Recreating the “Pure Desire” Photo Style with AI: A ComfyUI Workflow for Masterful Light and Mood

Effect Preview: Pursuing Essence Over Likeness

The goal of AI recreation is not pixel‑perfect duplication but capturing the original image’s spirit, atmosphere, and stylistic feeling. Side‑by‑side comparisons of source photos and AI‑generated results illustrate this approach.

Core Workflow Breakdown

The process mimics a professional photographer’s post‑production pipeline, moving from analysis and prompt creation to generation, refinement, upscaling, and final grading.

Workflow Overview

The following node diagram (shown below) represents the full ComfyUI workflow; each module is explained in subsequent sections.

Step 1: Prompt Engineering – Directing the AI

This step is decisive. Rather than a simple scene description, the author writes a detailed shooting script from the perspective of a “photography director.”

Prompt reverse and strengthening : an image‑to‑text model first extracts a basic description, which is then professionally enriched.

Core prompt structure :

Role setting :

你是摄影大师，擅长设计女性肖像，尤其是纯欲风格的摄影。你有光之艺术的手法，擅长敏锐捕捉人物的神情、光线...

Descriptive dimensions : composition, ambience, pose (makeup, emotion, hairstyle, action), lens language (focal length, depth of field), lighting style (light source type, quality), clothing details, and post‑processing techniques.

Style reinforcement commands (critical constraints):

【非常重要】务必表达纯欲，中国女性，不裸露的含义。
【非常重要】人物真实，淡妆，脸部偏可爱风格，肌肤肌理细腻，可见毛孔，16K，真实拍摄。
【非常注意】妆容为淡妆，不要太突出的卧蚕，不要眼袋。
【非常注意】强化氛围和场景。大师级调色。
【非常注意】16K画质、超写实风格、超高质量画质，大师级调色、超高画面细节。

Step 2: Generate Image with Z‑Image

The refined prompt is fed into the standard Z‑Image generation pipeline, producing the initial portrait.

Step 3: Optional Fine‑Tuning

Z‑Image sometimes yields “mask‑like” faces or details that clash with the pure‑desire aesthetic (e.g., visible eye bags). The author uses Qwen‑Edit or similar image‑to‑image models for localized correction.

Modification prompt example :

减淡女孩的眼袋和卧蚕，脸部变得圆润些。其他100%不要改变！

Before/after comparison:

Note: This step’s model stability needs improvement and may cause pixel loss; multiple attempts are often required.

Step 4: Lossless Upscaling

A dedicated Topaz upscaling node enlarges the image while preserving fine details, essential for print‑quality output.

Dynamic upscaling demonstration:

For detailed usage of this node, refer to the earlier article “可能是最好的 ComfyUI 无损放大方案？”

Step 5: Color Grading and Grain – Adding Mood and Cinematic Feel

The final step adds film‑like texture and cinematic color tones using a filter node that bundles many popular presets.

brooklyn : warm tones, skin‑tone warmth, overall bright and soft.

1977 : warm, added red/yellow, faded 70s look.

aden : low saturation, cool‑greenish, fresh.

brannan : high contrast, brownish, vintage feel.

clarendon : high contrast, cool, emphasizes blues and highlights.

earlybird : warm yellow, slight vignette, retro.

gingham : low saturation, cool‑gray, clean.

hudson : cool blue, high contrast, dark blues.

inkwell : black‑white, high contrast.

kelvin : warm orange, incandescent feel.

lark : low saturation, bright, cool.

lofi : high saturation, contrast, vignette.

maven : greenish, dim, retro.

mayfair : pink, soft skin, bright warmth.

moon : black‑white, low contrast, soft gray.

nashville : pink‑purple, desaturated, faded.

perpetua : cool green, fresh, slight cinematic.

reyes : lowered contrast, warm brown, soft fade.

rise : warm, soft light, gentle skin.

slumber : desaturated, gray‑blue, hazy.

stinson : cool, low contrast, gentle.

toaster : warm orange, strong vignette, retro drama.

valencia : warm yellow, bright, soft fade.

walden : high saturation, blue shift, bright.

willow : black‑white, gray‑green, soft low‑key.

xpro2 : high contrast, vivid, cyan‑tint, cross‑processed film look.

Final Result Showcase

After completing the five steps, a high‑resolution final image is obtained. Even after compression by the platform, the texture remains evident.

Detail Zoom : close‑ups reveal skin texture, hair strands, and nuanced light‑shadow layers.

Extending the Workflow

The workflow’s core lies in reusable prompt engineering and a standardized post‑processing chain. By swapping style descriptors and adjusting filters, virtually any photographic style can be reproduced.

Example: recreating a “hippocampus”‑style Spring Festival portrait.

Conclusion: AI as the New Darkroom

From silver‑based darkrooms to digital labs and now AI workflows, tools evolve while photography’s core—control of light, capture of emotion, definition of beauty—remains unchanged. This workflow is not merely a technical solution but a design mindset that deconstructs an artistic style into describable, optimizable, and reproducible steps, giving creators a powerful prototype tool to explore visual boundaries and focus more on creativity.