 英雄, Chinese kung fu, rendered with Stable Diffusion and Virile Fusion, using Control Net with images from Midjourney.

Snow Storm

Snow Storm, Hero 英雄, Chinese kung fu, rendered with Stable Diffusion and Virile Fusion, using Control Net with images from Midjourney.
Snow Storm, Hero 英雄, Chinese kung fu, rendered with Stable Diffusion and Virile Fusion, using Control Net with images from Midjourney.
Snow Storm, Hero 英雄, Chinese kung fu, rendered with Stable Diffusion and Virile Fusion, using Control Net with images from Midjourney.
Snow Storm, Hero 英雄, Chinese kung fu, rendered with Stable Diffusion and Virile Fusion, using Control Net with images from Midjourney.
Snow Storm, Hero 英雄, Chinese kung fu, rendered with Stable Diffusion and Virile Fusion, using Control Net with images from Midjourney.
Snow Storm, Hero 英雄, Chinese kung fu, rendered with Stable Diffusion and Virile Fusion, using Control Net with images from Midjourney.

Bamboo Forest

Bamboo Forest, Hero 英雄, Chinese kung fu, rendered with Stable Diffusion and Virile Fusion, using Control Net with images from Midjourney.
Bamboo Forest, Hero 英雄, Chinese kung fu, rendered with Stable Diffusion and Virile Fusion, using Control Net with images from Midjourney.
Bamboo Forest, Hero 英雄, Chinese kung fu, rendered with Stable Diffusion and Virile Fusion, using Control Net with images from Midjourney.
Bamboo Forest, Hero 英雄, Chinese kung fu, rendered with Stable Diffusion and Virile Fusion, using Control Net with images from Midjourney.

Technical Parameters

  • Sampler: DPM++ SDE
  • CFG scale: 6
  • Size: 768x768
  • Model hash: f1de8faa49
  • Model: virileFusion_v20
  • Denoising strength: 0.5
  • ADetailer model: face_yolov8n.pt
  • ADetailer prompt: handsome chinese face
  • ADetailer confidence: 0.3
  • ADetailer dilate/erode: 4
  • ADetailer mask blur: 4
  • ADetailer denoising strength: 0.4
  • ADetailer inpaint only masked: True
  • ADetailer inpaint padding: 32
  • ADetailer version: 23.7.8
  • ControlNet 0:
    • preprocessor: reference_only
    • model: None
    • weight: 1
    • starting/ending: (0, 1)
    • resize mode: Crop and Resize
    • pixel perfect: False
    • control mode: ControlNet is more important
    • preprocessor params: (-1, 0.5, -1)
  • ControlNet 1:
    • preprocessor: shuffle
    • model: control_v11e_sd15_shuffle 526bfdae
    • weight: 1
    • starting/ending: (0, 1)
    • resize mode: Crop and Resize
    • pixel perfect: False
    • control mode: ControlNet is more important
    • preprocessor params: (1536, -1, -1)
  • Hires upscale: 2 (1536x1536)
  • Hires steps: 10
  • Hires upscaler: 4x_foolhardy_Remacri
  • Lora hashes:
    • epi_noiseoffset2: d1131f7207d6
    • add_detail: 7c6bad76eb54"
  • Version: v1.3.2
  • Post Upscale: Topaz Gigapixel AI HQ 2x (3072x3072)
  • Post: Adobe Photoshop AI Beta (hand fix)
  • Post: Adobe Lightroom (color correction)