This JSON file describes a ComfyUI workflow designed to create cinematic-quality backlit portrait photography with an advanced aesthetic. The workflow specializes in generating high-end portrait images with dramatic lighting effects, particularly backlighting scenarios. Here's a detailed breakdown:

Core Components

1. Model Loading

• CheckpointLoaderSimple (Node 4 & 44):

  • Loads the base model: 高级质感人像-FLUX_v1.0 (High-Quality Portrait FLUX v1.0)

  • Provides MODEL, CLIP, and VAE outputs for image generation.

2. LoRA Integration

• LoraLoader (Nodes 11 & 39):

  • Applies the specialized LoRA: miluo丨逆光镜头人像_极致高级感_v1.4 (Backlit Portraits - Ultimate Premium v1.4)

  • Adjustable strength (0.4–1.2) to fine-tune the backlighting effect.

3. Text Encoding

• CLIPTextEncode (Nodes 6, 7, 31):

  • Positive Prompt: Describes a cinematic Asian woman in a black coat with a mustard scarf, standing in sunset-lit streets. Keywords include:

    • backlight, golden hour lighting, bokeh effect, shallow depth of field, film texture.

  • Negative Prompt: Left empty (defaults to generic negatives).

4. Latent Image Generation

• EmptyLatentImage (Nodes 5 & 30):

  • Sets resolution: 768x1248 (portrait-oriented) and 768x1024.

• KSampler (Node 3) & SamplerCustomAdvanced (Node 34):

  • Uses euler sampler with 30 steps and simple scheduler.

  • Emphasizes controlled noise for cinematic grain.

5. Image Decoding & Output

• VAEDecode (Nodes 8 & 32):

  • Decodes latent images using the loaded VAE (ae.sft).

• PreviewImage (Nodes 27 & 42):

  • Displays results for both online and local workflows.

• SaveImage (Implicit in workflow logic).

Key Features

1. Backlighting Optimization:

   • The LoRA miluo丨逆光镜头人像_极致高级感_v1.4 specializes in simulating natural backlighting (e.g., sunset hair glow, floating dust particles).

   • Prompt engineering includes terms like flickering light spots and large aperture for lens effects.

2. Cinematic Quality:

   • Leverages film and television texture and soft focus for a cinematic look.

   • High resolution (768x1248) ensures detail retention.

3. Dual Workflow Support:

   • Online: Optimized for cloud-based execution (default).

   • Local: Alternative path for offline use (disabled by default).

4. Dynamic Conditioning:

   • The ttN text node (Node 47) provides instructions for customizing prompts while preserving core keywords (miluo, backlight).

Workflow Steps

1. Initialization:

   • Load the base model (高级质感人像-FLUX_v1.0) and LoRA.

   • Encode prompts with CLIP.

2. Image Generation:

   • Generate latent space images using the KSampler with backlight-focused conditions.

   • Refine details via SamplerCustomAdvanced (local path).

3. Post-Processing:

   • Decode latent images to RGB.

   • Preview/save outputs.

Technical Notes

• Trigger Words: miluo, backlight (required for LoRA activation).

• Model Links:

  • Base Model: liblib.art/高级质感人像

  • LoRA: liblib.art/逆光镜头人像

• Seed Control: Randomized by default (randomize enabled).

Use Cases

• Photorealistic Portraits: Ideal for fashion, promotional content, or artistic projects.

• Lighting Studies: Demonstrates advanced backlighting techniques in AI-generated imagery.

• Local/Cloud Flexibility: Adaptable to different hardware environments.

Output Example

The workflow produces images resembling:
"A young Asian woman in a black coat, bathed in golden sunset light, with hair illuminated and atmospheric dust particles, rendered in a shallow-focus cinematic style."

This workflow exemplifies how LoRA specialization and prompt engineering can achieve targeted photographic effects in AI-generated art. The dual-path design ensures versatility across deployment scenarios.